ia64/linux-2.6.18-xen.hg

view fs/jffs2/readinode.c @ 524:7f8b544237bf

netfront: Allow netfront in domain 0.

This is useful if your physical network device is in a utility domain.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Tue Apr 15 15:18:58 2008 +0100 (2008-04-15)
parents 831230e53067
children
line source
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: readinode.c,v 1.143 2005/11/07 11:14:41 gleixner Exp $
11 *
12 */
14 #include <linux/kernel.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/fs.h>
18 #include <linux/crc32.h>
19 #include <linux/pagemap.h>
20 #include <linux/mtd/mtd.h>
21 #include <linux/compiler.h>
22 #include "nodelist.h"
24 /*
25 * Put a new tmp_dnode_info into the temporaty RB-tree, keeping the list in
26 * order of increasing version.
27 */
28 static void jffs2_add_tn_to_tree(struct jffs2_tmp_dnode_info *tn, struct rb_root *list)
29 {
30 struct rb_node **p = &list->rb_node;
31 struct rb_node * parent = NULL;
32 struct jffs2_tmp_dnode_info *this;
34 while (*p) {
35 parent = *p;
36 this = rb_entry(parent, struct jffs2_tmp_dnode_info, rb);
38 /* There may actually be a collision here, but it doesn't
39 actually matter. As long as the two nodes with the same
40 version are together, it's all fine. */
41 if (tn->version > this->version)
42 p = &(*p)->rb_left;
43 else
44 p = &(*p)->rb_right;
45 }
47 rb_link_node(&tn->rb, parent, p);
48 rb_insert_color(&tn->rb, list);
49 }
51 static void jffs2_free_tmp_dnode_info_list(struct rb_root *list)
52 {
53 struct rb_node *this;
54 struct jffs2_tmp_dnode_info *tn;
56 this = list->rb_node;
58 /* Now at bottom of tree */
59 while (this) {
60 if (this->rb_left)
61 this = this->rb_left;
62 else if (this->rb_right)
63 this = this->rb_right;
64 else {
65 tn = rb_entry(this, struct jffs2_tmp_dnode_info, rb);
66 jffs2_free_full_dnode(tn->fn);
67 jffs2_free_tmp_dnode_info(tn);
69 this = rb_parent(this);
70 if (!this)
71 break;
73 if (this->rb_left == &tn->rb)
74 this->rb_left = NULL;
75 else if (this->rb_right == &tn->rb)
76 this->rb_right = NULL;
77 else BUG();
78 }
79 }
80 list->rb_node = NULL;
81 }
83 static void jffs2_free_full_dirent_list(struct jffs2_full_dirent *fd)
84 {
85 struct jffs2_full_dirent *next;
87 while (fd) {
88 next = fd->next;
89 jffs2_free_full_dirent(fd);
90 fd = next;
91 }
92 }
94 /* Returns first valid node after 'ref'. May return 'ref' */
95 static struct jffs2_raw_node_ref *jffs2_first_valid_node(struct jffs2_raw_node_ref *ref)
96 {
97 while (ref && ref->next_in_ino) {
98 if (!ref_obsolete(ref))
99 return ref;
100 dbg_noderef("node at 0x%08x is obsoleted. Ignoring.\n", ref_offset(ref));
101 ref = ref->next_in_ino;
102 }
103 return NULL;
104 }
106 /*
107 * Helper function for jffs2_get_inode_nodes().
108 * It is called every time an directory entry node is found.
109 *
110 * Returns: 0 on succes;
111 * 1 if the node should be marked obsolete;
112 * negative error code on failure.
113 */
114 static inline int read_direntry(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
115 struct jffs2_raw_dirent *rd, size_t read, struct jffs2_full_dirent **fdp,
116 uint32_t *latest_mctime, uint32_t *mctime_ver)
117 {
118 struct jffs2_full_dirent *fd;
119 uint32_t crc;
121 /* Obsoleted. This cannot happen, surely? dwmw2 20020308 */
122 BUG_ON(ref_obsolete(ref));
124 crc = crc32(0, rd, sizeof(*rd) - 8);
125 if (unlikely(crc != je32_to_cpu(rd->node_crc))) {
126 JFFS2_NOTICE("header CRC failed on dirent node at %#08x: read %#08x, calculated %#08x\n",
127 ref_offset(ref), je32_to_cpu(rd->node_crc), crc);
128 return 1;
129 }
131 /* If we've never checked the CRCs on this node, check them now */
132 if (ref_flags(ref) == REF_UNCHECKED) {
133 struct jffs2_eraseblock *jeb;
134 int len;
136 /* Sanity check */
137 if (unlikely(PAD((rd->nsize + sizeof(*rd))) != PAD(je32_to_cpu(rd->totlen)))) {
138 JFFS2_ERROR("illegal nsize in node at %#08x: nsize %#02x, totlen %#04x\n",
139 ref_offset(ref), rd->nsize, je32_to_cpu(rd->totlen));
140 return 1;
141 }
143 jeb = &c->blocks[ref->flash_offset / c->sector_size];
144 len = ref_totlen(c, jeb, ref);
146 spin_lock(&c->erase_completion_lock);
147 jeb->used_size += len;
148 jeb->unchecked_size -= len;
149 c->used_size += len;
150 c->unchecked_size -= len;
151 ref->flash_offset = ref_offset(ref) | REF_PRISTINE;
152 spin_unlock(&c->erase_completion_lock);
153 }
155 fd = jffs2_alloc_full_dirent(rd->nsize + 1);
156 if (unlikely(!fd))
157 return -ENOMEM;
159 fd->raw = ref;
160 fd->version = je32_to_cpu(rd->version);
161 fd->ino = je32_to_cpu(rd->ino);
162 fd->type = rd->type;
164 /* Pick out the mctime of the latest dirent */
165 if(fd->version > *mctime_ver && je32_to_cpu(rd->mctime)) {
166 *mctime_ver = fd->version;
167 *latest_mctime = je32_to_cpu(rd->mctime);
168 }
170 /*
171 * Copy as much of the name as possible from the raw
172 * dirent we've already read from the flash.
173 */
174 if (read > sizeof(*rd))
175 memcpy(&fd->name[0], &rd->name[0],
176 min_t(uint32_t, rd->nsize, (read - sizeof(*rd)) ));
178 /* Do we need to copy any more of the name directly from the flash? */
179 if (rd->nsize + sizeof(*rd) > read) {
180 /* FIXME: point() */
181 int err;
182 int already = read - sizeof(*rd);
184 err = jffs2_flash_read(c, (ref_offset(ref)) + read,
185 rd->nsize - already, &read, &fd->name[already]);
186 if (unlikely(read != rd->nsize - already) && likely(!err))
187 return -EIO;
189 if (unlikely(err)) {
190 JFFS2_ERROR("read remainder of name: error %d\n", err);
191 jffs2_free_full_dirent(fd);
192 return -EIO;
193 }
194 }
196 fd->nhash = full_name_hash(fd->name, rd->nsize);
197 fd->next = NULL;
198 fd->name[rd->nsize] = '\0';
200 /*
201 * Wheee. We now have a complete jffs2_full_dirent structure, with
202 * the name in it and everything. Link it into the list
203 */
204 jffs2_add_fd_to_list(c, fd, fdp);
206 return 0;
207 }
209 /*
210 * Helper function for jffs2_get_inode_nodes().
211 * It is called every time an inode node is found.
212 *
213 * Returns: 0 on succes;
214 * 1 if the node should be marked obsolete;
215 * negative error code on failure.
216 */
217 static inline int read_dnode(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
218 struct jffs2_raw_inode *rd, struct rb_root *tnp, int rdlen,
219 uint32_t *latest_mctime, uint32_t *mctime_ver)
220 {
221 struct jffs2_tmp_dnode_info *tn;
222 uint32_t len, csize;
223 int ret = 1;
224 uint32_t crc;
226 /* Obsoleted. This cannot happen, surely? dwmw2 20020308 */
227 BUG_ON(ref_obsolete(ref));
229 crc = crc32(0, rd, sizeof(*rd) - 8);
230 if (unlikely(crc != je32_to_cpu(rd->node_crc))) {
231 JFFS2_NOTICE("node CRC failed on dnode at %#08x: read %#08x, calculated %#08x\n",
232 ref_offset(ref), je32_to_cpu(rd->node_crc), crc);
233 return 1;
234 }
236 tn = jffs2_alloc_tmp_dnode_info();
237 if (!tn) {
238 JFFS2_ERROR("failed to allocate tn (%zu bytes).\n", sizeof(*tn));
239 return -ENOMEM;
240 }
242 tn->partial_crc = 0;
243 csize = je32_to_cpu(rd->csize);
245 /* If we've never checked the CRCs on this node, check them now */
246 if (ref_flags(ref) == REF_UNCHECKED) {
248 /* Sanity checks */
249 if (unlikely(je32_to_cpu(rd->offset) > je32_to_cpu(rd->isize)) ||
250 unlikely(PAD(je32_to_cpu(rd->csize) + sizeof(*rd)) != PAD(je32_to_cpu(rd->totlen)))) {
251 JFFS2_WARNING("inode node header CRC is corrupted at %#08x\n", ref_offset(ref));
252 jffs2_dbg_dump_node(c, ref_offset(ref));
253 goto free_out;
254 }
256 if (jffs2_is_writebuffered(c) && csize != 0) {
257 /* At this point we are supposed to check the data CRC
258 * of our unchecked node. But thus far, we do not
259 * know whether the node is valid or obsolete. To
260 * figure this out, we need to walk all the nodes of
261 * the inode and build the inode fragtree. We don't
262 * want to spend time checking data of nodes which may
263 * later be found to be obsolete. So we put off the full
264 * data CRC checking until we have read all the inode
265 * nodes and have started building the fragtree.
266 *
267 * The fragtree is being built starting with nodes
268 * having the highest version number, so we'll be able
269 * to detect whether a node is valid (i.e., it is not
270 * overlapped by a node with higher version) or not.
271 * And we'll be able to check only those nodes, which
272 * are not obsolete.
273 *
274 * Of course, this optimization only makes sense in case
275 * of NAND flashes (or other flashes whith
276 * !jffs2_can_mark_obsolete()), since on NOR flashes
277 * nodes are marked obsolete physically.
278 *
279 * Since NAND flashes (or other flashes with
280 * jffs2_is_writebuffered(c)) are anyway read by
281 * fractions of c->wbuf_pagesize, and we have just read
282 * the node header, it is likely that the starting part
283 * of the node data is also read when we read the
284 * header. So we don't mind to check the CRC of the
285 * starting part of the data of the node now, and check
286 * the second part later (in jffs2_check_node_data()).
287 * Of course, we will not need to re-read and re-check
288 * the NAND page which we have just read. This is why we
289 * read the whole NAND page at jffs2_get_inode_nodes(),
290 * while we needed only the node header.
291 */
292 unsigned char *buf;
294 /* 'buf' will point to the start of data */
295 buf = (unsigned char *)rd + sizeof(*rd);
296 /* len will be the read data length */
297 len = min_t(uint32_t, rdlen - sizeof(*rd), csize);
298 tn->partial_crc = crc32(0, buf, len);
300 dbg_readinode("Calculates CRC (%#08x) for %d bytes, csize %d\n", tn->partial_crc, len, csize);
302 /* If we actually calculated the whole data CRC
303 * and it is wrong, drop the node. */
304 if (len >= csize && unlikely(tn->partial_crc != je32_to_cpu(rd->data_crc))) {
305 JFFS2_NOTICE("wrong data CRC in data node at 0x%08x: read %#08x, calculated %#08x.\n",
306 ref_offset(ref), tn->partial_crc, je32_to_cpu(rd->data_crc));
307 goto free_out;
308 }
310 } else if (csize == 0) {
311 /*
312 * We checked the header CRC. If the node has no data, adjust
313 * the space accounting now. For other nodes this will be done
314 * later either when the node is marked obsolete or when its
315 * data is checked.
316 */
317 struct jffs2_eraseblock *jeb;
319 dbg_readinode("the node has no data.\n");
320 jeb = &c->blocks[ref->flash_offset / c->sector_size];
321 len = ref_totlen(c, jeb, ref);
323 spin_lock(&c->erase_completion_lock);
324 jeb->used_size += len;
325 jeb->unchecked_size -= len;
326 c->used_size += len;
327 c->unchecked_size -= len;
328 ref->flash_offset = ref_offset(ref) | REF_NORMAL;
329 spin_unlock(&c->erase_completion_lock);
330 }
331 }
333 tn->fn = jffs2_alloc_full_dnode();
334 if (!tn->fn) {
335 JFFS2_ERROR("alloc fn failed\n");
336 ret = -ENOMEM;
337 goto free_out;
338 }
340 tn->version = je32_to_cpu(rd->version);
341 tn->fn->ofs = je32_to_cpu(rd->offset);
342 tn->data_crc = je32_to_cpu(rd->data_crc);
343 tn->csize = csize;
344 tn->fn->raw = ref;
346 /* There was a bug where we wrote hole nodes out with
347 csize/dsize swapped. Deal with it */
348 if (rd->compr == JFFS2_COMPR_ZERO && !je32_to_cpu(rd->dsize) && csize)
349 tn->fn->size = csize;
350 else // normal case...
351 tn->fn->size = je32_to_cpu(rd->dsize);
353 dbg_readinode("dnode @%08x: ver %u, offset %#04x, dsize %#04x, csize %#04x\n",
354 ref_offset(ref), je32_to_cpu(rd->version), je32_to_cpu(rd->offset), je32_to_cpu(rd->dsize), csize);
356 jffs2_add_tn_to_tree(tn, tnp);
358 return 0;
360 free_out:
361 jffs2_free_tmp_dnode_info(tn);
362 return ret;
363 }
365 /*
366 * Helper function for jffs2_get_inode_nodes().
367 * It is called every time an unknown node is found.
368 *
369 * Returns: 0 on success;
370 * 1 if the node should be marked obsolete;
371 * negative error code on failure.
372 */
373 static inline int read_unknown(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref, struct jffs2_unknown_node *un)
374 {
375 /* We don't mark unknown nodes as REF_UNCHECKED */
376 BUG_ON(ref_flags(ref) == REF_UNCHECKED);
378 un->nodetype = cpu_to_je16(JFFS2_NODE_ACCURATE | je16_to_cpu(un->nodetype));
380 switch(je16_to_cpu(un->nodetype) & JFFS2_COMPAT_MASK) {
382 case JFFS2_FEATURE_INCOMPAT:
383 JFFS2_ERROR("unknown INCOMPAT nodetype %#04X at %#08x\n",
384 je16_to_cpu(un->nodetype), ref_offset(ref));
385 /* EEP */
386 BUG();
387 break;
389 case JFFS2_FEATURE_ROCOMPAT:
390 JFFS2_ERROR("unknown ROCOMPAT nodetype %#04X at %#08x\n",
391 je16_to_cpu(un->nodetype), ref_offset(ref));
392 BUG_ON(!(c->flags & JFFS2_SB_FLAG_RO));
393 break;
395 case JFFS2_FEATURE_RWCOMPAT_COPY:
396 JFFS2_NOTICE("unknown RWCOMPAT_COPY nodetype %#04X at %#08x\n",
397 je16_to_cpu(un->nodetype), ref_offset(ref));
398 break;
400 case JFFS2_FEATURE_RWCOMPAT_DELETE:
401 JFFS2_NOTICE("unknown RWCOMPAT_DELETE nodetype %#04X at %#08x\n",
402 je16_to_cpu(un->nodetype), ref_offset(ref));
403 return 1;
404 }
406 return 0;
407 }
409 /*
410 * Helper function for jffs2_get_inode_nodes().
411 * The function detects whether more data should be read and reads it if yes.
412 *
413 * Returns: 0 on succes;
414 * negative error code on failure.
415 */
416 static int read_more(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
417 int right_size, int *rdlen, unsigned char *buf, unsigned char *bufstart)
418 {
419 int right_len, err, len;
420 size_t retlen;
421 uint32_t offs;
423 if (jffs2_is_writebuffered(c)) {
424 right_len = c->wbuf_pagesize - (bufstart - buf);
425 if (right_size + (int)(bufstart - buf) > c->wbuf_pagesize)
426 right_len += c->wbuf_pagesize;
427 } else
428 right_len = right_size;
430 if (*rdlen == right_len)
431 return 0;
433 /* We need to read more data */
434 offs = ref_offset(ref) + *rdlen;
435 if (jffs2_is_writebuffered(c)) {
436 bufstart = buf + c->wbuf_pagesize;
437 len = c->wbuf_pagesize;
438 } else {
439 bufstart = buf + *rdlen;
440 len = right_size - *rdlen;
441 }
443 dbg_readinode("read more %d bytes\n", len);
445 err = jffs2_flash_read(c, offs, len, &retlen, bufstart);
446 if (err) {
447 JFFS2_ERROR("can not read %d bytes from 0x%08x, "
448 "error code: %d.\n", len, offs, err);
449 return err;
450 }
452 if (retlen < len) {
453 JFFS2_ERROR("short read at %#08x: %zu instead of %d.\n",
454 offs, retlen, len);
455 return -EIO;
456 }
458 *rdlen = right_len;
460 return 0;
461 }
463 /* Get tmp_dnode_info and full_dirent for all non-obsolete nodes associated
464 with this ino, returning the former in order of version */
465 static int jffs2_get_inode_nodes(struct jffs2_sb_info *c, struct jffs2_inode_info *f,
466 struct rb_root *tnp, struct jffs2_full_dirent **fdp,
467 uint32_t *highest_version, uint32_t *latest_mctime,
468 uint32_t *mctime_ver)
469 {
470 struct jffs2_raw_node_ref *ref, *valid_ref;
471 struct rb_root ret_tn = RB_ROOT;
472 struct jffs2_full_dirent *ret_fd = NULL;
473 unsigned char *buf = NULL;
474 union jffs2_node_union *node;
475 size_t retlen;
476 int len, err;
478 *mctime_ver = 0;
480 dbg_readinode("ino #%u\n", f->inocache->ino);
482 if (jffs2_is_writebuffered(c)) {
483 /*
484 * If we have the write buffer, we assume the minimal I/O unit
485 * is c->wbuf_pagesize. We implement some optimizations which in
486 * this case and we need a temporary buffer of size =
487 * 2*c->wbuf_pagesize bytes (see comments in read_dnode()).
488 * Basically, we want to read not only the node header, but the
489 * whole wbuf (NAND page in case of NAND) or 2, if the node
490 * header overlaps the border between the 2 wbufs.
491 */
492 len = 2*c->wbuf_pagesize;
493 } else {
494 /*
495 * When there is no write buffer, the size of the temporary
496 * buffer is the size of the larges node header.
497 */
498 len = sizeof(union jffs2_node_union);
499 }
501 /* FIXME: in case of NOR and available ->point() this
502 * needs to be fixed. */
503 buf = kmalloc(len, GFP_KERNEL);
504 if (!buf)
505 return -ENOMEM;
507 spin_lock(&c->erase_completion_lock);
508 valid_ref = jffs2_first_valid_node(f->inocache->nodes);
509 if (!valid_ref && f->inocache->ino != 1)
510 JFFS2_WARNING("Eep. No valid nodes for ino #%u.\n", f->inocache->ino);
511 while (valid_ref) {
512 unsigned char *bufstart;
514 /* We can hold a pointer to a non-obsolete node without the spinlock,
515 but _obsolete_ nodes may disappear at any time, if the block
516 they're in gets erased. So if we mark 'ref' obsolete while we're
517 not holding the lock, it can go away immediately. For that reason,
518 we find the next valid node first, before processing 'ref'.
519 */
520 ref = valid_ref;
521 valid_ref = jffs2_first_valid_node(ref->next_in_ino);
522 spin_unlock(&c->erase_completion_lock);
524 cond_resched();
526 /*
527 * At this point we don't know the type of the node we're going
528 * to read, so we do not know the size of its header. In order
529 * to minimize the amount of flash IO we assume the node has
530 * size = JFFS2_MIN_NODE_HEADER.
531 */
532 if (jffs2_is_writebuffered(c)) {
533 /*
534 * We treat 'buf' as 2 adjacent wbufs. We want to
535 * adjust bufstart such as it points to the
536 * beginning of the node within this wbuf.
537 */
538 bufstart = buf + (ref_offset(ref) % c->wbuf_pagesize);
539 /* We will read either one wbuf or 2 wbufs. */
540 len = c->wbuf_pagesize - (bufstart - buf);
541 if (JFFS2_MIN_NODE_HEADER + (int)(bufstart - buf) > c->wbuf_pagesize) {
542 /* The header spans the border of the first wbuf */
543 len += c->wbuf_pagesize;
544 }
545 } else {
546 bufstart = buf;
547 len = JFFS2_MIN_NODE_HEADER;
548 }
550 dbg_readinode("read %d bytes at %#08x(%d).\n", len, ref_offset(ref), ref_flags(ref));
552 /* FIXME: point() */
553 err = jffs2_flash_read(c, ref_offset(ref), len,
554 &retlen, bufstart);
555 if (err) {
556 JFFS2_ERROR("can not read %d bytes from 0x%08x, " "error code: %d.\n", len, ref_offset(ref), err);
557 goto free_out;
558 }
560 if (retlen < len) {
561 JFFS2_ERROR("short read at %#08x: %zu instead of %d.\n", ref_offset(ref), retlen, len);
562 err = -EIO;
563 goto free_out;
564 }
566 node = (union jffs2_node_union *)bufstart;
568 /* No need to mask in the valid bit; it shouldn't be invalid */
569 if (je32_to_cpu(node->u.hdr_crc) != crc32(0, node, sizeof(node->u)-4)) {
570 JFFS2_NOTICE("Node header CRC failed at %#08x. {%04x,%04x,%08x,%08x}\n",
571 ref_offset(ref), je16_to_cpu(node->u.magic),
572 je16_to_cpu(node->u.nodetype),
573 je32_to_cpu(node->u.totlen),
574 je32_to_cpu(node->u.hdr_crc));
575 jffs2_dbg_dump_node(c, ref_offset(ref));
576 jffs2_mark_node_obsolete(c, ref);
577 goto cont;
578 }
580 switch (je16_to_cpu(node->u.nodetype)) {
582 case JFFS2_NODETYPE_DIRENT:
584 if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_raw_dirent)) {
585 err = read_more(c, ref, sizeof(struct jffs2_raw_dirent), &len, buf, bufstart);
586 if (unlikely(err))
587 goto free_out;
588 }
590 err = read_direntry(c, ref, &node->d, retlen, &ret_fd, latest_mctime, mctime_ver);
591 if (err == 1) {
592 jffs2_mark_node_obsolete(c, ref);
593 break;
594 } else if (unlikely(err))
595 goto free_out;
597 if (je32_to_cpu(node->d.version) > *highest_version)
598 *highest_version = je32_to_cpu(node->d.version);
600 break;
602 case JFFS2_NODETYPE_INODE:
604 if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_raw_inode)) {
605 err = read_more(c, ref, sizeof(struct jffs2_raw_inode), &len, buf, bufstart);
606 if (unlikely(err))
607 goto free_out;
608 }
610 err = read_dnode(c, ref, &node->i, &ret_tn, len, latest_mctime, mctime_ver);
611 if (err == 1) {
612 jffs2_mark_node_obsolete(c, ref);
613 break;
614 } else if (unlikely(err))
615 goto free_out;
617 if (je32_to_cpu(node->i.version) > *highest_version)
618 *highest_version = je32_to_cpu(node->i.version);
620 break;
622 default:
623 if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_unknown_node)) {
624 err = read_more(c, ref, sizeof(struct jffs2_unknown_node), &len, buf, bufstart);
625 if (unlikely(err))
626 goto free_out;
627 }
629 err = read_unknown(c, ref, &node->u);
630 if (err == 1) {
631 jffs2_mark_node_obsolete(c, ref);
632 break;
633 } else if (unlikely(err))
634 goto free_out;
636 }
637 cont:
638 spin_lock(&c->erase_completion_lock);
639 }
641 spin_unlock(&c->erase_completion_lock);
642 *tnp = ret_tn;
643 *fdp = ret_fd;
644 kfree(buf);
646 dbg_readinode("nodes of inode #%u were read, the highest version is %u, latest_mctime %u, mctime_ver %u.\n",
647 f->inocache->ino, *highest_version, *latest_mctime, *mctime_ver);
648 return 0;
650 free_out:
651 jffs2_free_tmp_dnode_info_list(&ret_tn);
652 jffs2_free_full_dirent_list(ret_fd);
653 kfree(buf);
654 return err;
655 }
657 static int jffs2_do_read_inode_internal(struct jffs2_sb_info *c,
658 struct jffs2_inode_info *f,
659 struct jffs2_raw_inode *latest_node)
660 {
661 struct jffs2_tmp_dnode_info *tn;
662 struct rb_root tn_list;
663 struct rb_node *rb, *repl_rb;
664 struct jffs2_full_dirent *fd_list;
665 struct jffs2_full_dnode *fn, *first_fn = NULL;
666 uint32_t crc;
667 uint32_t latest_mctime, mctime_ver;
668 size_t retlen;
669 int ret;
671 dbg_readinode("ino #%u nlink is %d\n", f->inocache->ino, f->inocache->nlink);
673 /* Grab all nodes relevant to this ino */
674 ret = jffs2_get_inode_nodes(c, f, &tn_list, &fd_list, &f->highest_version, &latest_mctime, &mctime_ver);
676 if (ret) {
677 JFFS2_ERROR("cannot read nodes for ino %u, returned error is %d\n", f->inocache->ino, ret);
678 if (f->inocache->state == INO_STATE_READING)
679 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
680 return ret;
681 }
682 f->dents = fd_list;
684 rb = rb_first(&tn_list);
686 while (rb) {
687 cond_resched();
688 tn = rb_entry(rb, struct jffs2_tmp_dnode_info, rb);
689 fn = tn->fn;
690 ret = 1;
691 dbg_readinode("consider node ver %u, phys offset "
692 "%#08x(%d), range %u-%u.\n", tn->version,
693 ref_offset(fn->raw), ref_flags(fn->raw),
694 fn->ofs, fn->ofs + fn->size);
696 if (fn->size) {
697 ret = jffs2_add_older_frag_to_fragtree(c, f, tn);
698 /* TODO: the error code isn't checked, check it */
699 jffs2_dbg_fragtree_paranoia_check_nolock(f);
700 BUG_ON(ret < 0);
701 if (!first_fn && ret == 0)
702 first_fn = fn;
703 } else if (!first_fn) {
704 first_fn = fn;
705 f->metadata = fn;
706 ret = 0; /* Prevent freeing the metadata update node */
707 } else
708 jffs2_mark_node_obsolete(c, fn->raw);
710 BUG_ON(rb->rb_left);
711 if (rb_parent(rb) && rb_parent(rb)->rb_left == rb) {
712 /* We were then left-hand child of our parent. We need
713 * to move our own right-hand child into our place. */
714 repl_rb = rb->rb_right;
715 if (repl_rb)
716 rb_set_parent(repl_rb, rb_parent(rb));
717 } else
718 repl_rb = NULL;
720 rb = rb_next(rb);
722 /* Remove the spent tn from the tree; don't bother rebalancing
723 * but put our right-hand child in our own place. */
724 if (rb_parent(&tn->rb)) {
725 if (rb_parent(&tn->rb)->rb_left == &tn->rb)
726 rb_parent(&tn->rb)->rb_left = repl_rb;
727 else if (rb_parent(&tn->rb)->rb_right == &tn->rb)
728 rb_parent(&tn->rb)->rb_right = repl_rb;
729 else BUG();
730 } else if (tn->rb.rb_right)
731 rb_set_parent(tn->rb.rb_right, NULL);
733 jffs2_free_tmp_dnode_info(tn);
734 if (ret) {
735 dbg_readinode("delete dnode %u-%u.\n",
736 fn->ofs, fn->ofs + fn->size);
737 jffs2_free_full_dnode(fn);
738 }
739 }
740 jffs2_dbg_fragtree_paranoia_check_nolock(f);
742 BUG_ON(first_fn && ref_obsolete(first_fn->raw));
744 fn = first_fn;
745 if (unlikely(!first_fn)) {
746 /* No data nodes for this inode. */
747 if (f->inocache->ino != 1) {
748 JFFS2_WARNING("no data nodes found for ino #%u\n", f->inocache->ino);
749 if (!fd_list) {
750 if (f->inocache->state == INO_STATE_READING)
751 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
752 return -EIO;
753 }
754 JFFS2_NOTICE("but it has children so we fake some modes for it\n");
755 }
756 latest_node->mode = cpu_to_jemode(S_IFDIR|S_IRUGO|S_IWUSR|S_IXUGO);
757 latest_node->version = cpu_to_je32(0);
758 latest_node->atime = latest_node->ctime = latest_node->mtime = cpu_to_je32(0);
759 latest_node->isize = cpu_to_je32(0);
760 latest_node->gid = cpu_to_je16(0);
761 latest_node->uid = cpu_to_je16(0);
762 if (f->inocache->state == INO_STATE_READING)
763 jffs2_set_inocache_state(c, f->inocache, INO_STATE_PRESENT);
764 return 0;
765 }
767 ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(*latest_node), &retlen, (void *)latest_node);
768 if (ret || retlen != sizeof(*latest_node)) {
769 JFFS2_ERROR("failed to read from flash: error %d, %zd of %zd bytes read\n",
770 ret, retlen, sizeof(*latest_node));
771 /* FIXME: If this fails, there seems to be a memory leak. Find it. */
772 up(&f->sem);
773 jffs2_do_clear_inode(c, f);
774 return ret?ret:-EIO;
775 }
777 crc = crc32(0, latest_node, sizeof(*latest_node)-8);
778 if (crc != je32_to_cpu(latest_node->node_crc)) {
779 JFFS2_ERROR("CRC failed for read_inode of inode %u at physical location 0x%x\n",
780 f->inocache->ino, ref_offset(fn->raw));
781 up(&f->sem);
782 jffs2_do_clear_inode(c, f);
783 return -EIO;
784 }
786 switch(jemode_to_cpu(latest_node->mode) & S_IFMT) {
787 case S_IFDIR:
788 if (mctime_ver > je32_to_cpu(latest_node->version)) {
789 /* The times in the latest_node are actually older than
790 mctime in the latest dirent. Cheat. */
791 latest_node->ctime = latest_node->mtime = cpu_to_je32(latest_mctime);
792 }
793 break;
796 case S_IFREG:
797 /* If it was a regular file, truncate it to the latest node's isize */
798 jffs2_truncate_fragtree(c, &f->fragtree, je32_to_cpu(latest_node->isize));
799 break;
801 case S_IFLNK:
802 /* Hack to work around broken isize in old symlink code.
803 Remove this when dwmw2 comes to his senses and stops
804 symlinks from being an entirely gratuitous special
805 case. */
806 if (!je32_to_cpu(latest_node->isize))
807 latest_node->isize = latest_node->dsize;
809 if (f->inocache->state != INO_STATE_CHECKING) {
810 /* Symlink's inode data is the target path. Read it and
811 * keep in RAM to facilitate quick follow symlink
812 * operation. */
813 f->target = kmalloc(je32_to_cpu(latest_node->csize) + 1, GFP_KERNEL);
814 if (!f->target) {
815 JFFS2_ERROR("can't allocate %d bytes of memory for the symlink target path cache\n", je32_to_cpu(latest_node->csize));
816 up(&f->sem);
817 jffs2_do_clear_inode(c, f);
818 return -ENOMEM;
819 }
821 ret = jffs2_flash_read(c, ref_offset(fn->raw) + sizeof(*latest_node),
822 je32_to_cpu(latest_node->csize), &retlen, (char *)f->target);
824 if (ret || retlen != je32_to_cpu(latest_node->csize)) {
825 if (retlen != je32_to_cpu(latest_node->csize))
826 ret = -EIO;
827 kfree(f->target);
828 f->target = NULL;
829 up(&f->sem);
830 jffs2_do_clear_inode(c, f);
831 return -ret;
832 }
834 f->target[je32_to_cpu(latest_node->csize)] = '\0';
835 dbg_readinode("symlink's target '%s' cached\n", f->target);
836 }
838 /* fall through... */
840 case S_IFBLK:
841 case S_IFCHR:
842 /* Certain inode types should have only one data node, and it's
843 kept as the metadata node */
844 if (f->metadata) {
845 JFFS2_ERROR("Argh. Special inode #%u with mode 0%o had metadata node\n",
846 f->inocache->ino, jemode_to_cpu(latest_node->mode));
847 up(&f->sem);
848 jffs2_do_clear_inode(c, f);
849 return -EIO;
850 }
851 if (!frag_first(&f->fragtree)) {
852 JFFS2_ERROR("Argh. Special inode #%u with mode 0%o has no fragments\n",
853 f->inocache->ino, jemode_to_cpu(latest_node->mode));
854 up(&f->sem);
855 jffs2_do_clear_inode(c, f);
856 return -EIO;
857 }
858 /* ASSERT: f->fraglist != NULL */
859 if (frag_next(frag_first(&f->fragtree))) {
860 JFFS2_ERROR("Argh. Special inode #%u with mode 0x%x had more than one node\n",
861 f->inocache->ino, jemode_to_cpu(latest_node->mode));
862 /* FIXME: Deal with it - check crc32, check for duplicate node, check times and discard the older one */
863 up(&f->sem);
864 jffs2_do_clear_inode(c, f);
865 return -EIO;
866 }
867 /* OK. We're happy */
868 f->metadata = frag_first(&f->fragtree)->node;
869 jffs2_free_node_frag(frag_first(&f->fragtree));
870 f->fragtree = RB_ROOT;
871 break;
872 }
873 if (f->inocache->state == INO_STATE_READING)
874 jffs2_set_inocache_state(c, f->inocache, INO_STATE_PRESENT);
876 return 0;
877 }
879 /* Scan the list of all nodes present for this ino, build map of versions, etc. */
880 int jffs2_do_read_inode(struct jffs2_sb_info *c, struct jffs2_inode_info *f,
881 uint32_t ino, struct jffs2_raw_inode *latest_node)
882 {
883 dbg_readinode("read inode #%u\n", ino);
885 retry_inocache:
886 spin_lock(&c->inocache_lock);
887 f->inocache = jffs2_get_ino_cache(c, ino);
889 if (f->inocache) {
890 /* Check its state. We may need to wait before we can use it */
891 switch(f->inocache->state) {
892 case INO_STATE_UNCHECKED:
893 case INO_STATE_CHECKEDABSENT:
894 f->inocache->state = INO_STATE_READING;
895 break;
897 case INO_STATE_CHECKING:
898 case INO_STATE_GC:
899 /* If it's in either of these states, we need
900 to wait for whoever's got it to finish and
901 put it back. */
902 dbg_readinode("waiting for ino #%u in state %d\n", ino, f->inocache->state);
903 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
904 goto retry_inocache;
906 case INO_STATE_READING:
907 case INO_STATE_PRESENT:
908 /* Eep. This should never happen. It can
909 happen if Linux calls read_inode() again
910 before clear_inode() has finished though. */
911 JFFS2_ERROR("Eep. Trying to read_inode #%u when it's already in state %d!\n", ino, f->inocache->state);
912 /* Fail. That's probably better than allowing it to succeed */
913 f->inocache = NULL;
914 break;
916 default:
917 BUG();
918 }
919 }
920 spin_unlock(&c->inocache_lock);
922 if (!f->inocache && ino == 1) {
923 /* Special case - no root inode on medium */
924 f->inocache = jffs2_alloc_inode_cache();
925 if (!f->inocache) {
926 JFFS2_ERROR("cannot allocate inocache for root inode\n");
927 return -ENOMEM;
928 }
929 dbg_readinode("creating inocache for root inode\n");
930 memset(f->inocache, 0, sizeof(struct jffs2_inode_cache));
931 f->inocache->ino = f->inocache->nlink = 1;
932 f->inocache->nodes = (struct jffs2_raw_node_ref *)f->inocache;
933 f->inocache->state = INO_STATE_READING;
934 jffs2_add_ino_cache(c, f->inocache);
935 }
936 if (!f->inocache) {
937 JFFS2_ERROR("requestied to read an nonexistent ino %u\n", ino);
938 return -ENOENT;
939 }
941 return jffs2_do_read_inode_internal(c, f, latest_node);
942 }
944 int jffs2_do_crccheck_inode(struct jffs2_sb_info *c, struct jffs2_inode_cache *ic)
945 {
946 struct jffs2_raw_inode n;
947 struct jffs2_inode_info *f = kmalloc(sizeof(*f), GFP_KERNEL);
948 int ret;
950 if (!f)
951 return -ENOMEM;
953 memset(f, 0, sizeof(*f));
954 init_MUTEX_LOCKED(&f->sem);
955 f->inocache = ic;
957 ret = jffs2_do_read_inode_internal(c, f, &n);
958 if (!ret) {
959 up(&f->sem);
960 jffs2_do_clear_inode(c, f);
961 }
962 kfree (f);
963 return ret;
964 }
966 void jffs2_do_clear_inode(struct jffs2_sb_info *c, struct jffs2_inode_info *f)
967 {
968 struct jffs2_full_dirent *fd, *fds;
969 int deleted;
971 jffs2_clear_acl(f);
972 jffs2_xattr_delete_inode(c, f->inocache);
973 down(&f->sem);
974 deleted = f->inocache && !f->inocache->nlink;
976 if (f->inocache && f->inocache->state != INO_STATE_CHECKING)
977 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CLEARING);
979 if (f->metadata) {
980 if (deleted)
981 jffs2_mark_node_obsolete(c, f->metadata->raw);
982 jffs2_free_full_dnode(f->metadata);
983 }
985 jffs2_kill_fragtree(&f->fragtree, deleted?c:NULL);
987 if (f->target) {
988 kfree(f->target);
989 f->target = NULL;
990 }
992 fds = f->dents;
993 while(fds) {
994 fd = fds;
995 fds = fd->next;
996 jffs2_free_full_dirent(fd);
997 }
999 if (f->inocache && f->inocache->state != INO_STATE_CHECKING) {
1000 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1001 if (f->inocache->nodes == (void *)f->inocache)
1002 jffs2_del_ino_cache(c, f->inocache);
1005 up(&f->sem);