3 * Copyright (C) 1992, 1993, 1994, 1995
4 * Remy Card (card@masi.ibp.fr)
5 * Laboratoire MASI - Institut Blaise Pascal
6 * Universite Pierre et Marie Curie (Paris VI)
10 * linux/fs/minix/dir.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * ext2 directory handling functions
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
20 * All code that works with directory layout had been switched to pagecache
23 * Adapted for Lustre Light
24 * Copyright (C) 2002, Cluster File Systems, Inc.
29 #include <linux/ext2_fs.h>
30 #include <linux/pagemap.h>
32 #include <linux/locks.h>
33 #include <asm/uaccess.h>
35 #define DEBUG_SUBSYSTEM S_LLIGHT
37 #include <linux/obd_support.h>
38 #include <linux/lustre_lib.h>
39 #include <linux/lustre_idl.h>
40 #include <linux/lustre_mds.h>
41 #include <linux/lustre_light.h>
43 typedef struct ext2_dir_entry_2 ext2_dirent;
45 #define PageChecked(page) test_bit(PG_checked, &(page)->flags)
46 #define SetPageChecked(page) set_bit(PG_checked, &(page)->flags)
49 static int ll_dir_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
54 /* returns the page unlocked, but with a reference */
55 static int ll_dir_readpage(struct file *file, struct page *page)
57 struct inode *inode = page->mapping->host;
58 struct ll_sb_info *sbi = ll_i2sbi(inode);
62 struct ptlrep_hdr *hdr;
66 if ( ((inode->i_size + PAGE_CACHE_SIZE -1)>>PAGE_SHIFT)
68 memset(kmap(page), 0, PAGE_CACHE_SIZE);
73 if (Page_Uptodate(page)) {
78 offset = page->index << PAGE_SHIFT;
80 rc = mdc_readpage(&sbi->ll_mds_client, inode->i_ino, S_IFDIR, offset,
88 if ((rc = hdr->status)) {
93 /* PDEBUG(page, "READ"); */
95 SetPageUptodate(page);
97 obd_unlock_page(page);
100 } /* ll_dir_readpage */
102 struct address_space_operations ll_dir_aops = {
103 readpage: ll_dir_readpage,
104 prepare_write: ll_dir_prepare_write
107 int waitfor_one_page(struct page *page)
110 struct buffer_head *bh, *head = page->buffers;
115 if (buffer_req(bh) && !buffer_uptodate(bh))
117 } while ((bh = bh->b_this_page) != head);
122 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
123 * more robust, but we have what we have
125 static inline unsigned ext2_chunk_size(struct inode *inode)
127 //return inode->i_sb->s_blocksize;
131 static inline void ext2_put_page(struct page *page)
134 page_cache_release(page);
137 static inline unsigned long dir_pages(struct inode *inode)
139 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
142 extern void set_page_clean(struct page *page);
144 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
146 struct inode *dir = page->mapping->host;
149 dir->i_version = ++event;
150 dir->i_size = (page->index << PAGE_CACHE_SHIFT) + to;
151 SetPageUptodate(page);
152 set_page_clean(page);
154 //page->mapping->a_ops->commit_write(NULL, page, from, to);
156 // err = waitfor_one_page(page);
160 static void ext2_check_page(struct page *page)
162 struct inode *dir = page->mapping->host;
163 unsigned chunk_size = ext2_chunk_size(dir);
164 char *kaddr = page_address(page);
165 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
166 unsigned offs, rec_len;
167 unsigned limit = PAGE_CACHE_SIZE;
171 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
172 limit = dir->i_size & ~PAGE_CACHE_MASK;
173 if (limit & (chunk_size - 1)) {
174 CERROR("limit %d dir size %lld index %ld\n",
175 limit, dir->i_size, page->index);
178 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
179 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
180 p->rec_len = cpu_to_le16(chunk_size);
185 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
186 p = (ext2_dirent *)(kaddr + offs);
187 rec_len = le16_to_cpu(p->rec_len);
189 if (rec_len < EXT2_DIR_REC_LEN(1))
193 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
195 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
197 // if (le32_to_cpu(p->inode) > max_inumber)
203 SetPageChecked(page);
206 /* Too bad, we had an error */
209 CERROR("ext2_check_page"
210 "size of directory #%lu is not a multiple of chunk size\n",
215 error = "rec_len is smaller than minimal";
218 error = "unaligned directory entry";
221 error = "rec_len is too small for name_len";
224 error = "directory entry across blocks";
227 // error = "inode out of bounds";
229 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
230 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
231 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
232 (unsigned long) le32_to_cpu(p->inode),
233 rec_len, p->name_len);
236 p = (ext2_dirent *)(kaddr + offs);
237 CERROR("ext2_check_page"
238 "entry in directory #%lu spans the page boundary"
239 "offset=%lu, inode=%lu",
240 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
241 (unsigned long) le32_to_cpu(p->inode));
243 SetPageChecked(page);
247 static struct page * ext2_get_page(struct inode *dir, unsigned long n)
249 struct address_space *mapping = dir->i_mapping;
250 struct page *page = read_cache_page(mapping, n,
251 (filler_t*)mapping->a_ops->readpage, NULL);
255 if (!Page_Uptodate(page))
257 if (!PageChecked(page))
258 ext2_check_page(page);
266 return ERR_PTR(-EIO);
270 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
272 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
274 static inline int ext2_match (int len, const char * const name,
275 struct ext2_dir_entry_2 * de)
277 if (len != de->name_len)
281 return !memcmp(name, de->name, len);
285 * p is at least 6 bytes before the end of page
287 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
289 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
292 static inline unsigned
293 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
295 ext2_dirent *de = (ext2_dirent*)(base + offset);
296 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
297 while ((char*)p < (char*)de)
298 p = ext2_next_entry(p);
299 return (char *)p - base;
302 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
303 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
304 [EXT2_FT_REG_FILE] DT_REG,
305 [EXT2_FT_DIR] DT_DIR,
306 [EXT2_FT_CHRDEV] DT_CHR,
307 [EXT2_FT_BLKDEV] DT_BLK,
308 [EXT2_FT_FIFO] DT_FIFO,
309 [EXT2_FT_SOCK] DT_SOCK,
310 [EXT2_FT_SYMLINK] DT_LNK,
313 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
315 [EXT2_FT_REG_FILE] S_IFREG,
316 [EXT2_FT_DIR] S_IFDIR,
317 [EXT2_FT_CHRDEV] S_IFCHR,
318 [EXT2_FT_BLKDEV] S_IFBLK,
319 [EXT2_FT_FIFO] S_IFIFO,
320 [EXT2_FT_SOCK] S_IFSOCK,
321 [EXT2_FT_SYMLINK] S_IFLNK
325 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
326 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
327 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
328 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
329 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
330 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
331 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
332 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
335 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
337 mode_t mode = inode->i_mode;
338 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
342 new_ll_readdir (struct file * filp, void * dirent, filldir_t filldir)
344 loff_t pos = filp->f_pos;
345 struct inode *inode = filp->f_dentry->d_inode;
346 // XXX struct super_block *sb = inode->i_sb;
347 unsigned offset = pos & ~PAGE_CACHE_MASK;
348 unsigned long n = pos >> PAGE_CACHE_SHIFT;
349 unsigned long npages = dir_pages(inode);
350 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
351 unsigned char *types = NULL;
352 int need_revalidate = (filp->f_version != inode->i_version);
354 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
357 types = ext2_filetype_table;
359 for ( ; n < npages; n++, offset = 0) {
362 struct page *page = ext2_get_page(inode, n);
366 kaddr = page_address(page);
367 if (need_revalidate) {
368 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
371 de = (ext2_dirent *)(kaddr+offset);
372 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
373 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
376 unsigned char d_type = DT_UNKNOWN;
378 if (types && de->file_type < EXT2_FT_MAX)
379 d_type = types[de->file_type];
381 offset = (char *)de - kaddr;
382 over = filldir(dirent, de->name, de->name_len,
383 (n<<PAGE_CACHE_SHIFT) | offset,
384 le32_to_cpu(de->inode), d_type);
394 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
395 filp->f_version = inode->i_version;
403 * finds an entry in the specified directory with the wanted name. It
404 * returns the page in which the entry was found, and the entry itself
405 * (as a parameter - res_dir). Page is returned mapped and unlocked.
406 * Entry is guaranteed to be valid.
408 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
409 struct dentry *dentry, struct page ** res_page)
411 const char *name = dentry->d_name.name;
412 int namelen = dentry->d_name.len;
413 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
414 unsigned long start, n;
415 unsigned long npages = dir_pages(dir);
416 struct page *page = NULL;
422 // start = dir->u.ext2_i.i_dir_start_lookup;
429 page = ext2_get_page(dir, n);
431 kaddr = page_address(page);
432 de = (ext2_dirent *) kaddr;
433 kaddr += PAGE_CACHE_SIZE - reclen;
434 while ((char *) de <= kaddr) {
435 if (ext2_match (namelen, name, de))
437 de = ext2_next_entry(de);
443 } while (n != start);
448 // dir->u.ext2_i.i_dir_start_lookup = n;
452 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
454 struct page *page = ext2_get_page(dir, 0);
455 ext2_dirent *de = NULL;
458 de = ext2_next_entry((ext2_dirent *) page_address(page));
464 ino_t ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
467 struct ext2_dir_entry_2 * de;
470 de = ext2_find_entry (dir, dentry, &page);
472 res = le32_to_cpu(de->inode);
473 *type = ll_dt2fmt[de->file_type];
475 page_cache_release(page);
480 /* Releases the page */
481 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
482 struct page *page, struct inode *inode)
484 unsigned from = (char *) de - (char *) page_address(page);
485 unsigned to = from + le16_to_cpu(de->rec_len);
489 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
492 de->inode = cpu_to_le32(inode->i_ino);
493 ext2_set_de_type (de, inode);
494 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
495 err = ext2_commit_chunk(page, from, to);
503 int ll_add_link (struct dentry *dentry, struct inode *inode)
505 struct inode *dir = dentry->d_parent->d_inode;
506 const char *name = dentry->d_name.name;
507 int namelen = dentry->d_name.len;
508 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
509 unsigned short rec_len, name_len;
510 struct page *page = NULL;
512 unsigned long npages = dir_pages(dir);
518 /* We take care of directory expansion in the same loop */
519 for (n = 0; n <= npages; n++) {
520 page = ext2_get_page(dir, n);
524 kaddr = page_address(page);
525 de = (ext2_dirent *)kaddr;
526 kaddr += PAGE_CACHE_SIZE - reclen;
527 while ((char *)de <= kaddr) {
529 if (ext2_match (namelen, name, de))
531 name_len = EXT2_DIR_REC_LEN(de->name_len);
532 rec_len = le16_to_cpu(de->rec_len);
533 if ( n==npages && rec_len == 0) {
534 CERROR("Fatal dir behaviour\n");
537 if (!de->inode && rec_len >= reclen)
539 if (rec_len >= name_len + reclen)
541 de = (ext2_dirent *) ((char *) de + rec_len);
549 from = (char*)de - (char*)page_address(page);
552 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
556 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
557 de1->rec_len = cpu_to_le16(rec_len - name_len);
558 de->rec_len = cpu_to_le16(name_len);
561 de->name_len = namelen;
562 memcpy (de->name, name, namelen);
563 de->inode = cpu_to_le32(inode->i_ino);
564 ext2_set_de_type (de, inode);
565 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
566 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
567 err = ext2_commit_chunk(page, from, to);
569 // change_inode happens with the commit_chunk
570 /* XXX OFFSET_CACHE */
581 * ext2_delete_entry deletes a directory entry by merging it with the
582 * previous entry. Page is up-to-date. Releases the page.
584 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
586 struct address_space *mapping = page->mapping;
587 struct inode *inode = mapping->host;
588 char *kaddr = page_address(page);
589 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
590 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
591 ext2_dirent * pde = NULL;
592 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
595 while ((char*)de < (char*)dir) {
597 de = ext2_next_entry(de);
600 from = (char*)pde - (char*)page_address(page);
602 err = mapping->a_ops->prepare_write(NULL, page, from, to);
606 pde->rec_len = cpu_to_le16(to-from);
608 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
609 err = ext2_commit_chunk(page, from, to);
616 * Set the first fragment of directory.
618 int ext2_make_empty(struct inode *inode, struct inode *parent)
620 struct address_space *mapping = inode->i_mapping;
621 struct page *page = grab_cache_page(mapping, 0);
622 unsigned chunk_size = ext2_chunk_size(inode);
623 struct ext2_dir_entry_2 * de;
630 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
634 base = page_address(page);
636 de = (struct ext2_dir_entry_2 *) base;
638 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
639 memcpy (de->name, ".\0\0", 4);
640 de->inode = cpu_to_le32(inode->i_ino);
641 ext2_set_de_type (de, inode);
643 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
645 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
646 de->inode = cpu_to_le32(parent->i_ino);
647 memcpy (de->name, "..\0", 4);
648 ext2_set_de_type (de, inode);
650 err = ext2_commit_chunk(page, 0, chunk_size);
653 page_cache_release(page);
659 * routine to check that the specified directory is empty (for rmdir)
661 int ext2_empty_dir (struct inode * inode)
663 struct page *page = NULL;
664 unsigned long i, npages = dir_pages(inode);
666 for (i = 0; i < npages; i++) {
669 page = ext2_get_page(inode, i);
674 kaddr = page_address(page);
675 de = (ext2_dirent *)kaddr;
676 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
678 while ((char *)de <= kaddr) {
679 if (de->inode != 0) {
680 /* check for . and .. */
681 if (de->name[0] != '.')
683 if (de->name_len > 2)
685 if (de->name_len < 2) {
687 cpu_to_le32(inode->i_ino))
689 } else if (de->name[1] != '.')
692 de = ext2_next_entry(de);
703 struct file_operations ll_dir_operations = {
704 read: generic_read_dir,
705 readdir: new_ll_readdir