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/obd_support.h>
33 #include <linux/locks.h>
34 #include <asm/uaccess.h>
35 #include <linux/lustre_lib.h>
36 #include <linux/lustre_idl.h>
37 #include <linux/lustre_mds.h>
38 #include <linux/lustre_light.h>
40 typedef struct ext2_dir_entry_2 ext2_dirent;
42 #define PageChecked(page) test_bit(PG_checked, &(page)->flags)
43 #define SetPageChecked(page) set_bit(PG_checked, &(page)->flags)
46 static int ll_dir_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
51 /* returns the page unlocked, but with a reference */
52 static int ll_dir_readpage(struct file *file, struct page *page)
54 struct inode *inode = page->mapping->host;
55 struct ll_sb_info *sbi = ll_i2sbi(inode);
59 struct ptlrep_hdr *hdr;
63 if ( ((inode->i_size + PAGE_CACHE_SIZE -1)>>PAGE_SHIFT)
65 memset(kmap(page), 0, PAGE_CACHE_SIZE);
70 if (Page_Uptodate(page)) {
75 offset = page->index << PAGE_SHIFT;
77 rc = mdc_readpage(sbi->ll_peer_ptr, inode->i_ino, S_IFDIR, offset, buf,
85 if ((rc = hdr->status)) {
90 /* PDEBUG(page, "READ"); */
92 SetPageUptodate(page);
94 obd_unlock_page(page);
97 } /* ll_dir_readpage */
99 struct address_space_operations ll_dir_aops = {
100 readpage: ll_dir_readpage,
101 prepare_write: ll_dir_prepare_write
104 int waitfor_one_page(struct page *page)
107 struct buffer_head *bh, *head = page->buffers;
112 if (buffer_req(bh) && !buffer_uptodate(bh))
114 } while ((bh = bh->b_this_page) != head);
119 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
120 * more robust, but we have what we have
122 static inline unsigned ext2_chunk_size(struct inode *inode)
124 //return inode->i_sb->s_blocksize;
128 static inline void ext2_put_page(struct page *page)
131 page_cache_release(page);
134 static inline unsigned long dir_pages(struct inode *inode)
136 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
139 extern void set_page_clean(struct page *page);
141 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
143 struct inode *dir = page->mapping->host;
146 dir->i_version = ++event;
147 dir->i_size = (page->index << PAGE_CACHE_SHIFT) + to;
148 SetPageUptodate(page);
149 set_page_clean(page);
151 //page->mapping->a_ops->commit_write(NULL, page, from, to);
153 // err = waitfor_one_page(page);
157 static void ext2_check_page(struct page *page)
159 struct inode *dir = page->mapping->host;
160 unsigned chunk_size = ext2_chunk_size(dir);
161 char *kaddr = page_address(page);
162 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
163 unsigned offs, rec_len;
164 unsigned limit = PAGE_CACHE_SIZE;
168 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
169 limit = dir->i_size & ~PAGE_CACHE_MASK;
170 if (limit & (chunk_size - 1))
172 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
173 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
174 p->rec_len = cpu_to_le16(chunk_size);
179 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
180 p = (ext2_dirent *)(kaddr + offs);
181 rec_len = le16_to_cpu(p->rec_len);
183 if (rec_len < EXT2_DIR_REC_LEN(1))
187 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
189 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
191 // if (le32_to_cpu(p->inode) > max_inumber)
197 SetPageChecked(page);
200 /* Too bad, we had an error */
203 printk("ext2_check_page"
204 "size of directory #%lu is not a multiple of chunk size",
209 error = "rec_len is smaller than minimal";
212 error = "unaligned directory entry";
215 error = "rec_len is too small for name_len";
218 error = "directory entry across blocks";
221 // error = "inode out of bounds";
223 printk("ext2_check_page: bad entry in directory #%lu: %s - "
224 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
225 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
226 (unsigned long) le32_to_cpu(p->inode),
227 rec_len, p->name_len);
230 p = (ext2_dirent *)(kaddr + offs);
231 printk("ext2_check_page"
232 "entry in directory #%lu spans the page boundary"
233 "offset=%lu, inode=%lu",
234 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
235 (unsigned long) le32_to_cpu(p->inode));
237 SetPageChecked(page);
241 static struct page * ext2_get_page(struct inode *dir, unsigned long n)
243 struct address_space *mapping = dir->i_mapping;
244 struct page *page = read_cache_page(mapping, n,
245 (filler_t*)mapping->a_ops->readpage, NULL);
249 if (!Page_Uptodate(page))
251 if (!PageChecked(page))
252 ext2_check_page(page);
260 return ERR_PTR(-EIO);
264 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
266 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
268 static inline int ext2_match (int len, const char * const name,
269 struct ext2_dir_entry_2 * de)
271 if (len != de->name_len)
275 return !memcmp(name, de->name, len);
279 * p is at least 6 bytes before the end of page
281 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
283 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
286 static inline unsigned
287 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
289 ext2_dirent *de = (ext2_dirent*)(base + offset);
290 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
291 while ((char*)p < (char*)de)
292 p = ext2_next_entry(p);
293 return (char *)p - base;
296 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
297 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
298 [EXT2_FT_REG_FILE] DT_REG,
299 [EXT2_FT_DIR] DT_DIR,
300 [EXT2_FT_CHRDEV] DT_CHR,
301 [EXT2_FT_BLKDEV] DT_BLK,
302 [EXT2_FT_FIFO] DT_FIFO,
303 [EXT2_FT_SOCK] DT_SOCK,
304 [EXT2_FT_SYMLINK] DT_LNK,
307 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
309 [EXT2_FT_REG_FILE] S_IFREG,
310 [EXT2_FT_DIR] S_IFDIR,
311 [EXT2_FT_CHRDEV] S_IFCHR,
312 [EXT2_FT_BLKDEV] S_IFBLK,
313 [EXT2_FT_FIFO] S_IFIFO,
314 [EXT2_FT_SOCK] S_IFSOCK,
315 [EXT2_FT_SYMLINK] S_IFLNK
319 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
320 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
321 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
322 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
323 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
324 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
325 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
326 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
329 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
331 mode_t mode = inode->i_mode;
332 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
336 new_ll_readdir (struct file * filp, void * dirent, filldir_t filldir)
338 loff_t pos = filp->f_pos;
339 struct inode *inode = filp->f_dentry->d_inode;
340 // XXX struct super_block *sb = inode->i_sb;
341 unsigned offset = pos & ~PAGE_CACHE_MASK;
342 unsigned long n = pos >> PAGE_CACHE_SHIFT;
343 unsigned long npages = dir_pages(inode);
344 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
345 unsigned char *types = NULL;
346 int need_revalidate = (filp->f_version != inode->i_version);
348 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
351 types = ext2_filetype_table;
353 for ( ; n < npages; n++, offset = 0) {
356 struct page *page = ext2_get_page(inode, n);
360 kaddr = page_address(page);
361 if (need_revalidate) {
362 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
365 de = (ext2_dirent *)(kaddr+offset);
366 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
367 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
370 unsigned char d_type = DT_UNKNOWN;
372 if (types && de->file_type < EXT2_FT_MAX)
373 d_type = types[de->file_type];
375 offset = (char *)de - kaddr;
376 over = filldir(dirent, de->name, de->name_len,
377 (n<<PAGE_CACHE_SHIFT) | offset,
378 le32_to_cpu(de->inode), d_type);
388 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
389 filp->f_version = inode->i_version;
397 * finds an entry in the specified directory with the wanted name. It
398 * returns the page in which the entry was found, and the entry itself
399 * (as a parameter - res_dir). Page is returned mapped and unlocked.
400 * Entry is guaranteed to be valid.
402 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
403 struct dentry *dentry, struct page ** res_page)
405 const char *name = dentry->d_name.name;
406 int namelen = dentry->d_name.len;
407 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
408 unsigned long start, n;
409 unsigned long npages = dir_pages(dir);
410 struct page *page = NULL;
416 // start = dir->u.ext2_i.i_dir_start_lookup;
423 page = ext2_get_page(dir, n);
425 kaddr = page_address(page);
426 de = (ext2_dirent *) kaddr;
427 kaddr += PAGE_CACHE_SIZE - reclen;
428 while ((char *) de <= kaddr) {
429 if (ext2_match (namelen, name, de))
431 de = ext2_next_entry(de);
437 } while (n != start);
442 // dir->u.ext2_i.i_dir_start_lookup = n;
446 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
448 struct page *page = ext2_get_page(dir, 0);
449 ext2_dirent *de = NULL;
452 de = ext2_next_entry((ext2_dirent *) page_address(page));
458 ino_t ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
461 struct ext2_dir_entry_2 * de;
464 de = ext2_find_entry (dir, dentry, &page);
466 res = le32_to_cpu(de->inode);
467 *type = ll_dt2fmt[de->file_type];
469 page_cache_release(page);
474 /* Releases the page */
475 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
476 struct page *page, struct inode *inode)
478 unsigned from = (char *) de - (char *) page_address(page);
479 unsigned to = from + le16_to_cpu(de->rec_len);
483 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
486 de->inode = cpu_to_le32(inode->i_ino);
487 ext2_set_de_type (de, inode);
488 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
489 err = ext2_commit_chunk(page, from, to);
497 int ll_add_link (struct dentry *dentry, struct inode *inode)
499 struct inode *dir = dentry->d_parent->d_inode;
500 const char *name = dentry->d_name.name;
501 int namelen = dentry->d_name.len;
502 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
503 unsigned short rec_len, name_len;
504 struct page *page = NULL;
506 unsigned long npages = dir_pages(dir);
512 /* We take care of directory expansion in the same loop */
513 for (n = 0; n <= npages; n++) {
514 page = ext2_get_page(dir, n);
518 kaddr = page_address(page);
519 de = (ext2_dirent *)kaddr;
520 kaddr += PAGE_CACHE_SIZE - reclen;
521 while ((char *)de <= kaddr) {
523 if (ext2_match (namelen, name, de))
525 name_len = EXT2_DIR_REC_LEN(de->name_len);
526 rec_len = le16_to_cpu(de->rec_len);
527 if ( n==npages && rec_len == 0) {
528 printk("Fatal dir behaviour\n");
531 if (!de->inode && rec_len >= reclen)
533 if (rec_len >= name_len + reclen)
535 de = (ext2_dirent *) ((char *) de + rec_len);
543 from = (char*)de - (char*)page_address(page);
546 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
550 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
551 de1->rec_len = cpu_to_le16(rec_len - name_len);
552 de->rec_len = cpu_to_le16(name_len);
555 de->name_len = namelen;
556 memcpy (de->name, name, namelen);
557 de->inode = cpu_to_le32(inode->i_ino);
558 ext2_set_de_type (de, inode);
559 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
560 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
561 err = ext2_commit_chunk(page, from, to);
563 // change_inode happens with the commit_chunk
564 /* XXX OFFSET_CACHE */
575 * ext2_delete_entry deletes a directory entry by merging it with the
576 * previous entry. Page is up-to-date. Releases the page.
578 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
580 struct address_space *mapping = page->mapping;
581 struct inode *inode = mapping->host;
582 char *kaddr = page_address(page);
583 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
584 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
585 ext2_dirent * pde = NULL;
586 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
589 while ((char*)de < (char*)dir) {
591 de = ext2_next_entry(de);
594 from = (char*)pde - (char*)page_address(page);
596 err = mapping->a_ops->prepare_write(NULL, page, from, to);
600 pde->rec_len = cpu_to_le16(to-from);
602 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
603 err = ext2_commit_chunk(page, from, to);
610 * Set the first fragment of directory.
612 int ext2_make_empty(struct inode *inode, struct inode *parent)
614 struct address_space *mapping = inode->i_mapping;
615 struct page *page = grab_cache_page(mapping, 0);
616 unsigned chunk_size = ext2_chunk_size(inode);
617 struct ext2_dir_entry_2 * de;
624 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
628 base = page_address(page);
630 de = (struct ext2_dir_entry_2 *) base;
632 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
633 memcpy (de->name, ".\0\0", 4);
634 de->inode = cpu_to_le32(inode->i_ino);
635 ext2_set_de_type (de, inode);
637 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
639 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
640 de->inode = cpu_to_le32(parent->i_ino);
641 memcpy (de->name, "..\0", 4);
642 ext2_set_de_type (de, inode);
644 err = ext2_commit_chunk(page, 0, chunk_size);
647 page_cache_release(page);
653 * routine to check that the specified directory is empty (for rmdir)
655 int ext2_empty_dir (struct inode * inode)
657 struct page *page = NULL;
658 unsigned long i, npages = dir_pages(inode);
660 for (i = 0; i < npages; i++) {
663 page = ext2_get_page(inode, i);
668 kaddr = page_address(page);
669 de = (ext2_dirent *)kaddr;
670 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
672 while ((char *)de <= kaddr) {
673 if (de->inode != 0) {
674 /* check for . and .. */
675 if (de->name[0] != '.')
677 if (de->name_len > 2)
679 if (de->name_len < 2) {
681 cpu_to_le32(inode->i_ino))
683 } else if (de->name[1] != '.')
686 de = ext2_next_entry(de);
697 struct file_operations ll_dir_operations = {
698 read: generic_read_dir,
699 readdir: new_ll_readdir