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/obd_class.h>
39 #include <linux/lustre_lib.h>
40 #include <linux/lustre_idl.h>
41 #include <linux/lustre_mds.h>
42 #include <linux/lustre_light.h>
44 typedef struct ext2_dir_entry_2 ext2_dirent;
46 #define PageChecked(page) test_bit(PG_checked, &(page)->flags)
47 #define SetPageChecked(page) set_bit(PG_checked, &(page)->flags)
50 static int ll_dir_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
55 /* returns the page unlocked, but with a reference */
56 static int ll_dir_readpage(struct file *file, struct page *page)
58 struct inode *inode = page->mapping->host;
59 struct ll_sb_info *sbi = ll_i2sbi(inode);
63 struct ptlrep_hdr *hdr;
67 if ( ((inode->i_size + PAGE_CACHE_SIZE -1)>>PAGE_SHIFT)
69 memset(kmap(page), 0, PAGE_CACHE_SIZE);
75 if (Page_Uptodate(page)) {
80 offset = page->index << PAGE_SHIFT;
82 rc = mdc_readpage(&sbi->ll_mds_client, inode->i_ino, S_IFDIR, offset,
90 if ((rc = hdr->status)) {
95 /* PDEBUG(page, "READ"); */
97 SetPageUptodate(page);
102 } /* ll_dir_readpage */
104 struct address_space_operations ll_dir_aops = {
105 readpage: ll_dir_readpage,
106 prepare_write: ll_dir_prepare_write
109 int waitfor_one_page(struct page *page)
112 struct buffer_head *bh, *head = page->buffers;
117 if (buffer_req(bh) && !buffer_uptodate(bh))
119 } while ((bh = bh->b_this_page) != head);
124 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
125 * more robust, but we have what we have
127 static inline unsigned ext2_chunk_size(struct inode *inode)
129 //return inode->i_sb->s_blocksize;
133 static inline void ext2_put_page(struct page *page)
136 page_cache_release(page);
139 static inline unsigned long dir_pages(struct inode *inode)
141 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
144 extern void set_page_clean(struct page *page);
146 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
148 struct inode *dir = page->mapping->host;
151 dir->i_version = ++event;
152 dir->i_size = (page->index << PAGE_CACHE_SHIFT) + to;
153 SetPageUptodate(page);
154 set_page_clean(page);
156 //page->mapping->a_ops->commit_write(NULL, page, from, to);
158 // err = waitfor_one_page(page);
162 static void ext2_check_page(struct page *page)
164 struct inode *dir = page->mapping->host;
165 unsigned chunk_size = ext2_chunk_size(dir);
166 char *kaddr = page_address(page);
167 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
168 unsigned offs, rec_len;
169 unsigned limit = PAGE_CACHE_SIZE;
173 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
174 limit = dir->i_size & ~PAGE_CACHE_MASK;
175 if (limit & (chunk_size - 1)) {
176 CERROR("limit %d dir size %lld index %ld\n",
177 limit, dir->i_size, page->index);
180 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
181 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
182 p->rec_len = cpu_to_le16(chunk_size);
187 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
188 p = (ext2_dirent *)(kaddr + offs);
189 rec_len = le16_to_cpu(p->rec_len);
191 if (rec_len < EXT2_DIR_REC_LEN(1))
195 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
197 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
199 // if (le32_to_cpu(p->inode) > max_inumber)
205 SetPageChecked(page);
208 /* Too bad, we had an error */
211 CERROR("ext2_check_page"
212 "size of directory #%lu is not a multiple of chunk size\n",
217 error = "rec_len is smaller than minimal";
220 error = "unaligned directory entry";
223 error = "rec_len is too small for name_len";
226 error = "directory entry across blocks";
229 // error = "inode out of bounds";
231 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
232 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
233 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
234 (unsigned long) le32_to_cpu(p->inode),
235 rec_len, p->name_len);
238 p = (ext2_dirent *)(kaddr + offs);
239 CERROR("ext2_check_page"
240 "entry in directory #%lu spans the page boundary"
241 "offset=%lu, inode=%lu",
242 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
243 (unsigned long) le32_to_cpu(p->inode));
245 SetPageChecked(page);
249 static struct page * ext2_get_page(struct inode *dir, unsigned long n)
251 struct address_space *mapping = dir->i_mapping;
252 struct page *page = read_cache_page(mapping, n,
253 (filler_t*)mapping->a_ops->readpage, NULL);
257 if (!Page_Uptodate(page))
259 if (!PageChecked(page))
260 ext2_check_page(page);
268 return ERR_PTR(-EIO);
272 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
274 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
276 static inline int ext2_match (int len, const char * const name,
277 struct ext2_dir_entry_2 * de)
279 if (len != de->name_len)
283 return !memcmp(name, de->name, len);
287 * p is at least 6 bytes before the end of page
289 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
291 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
294 static inline unsigned
295 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
297 ext2_dirent *de = (ext2_dirent*)(base + offset);
298 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
299 while ((char*)p < (char*)de)
300 p = ext2_next_entry(p);
301 return (char *)p - base;
304 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
305 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
306 [EXT2_FT_REG_FILE] DT_REG,
307 [EXT2_FT_DIR] DT_DIR,
308 [EXT2_FT_CHRDEV] DT_CHR,
309 [EXT2_FT_BLKDEV] DT_BLK,
310 [EXT2_FT_FIFO] DT_FIFO,
311 [EXT2_FT_SOCK] DT_SOCK,
312 [EXT2_FT_SYMLINK] DT_LNK,
315 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
317 [EXT2_FT_REG_FILE] S_IFREG,
318 [EXT2_FT_DIR] S_IFDIR,
319 [EXT2_FT_CHRDEV] S_IFCHR,
320 [EXT2_FT_BLKDEV] S_IFBLK,
321 [EXT2_FT_FIFO] S_IFIFO,
322 [EXT2_FT_SOCK] S_IFSOCK,
323 [EXT2_FT_SYMLINK] S_IFLNK
327 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
328 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
329 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
330 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
331 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
332 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
333 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
334 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
337 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
339 mode_t mode = inode->i_mode;
340 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
344 new_ll_readdir (struct file * filp, void * dirent, filldir_t filldir)
346 loff_t pos = filp->f_pos;
347 struct inode *inode = filp->f_dentry->d_inode;
348 // XXX struct super_block *sb = inode->i_sb;
349 unsigned offset = pos & ~PAGE_CACHE_MASK;
350 unsigned long n = pos >> PAGE_CACHE_SHIFT;
351 unsigned long npages = dir_pages(inode);
352 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
353 unsigned char *types = NULL;
354 int need_revalidate = (filp->f_version != inode->i_version);
356 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
359 types = ext2_filetype_table;
361 for ( ; n < npages; n++, offset = 0) {
364 struct page *page = ext2_get_page(inode, n);
368 kaddr = page_address(page);
369 if (need_revalidate) {
370 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
373 de = (ext2_dirent *)(kaddr+offset);
374 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
375 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
378 unsigned char d_type = DT_UNKNOWN;
380 if (types && de->file_type < EXT2_FT_MAX)
381 d_type = types[de->file_type];
383 offset = (char *)de - kaddr;
384 over = filldir(dirent, de->name, de->name_len,
385 (n<<PAGE_CACHE_SHIFT) | offset,
386 le32_to_cpu(de->inode), d_type);
396 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
397 filp->f_version = inode->i_version;
405 * finds an entry in the specified directory with the wanted name. It
406 * returns the page in which the entry was found, and the entry itself
407 * (as a parameter - res_dir). Page is returned mapped and unlocked.
408 * Entry is guaranteed to be valid.
410 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
411 struct dentry *dentry, struct page ** res_page)
413 const char *name = dentry->d_name.name;
414 int namelen = dentry->d_name.len;
415 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
416 unsigned long start, n;
417 unsigned long npages = dir_pages(dir);
418 struct page *page = NULL;
424 // start = dir->u.ext2_i.i_dir_start_lookup;
431 page = ext2_get_page(dir, n);
433 kaddr = page_address(page);
434 de = (ext2_dirent *) kaddr;
435 kaddr += PAGE_CACHE_SIZE - reclen;
436 while ((char *) de <= kaddr) {
437 if (ext2_match (namelen, name, de))
439 de = ext2_next_entry(de);
445 } while (n != start);
450 // dir->u.ext2_i.i_dir_start_lookup = n;
454 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
456 struct page *page = ext2_get_page(dir, 0);
457 ext2_dirent *de = NULL;
460 de = ext2_next_entry((ext2_dirent *) page_address(page));
466 ino_t ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
469 struct ext2_dir_entry_2 * de;
472 de = ext2_find_entry (dir, dentry, &page);
474 res = le32_to_cpu(de->inode);
475 *type = ll_dt2fmt[de->file_type];
477 page_cache_release(page);
482 /* Releases the page */
483 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
484 struct page *page, struct inode *inode)
486 unsigned from = (char *) de - (char *) page_address(page);
487 unsigned to = from + le16_to_cpu(de->rec_len);
491 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
494 de->inode = cpu_to_le32(inode->i_ino);
495 ext2_set_de_type (de, inode);
496 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
497 err = ext2_commit_chunk(page, from, to);
505 int ll_add_link (struct dentry *dentry, struct inode *inode)
507 struct inode *dir = dentry->d_parent->d_inode;
508 const char *name = dentry->d_name.name;
509 int namelen = dentry->d_name.len;
510 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
511 unsigned short rec_len, name_len;
512 struct page *page = NULL;
514 unsigned long npages = dir_pages(dir);
520 /* We take care of directory expansion in the same loop */
521 for (n = 0; n <= npages; n++) {
522 page = ext2_get_page(dir, n);
526 kaddr = page_address(page);
527 de = (ext2_dirent *)kaddr;
528 kaddr += PAGE_CACHE_SIZE - reclen;
529 while ((char *)de <= kaddr) {
531 if (ext2_match (namelen, name, de))
533 name_len = EXT2_DIR_REC_LEN(de->name_len);
534 rec_len = le16_to_cpu(de->rec_len);
535 if ( n==npages && rec_len == 0) {
536 CERROR("Fatal dir behaviour\n");
539 if (!de->inode && rec_len >= reclen)
541 if (rec_len >= name_len + reclen)
543 de = (ext2_dirent *) ((char *) de + rec_len);
551 from = (char*)de - (char*)page_address(page);
554 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
558 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
559 de1->rec_len = cpu_to_le16(rec_len - name_len);
560 de->rec_len = cpu_to_le16(name_len);
563 de->name_len = namelen;
564 memcpy (de->name, name, namelen);
565 de->inode = cpu_to_le32(inode->i_ino);
566 ext2_set_de_type (de, inode);
567 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
568 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
569 err = ext2_commit_chunk(page, from, to);
571 // change_inode happens with the commit_chunk
572 /* XXX OFFSET_CACHE */
583 * ext2_delete_entry deletes a directory entry by merging it with the
584 * previous entry. Page is up-to-date. Releases the page.
586 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
588 struct address_space *mapping = page->mapping;
589 struct inode *inode = mapping->host;
590 char *kaddr = page_address(page);
591 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
592 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
593 ext2_dirent * pde = NULL;
594 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
597 while ((char*)de < (char*)dir) {
599 de = ext2_next_entry(de);
602 from = (char*)pde - (char*)page_address(page);
604 err = mapping->a_ops->prepare_write(NULL, page, from, to);
608 pde->rec_len = cpu_to_le16(to-from);
610 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
611 err = ext2_commit_chunk(page, from, to);
618 * Set the first fragment of directory.
620 int ext2_make_empty(struct inode *inode, struct inode *parent)
622 struct address_space *mapping = inode->i_mapping;
623 struct page *page = grab_cache_page(mapping, 0);
624 unsigned chunk_size = ext2_chunk_size(inode);
625 struct ext2_dir_entry_2 * de;
632 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
636 base = page_address(page);
638 de = (struct ext2_dir_entry_2 *) base;
640 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
641 memcpy (de->name, ".\0\0", 4);
642 de->inode = cpu_to_le32(inode->i_ino);
643 ext2_set_de_type (de, inode);
645 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
647 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
648 de->inode = cpu_to_le32(parent->i_ino);
649 memcpy (de->name, "..\0", 4);
650 ext2_set_de_type (de, inode);
652 err = ext2_commit_chunk(page, 0, chunk_size);
655 page_cache_release(page);
661 * routine to check that the specified directory is empty (for rmdir)
663 int ext2_empty_dir (struct inode * inode)
665 struct page *page = NULL;
666 unsigned long i, npages = dir_pages(inode);
668 for (i = 0; i < npages; i++) {
671 page = ext2_get_page(inode, i);
676 kaddr = page_address(page);
677 de = (ext2_dirent *)kaddr;
678 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
680 while ((char *)de <= kaddr) {
681 if (de->inode != 0) {
682 /* check for . and .. */
683 if (de->name[0] != '.')
685 if (de->name_len > 2)
687 if (de->name_len < 2) {
689 cpu_to_le32(inode->i_ino))
691 } else if (de->name[1] != '.')
694 de = ext2_next_entry(de);
705 struct file_operations ll_dir_operations = {
706 read: generic_read_dir,
707 readdir: new_ll_readdir