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 ptlrpc_request *request;
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,
85 ptlrpc_free_req(request);
87 SetPageUptodate(page);
93 } /* ll_dir_readpage */
95 struct address_space_operations ll_dir_aops = {
96 readpage: ll_dir_readpage,
97 prepare_write: ll_dir_prepare_write
100 int waitfor_one_page(struct page *page)
103 struct buffer_head *bh, *head = page->buffers;
108 if (buffer_req(bh) && !buffer_uptodate(bh))
110 } while ((bh = bh->b_this_page) != head);
115 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
116 * more robust, but we have what we have
118 static inline unsigned ext2_chunk_size(struct inode *inode)
120 //return inode->i_sb->s_blocksize;
124 static inline void ext2_put_page(struct page *page)
127 page_cache_release(page);
130 static inline unsigned long dir_pages(struct inode *inode)
132 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
135 extern void set_page_clean(struct page *page);
137 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
139 struct inode *dir = page->mapping->host;
142 dir->i_version = ++event;
143 dir->i_size = (page->index << PAGE_CACHE_SHIFT) + to;
144 SetPageUptodate(page);
145 set_page_clean(page);
147 //page->mapping->a_ops->commit_write(NULL, page, from, to);
149 // err = waitfor_one_page(page);
153 static void ext2_check_page(struct page *page)
155 struct inode *dir = page->mapping->host;
156 unsigned chunk_size = ext2_chunk_size(dir);
157 char *kaddr = page_address(page);
158 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
159 unsigned offs, rec_len;
160 unsigned limit = PAGE_CACHE_SIZE;
164 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
165 limit = dir->i_size & ~PAGE_CACHE_MASK;
166 if (limit & (chunk_size - 1)) {
167 CERROR("limit %d dir size %lld index %ld\n",
168 limit, dir->i_size, page->index);
171 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
172 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
173 p->rec_len = cpu_to_le16(chunk_size);
178 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
179 p = (ext2_dirent *)(kaddr + offs);
180 rec_len = le16_to_cpu(p->rec_len);
182 if (rec_len < EXT2_DIR_REC_LEN(1))
186 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
188 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
190 // if (le32_to_cpu(p->inode) > max_inumber)
196 SetPageChecked(page);
199 /* Too bad, we had an error */
202 CERROR("ext2_check_page"
203 "size of directory #%lu is not a multiple of chunk size\n",
208 error = "rec_len is smaller than minimal";
211 error = "unaligned directory entry";
214 error = "rec_len is too small for name_len";
217 error = "directory entry across blocks";
220 // error = "inode out of bounds";
222 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
223 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
224 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
225 (unsigned long) le32_to_cpu(p->inode),
226 rec_len, p->name_len);
229 p = (ext2_dirent *)(kaddr + offs);
230 CERROR("ext2_check_page"
231 "entry in directory #%lu spans the page boundary"
232 "offset=%lu, inode=%lu",
233 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
234 (unsigned long) le32_to_cpu(p->inode));
236 SetPageChecked(page);
240 static struct page * ext2_get_page(struct inode *dir, unsigned long n)
242 struct address_space *mapping = dir->i_mapping;
243 struct page *page = read_cache_page(mapping, n,
244 (filler_t*)mapping->a_ops->readpage, NULL);
248 if (!Page_Uptodate(page))
250 if (!PageChecked(page))
251 ext2_check_page(page);
259 return ERR_PTR(-EIO);
263 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
265 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
267 static inline int ext2_match (int len, const char * const name,
268 struct ext2_dir_entry_2 * de)
270 if (len != de->name_len)
274 return !memcmp(name, de->name, len);
278 * p is at least 6 bytes before the end of page
280 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
282 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
285 static inline unsigned
286 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
288 ext2_dirent *de = (ext2_dirent*)(base + offset);
289 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
290 while ((char*)p < (char*)de)
291 p = ext2_next_entry(p);
292 return (char *)p - base;
295 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
296 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
297 [EXT2_FT_REG_FILE] DT_REG,
298 [EXT2_FT_DIR] DT_DIR,
299 [EXT2_FT_CHRDEV] DT_CHR,
300 [EXT2_FT_BLKDEV] DT_BLK,
301 [EXT2_FT_FIFO] DT_FIFO,
302 [EXT2_FT_SOCK] DT_SOCK,
303 [EXT2_FT_SYMLINK] DT_LNK,
306 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
308 [EXT2_FT_REG_FILE] S_IFREG,
309 [EXT2_FT_DIR] S_IFDIR,
310 [EXT2_FT_CHRDEV] S_IFCHR,
311 [EXT2_FT_BLKDEV] S_IFBLK,
312 [EXT2_FT_FIFO] S_IFIFO,
313 [EXT2_FT_SOCK] S_IFSOCK,
314 [EXT2_FT_SYMLINK] S_IFLNK
318 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
319 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
320 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
321 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
322 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
323 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
324 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
325 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
328 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
330 mode_t mode = inode->i_mode;
331 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
335 new_ll_readdir (struct file * filp, void * dirent, filldir_t filldir)
337 loff_t pos = filp->f_pos;
338 struct inode *inode = filp->f_dentry->d_inode;
339 // XXX struct super_block *sb = inode->i_sb;
340 unsigned offset = pos & ~PAGE_CACHE_MASK;
341 unsigned long n = pos >> PAGE_CACHE_SHIFT;
342 unsigned long npages = dir_pages(inode);
343 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
344 unsigned char *types = NULL;
345 int need_revalidate = (filp->f_version != inode->i_version);
347 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
350 types = ext2_filetype_table;
352 for ( ; n < npages; n++, offset = 0) {
355 struct page *page = ext2_get_page(inode, n);
359 kaddr = page_address(page);
360 if (need_revalidate) {
361 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
364 de = (ext2_dirent *)(kaddr+offset);
365 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
366 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
369 unsigned char d_type = DT_UNKNOWN;
371 if (types && de->file_type < EXT2_FT_MAX)
372 d_type = types[de->file_type];
374 offset = (char *)de - kaddr;
375 over = filldir(dirent, de->name, de->name_len,
376 (n<<PAGE_CACHE_SHIFT) | offset,
377 le32_to_cpu(de->inode), d_type);
387 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
388 filp->f_version = inode->i_version;
396 * finds an entry in the specified directory with the wanted name. It
397 * returns the page in which the entry was found, and the entry itself
398 * (as a parameter - res_dir). Page is returned mapped and unlocked.
399 * Entry is guaranteed to be valid.
401 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
402 struct dentry *dentry, struct page ** res_page)
404 const char *name = dentry->d_name.name;
405 int namelen = dentry->d_name.len;
406 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
407 unsigned long start, n;
408 unsigned long npages = dir_pages(dir);
409 struct page *page = NULL;
415 // start = dir->u.ext2_i.i_dir_start_lookup;
422 page = ext2_get_page(dir, n);
424 kaddr = page_address(page);
425 de = (ext2_dirent *) kaddr;
426 kaddr += PAGE_CACHE_SIZE - reclen;
427 while ((char *) de <= kaddr) {
428 if (ext2_match (namelen, name, de))
430 de = ext2_next_entry(de);
436 } while (n != start);
441 // dir->u.ext2_i.i_dir_start_lookup = n;
445 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
447 struct page *page = ext2_get_page(dir, 0);
448 ext2_dirent *de = NULL;
451 de = ext2_next_entry((ext2_dirent *) page_address(page));
457 ino_t ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
460 struct ext2_dir_entry_2 * de;
463 de = ext2_find_entry (dir, dentry, &page);
465 res = le32_to_cpu(de->inode);
466 *type = ll_dt2fmt[de->file_type];
468 page_cache_release(page);
473 /* Releases the page */
474 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
475 struct page *page, struct inode *inode)
477 unsigned from = (char *) de - (char *) page_address(page);
478 unsigned to = from + le16_to_cpu(de->rec_len);
482 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
485 de->inode = cpu_to_le32(inode->i_ino);
486 ext2_set_de_type (de, inode);
487 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
488 err = ext2_commit_chunk(page, from, to);
496 int ll_add_link (struct dentry *dentry, struct inode *inode)
498 struct inode *dir = dentry->d_parent->d_inode;
499 const char *name = dentry->d_name.name;
500 int namelen = dentry->d_name.len;
501 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
502 unsigned short rec_len, name_len;
503 struct page *page = NULL;
505 unsigned long npages = dir_pages(dir);
511 /* We take care of directory expansion in the same loop */
512 for (n = 0; n <= npages; n++) {
513 page = ext2_get_page(dir, n);
517 kaddr = page_address(page);
518 de = (ext2_dirent *)kaddr;
519 kaddr += PAGE_CACHE_SIZE - reclen;
520 while ((char *)de <= kaddr) {
522 if (ext2_match (namelen, name, de))
524 name_len = EXT2_DIR_REC_LEN(de->name_len);
525 rec_len = le16_to_cpu(de->rec_len);
526 if ( n==npages && rec_len == 0) {
527 CERROR("Fatal dir behaviour\n");
530 if (!de->inode && rec_len >= reclen)
532 if (rec_len >= name_len + reclen)
534 de = (ext2_dirent *) ((char *) de + rec_len);
542 from = (char*)de - (char*)page_address(page);
545 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
549 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
550 de1->rec_len = cpu_to_le16(rec_len - name_len);
551 de->rec_len = cpu_to_le16(name_len);
554 de->name_len = namelen;
555 memcpy (de->name, name, namelen);
556 de->inode = cpu_to_le32(inode->i_ino);
557 ext2_set_de_type (de, inode);
558 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
559 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
560 err = ext2_commit_chunk(page, from, to);
562 // change_inode happens with the commit_chunk
563 /* XXX OFFSET_CACHE */
574 * ext2_delete_entry deletes a directory entry by merging it with the
575 * previous entry. Page is up-to-date. Releases the page.
577 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
579 struct address_space *mapping = page->mapping;
580 struct inode *inode = mapping->host;
581 char *kaddr = page_address(page);
582 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
583 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
584 ext2_dirent * pde = NULL;
585 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
588 while ((char*)de < (char*)dir) {
590 de = ext2_next_entry(de);
593 from = (char*)pde - (char*)page_address(page);
595 err = mapping->a_ops->prepare_write(NULL, page, from, to);
599 pde->rec_len = cpu_to_le16(to-from);
601 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
602 err = ext2_commit_chunk(page, from, to);
609 * Set the first fragment of directory.
611 int ext2_make_empty(struct inode *inode, struct inode *parent)
613 struct address_space *mapping = inode->i_mapping;
614 struct page *page = grab_cache_page(mapping, 0);
615 unsigned chunk_size = ext2_chunk_size(inode);
616 struct ext2_dir_entry_2 * de;
623 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
627 base = page_address(page);
629 de = (struct ext2_dir_entry_2 *) base;
631 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
632 memcpy (de->name, ".\0\0", 4);
633 de->inode = cpu_to_le32(inode->i_ino);
634 ext2_set_de_type (de, inode);
636 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
638 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
639 de->inode = cpu_to_le32(parent->i_ino);
640 memcpy (de->name, "..\0", 4);
641 ext2_set_de_type (de, inode);
643 err = ext2_commit_chunk(page, 0, chunk_size);
646 page_cache_release(page);
652 * routine to check that the specified directory is empty (for rmdir)
654 int ext2_empty_dir (struct inode * inode)
656 struct page *page = NULL;
657 unsigned long i, npages = dir_pages(inode);
659 for (i = 0; i < npages; i++) {
662 page = ext2_get_page(inode, i);
667 kaddr = page_address(page);
668 de = (ext2_dirent *)kaddr;
669 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
671 while ((char *)de <= kaddr) {
672 if (de->inode != 0) {
673 /* check for . and .. */
674 if (de->name[0] != '.')
676 if (de->name_len > 2)
678 if (de->name_len < 2) {
680 cpu_to_le32(inode->i_ino))
682 } else if (de->name[1] != '.')
685 de = ext2_next_entry(de);
696 struct file_operations ll_dir_operations = {
697 read: generic_read_dir,
698 readdir: new_ll_readdir