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)) {
76 CERROR("Explain this please?\n");
81 offset = page->index << PAGE_SHIFT;
83 rc = mdc_readpage(&sbi->ll_mds_client, inode->i_ino, S_IFDIR, offset,
86 ptlrpc_free_req(request);
91 SetPageUptodate(page);
95 } /* ll_dir_readpage */
97 struct address_space_operations ll_dir_aops = {
98 readpage: ll_dir_readpage,
99 prepare_write: ll_dir_prepare_write
102 int waitfor_one_page(struct page *page)
105 struct buffer_head *bh, *head = page->buffers;
110 if (buffer_req(bh) && !buffer_uptodate(bh))
112 } while ((bh = bh->b_this_page) != head);
117 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
118 * more robust, but we have what we have
120 static inline unsigned ext2_chunk_size(struct inode *inode)
122 //return inode->i_sb->s_blocksize;
126 static inline void ext2_put_page(struct page *page)
129 page_cache_release(page);
132 static inline unsigned long dir_pages(struct inode *inode)
134 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
137 extern void set_page_clean(struct page *page);
139 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
141 struct inode *dir = page->mapping->host;
144 dir->i_version = ++event;
145 dir->i_size = (page->index << PAGE_CACHE_SHIFT) + to;
146 SetPageUptodate(page);
147 set_page_clean(page);
149 //page->mapping->a_ops->commit_write(NULL, page, from, to);
151 // err = waitfor_one_page(page);
155 static void ext2_check_page(struct page *page)
157 struct inode *dir = page->mapping->host;
158 unsigned chunk_size = ext2_chunk_size(dir);
159 char *kaddr = page_address(page);
160 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
161 unsigned offs, rec_len;
162 unsigned limit = PAGE_CACHE_SIZE;
166 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
167 limit = dir->i_size & ~PAGE_CACHE_MASK;
168 if (limit & (chunk_size - 1)) {
169 CERROR("limit %d dir size %lld index %ld\n",
170 limit, dir->i_size, page->index);
173 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
174 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
175 p->rec_len = cpu_to_le16(chunk_size);
180 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
181 p = (ext2_dirent *)(kaddr + offs);
182 rec_len = le16_to_cpu(p->rec_len);
184 if (rec_len < EXT2_DIR_REC_LEN(1))
188 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
190 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
192 // if (le32_to_cpu(p->inode) > max_inumber)
198 SetPageChecked(page);
201 /* Too bad, we had an error */
204 CERROR("ext2_check_page"
205 "size of directory #%lu is not a multiple of chunk size\n",
210 error = "rec_len is smaller than minimal";
213 error = "unaligned directory entry";
216 error = "rec_len is too small for name_len";
219 error = "directory entry across blocks";
222 // error = "inode out of bounds";
224 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
225 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
226 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
227 (unsigned long) le32_to_cpu(p->inode),
228 rec_len, p->name_len);
231 p = (ext2_dirent *)(kaddr + offs);
232 CERROR("ext2_check_page"
233 "entry in directory #%lu spans the page boundary"
234 "offset=%lu, inode=%lu",
235 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
236 (unsigned long) le32_to_cpu(p->inode));
238 SetPageChecked(page);
243 static struct page * ext2_get_page(struct inode *dir, unsigned long n)
245 struct address_space *mapping = dir->i_mapping;
246 struct page *page = read_cache_page(mapping, n,
247 (filler_t*)mapping->a_ops->readpage, NULL);
251 if (!Page_Uptodate(page))
253 if (!PageChecked(page))
254 ext2_check_page(page);
262 return ERR_PTR(-EIO);
266 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
268 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
270 static inline int ext2_match (int len, const char * const name,
271 struct ext2_dir_entry_2 * de)
273 if (len != de->name_len)
277 return !memcmp(name, de->name, len);
281 * p is at least 6 bytes before the end of page
283 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
285 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
288 static inline unsigned
289 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
291 ext2_dirent *de = (ext2_dirent*)(base + offset);
292 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
293 while ((char*)p < (char*)de)
294 p = ext2_next_entry(p);
295 return (char *)p - base;
298 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
299 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
300 [EXT2_FT_REG_FILE] DT_REG,
301 [EXT2_FT_DIR] DT_DIR,
302 [EXT2_FT_CHRDEV] DT_CHR,
303 [EXT2_FT_BLKDEV] DT_BLK,
304 [EXT2_FT_FIFO] DT_FIFO,
305 [EXT2_FT_SOCK] DT_SOCK,
306 [EXT2_FT_SYMLINK] DT_LNK,
309 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
311 [EXT2_FT_REG_FILE] S_IFREG,
312 [EXT2_FT_DIR] S_IFDIR,
313 [EXT2_FT_CHRDEV] S_IFCHR,
314 [EXT2_FT_BLKDEV] S_IFBLK,
315 [EXT2_FT_FIFO] S_IFIFO,
316 [EXT2_FT_SOCK] S_IFSOCK,
317 [EXT2_FT_SYMLINK] S_IFLNK
321 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
322 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
323 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
324 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
325 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
326 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
327 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
328 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
331 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
333 mode_t mode = inode->i_mode;
334 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
338 new_ll_readdir (struct file * filp, void * dirent, filldir_t filldir)
340 loff_t pos = filp->f_pos;
341 struct inode *inode = filp->f_dentry->d_inode;
342 // XXX struct super_block *sb = inode->i_sb;
343 unsigned offset = pos & ~PAGE_CACHE_MASK;
344 unsigned long n = pos >> PAGE_CACHE_SHIFT;
345 unsigned long npages = dir_pages(inode);
346 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
347 unsigned char *types = NULL;
348 int need_revalidate = (filp->f_version != inode->i_version);
350 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
353 types = ext2_filetype_table;
355 for ( ; n < npages; n++, offset = 0) {
358 struct page *page = ext2_get_page(inode, n);
362 kaddr = page_address(page);
363 if (need_revalidate) {
364 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
367 de = (ext2_dirent *)(kaddr+offset);
368 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
369 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
372 unsigned char d_type = DT_UNKNOWN;
374 if (types && de->file_type < EXT2_FT_MAX)
375 d_type = types[de->file_type];
377 offset = (char *)de - kaddr;
378 over = filldir(dirent, de->name, de->name_len,
379 (n<<PAGE_CACHE_SHIFT) | offset,
380 le32_to_cpu(de->inode), d_type);
390 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
391 filp->f_version = inode->i_version;
399 * finds an entry in the specified directory with the wanted name. It
400 * returns the page in which the entry was found, and the entry itself
401 * (as a parameter - res_dir). Page is returned mapped and unlocked.
402 * Entry is guaranteed to be valid.
404 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
405 struct dentry *dentry, struct page ** res_page)
407 const char *name = dentry->d_name.name;
408 int namelen = dentry->d_name.len;
409 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
410 unsigned long start, n;
411 unsigned long npages = dir_pages(dir);
412 struct page *page = NULL;
418 // start = dir->u.ext2_i.i_dir_start_lookup;
425 page = ext2_get_page(dir, n);
427 kaddr = page_address(page);
428 de = (ext2_dirent *) kaddr;
429 kaddr += PAGE_CACHE_SIZE - reclen;
430 while ((char *) de <= kaddr) {
431 if (ext2_match (namelen, name, de))
433 de = ext2_next_entry(de);
439 } while (n != start);
444 // dir->u.ext2_i.i_dir_start_lookup = n;
448 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
450 struct page *page = ext2_get_page(dir, 0);
451 ext2_dirent *de = NULL;
454 de = ext2_next_entry((ext2_dirent *) page_address(page));
460 ino_t ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
463 struct ext2_dir_entry_2 * de;
466 de = ext2_find_entry (dir, dentry, &page);
468 res = le32_to_cpu(de->inode);
469 *type = ll_dt2fmt[de->file_type];
471 page_cache_release(page);
476 /* Releases the page */
477 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
478 struct page *page, struct inode *inode)
480 unsigned from = (char *) de - (char *) page_address(page);
481 unsigned to = from + le16_to_cpu(de->rec_len);
485 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
488 de->inode = cpu_to_le32(inode->i_ino);
489 ext2_set_de_type (de, inode);
490 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
491 err = ext2_commit_chunk(page, from, to);
499 int ll_add_link (struct dentry *dentry, struct inode *inode)
501 struct inode *dir = dentry->d_parent->d_inode;
502 const char *name = dentry->d_name.name;
503 int namelen = dentry->d_name.len;
504 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
505 unsigned short rec_len, name_len;
506 struct page *page = NULL;
508 unsigned long npages = dir_pages(dir);
514 /* We take care of directory expansion in the same loop */
515 for (n = 0; n <= npages; n++) {
516 page = ext2_get_page(dir, n);
520 kaddr = page_address(page);
521 de = (ext2_dirent *)kaddr;
522 kaddr += PAGE_CACHE_SIZE - reclen;
523 while ((char *)de <= kaddr) {
525 if (ext2_match (namelen, name, de))
527 name_len = EXT2_DIR_REC_LEN(de->name_len);
528 rec_len = le16_to_cpu(de->rec_len);
529 if ( n==npages && rec_len == 0) {
530 CERROR("Fatal dir behaviour\n");
533 if (!de->inode && rec_len >= reclen)
535 if (rec_len >= name_len + reclen)
537 de = (ext2_dirent *) ((char *) de + rec_len);
545 from = (char*)de - (char*)page_address(page);
548 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
552 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
553 de1->rec_len = cpu_to_le16(rec_len - name_len);
554 de->rec_len = cpu_to_le16(name_len);
557 de->name_len = namelen;
558 memcpy (de->name, name, namelen);
559 de->inode = cpu_to_le32(inode->i_ino);
560 ext2_set_de_type (de, inode);
561 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
562 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
563 err = ext2_commit_chunk(page, from, to);
565 // change_inode happens with the commit_chunk
566 /* XXX OFFSET_CACHE */
577 * ext2_delete_entry deletes a directory entry by merging it with the
578 * previous entry. Page is up-to-date. Releases the page.
580 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
582 struct address_space *mapping = page->mapping;
583 struct inode *inode = mapping->host;
584 char *kaddr = page_address(page);
585 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
586 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
587 ext2_dirent * pde = NULL;
588 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
591 while ((char*)de < (char*)dir) {
593 de = ext2_next_entry(de);
596 from = (char*)pde - (char*)page_address(page);
598 err = mapping->a_ops->prepare_write(NULL, page, from, to);
602 pde->rec_len = cpu_to_le16(to-from);
604 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
605 err = ext2_commit_chunk(page, from, to);
612 * Set the first fragment of directory.
614 int ext2_make_empty(struct inode *inode, struct inode *parent)
616 struct address_space *mapping = inode->i_mapping;
617 struct page *page = grab_cache_page(mapping, 0);
618 unsigned chunk_size = ext2_chunk_size(inode);
619 struct ext2_dir_entry_2 * de;
626 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
630 base = page_address(page);
632 de = (struct ext2_dir_entry_2 *) base;
634 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
635 memcpy (de->name, ".\0\0", 4);
636 de->inode = cpu_to_le32(inode->i_ino);
637 ext2_set_de_type (de, inode);
639 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
641 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
642 de->inode = cpu_to_le32(parent->i_ino);
643 memcpy (de->name, "..\0", 4);
644 ext2_set_de_type (de, inode);
646 err = ext2_commit_chunk(page, 0, chunk_size);
649 page_cache_release(page);
655 * routine to check that the specified directory is empty (for rmdir)
657 int ext2_empty_dir (struct inode * inode)
659 struct page *page = NULL;
660 unsigned long i, npages = dir_pages(inode);
662 for (i = 0; i < npages; i++) {
665 page = ext2_get_page(inode, i);
670 kaddr = page_address(page);
671 de = (ext2_dirent *)kaddr;
672 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
674 while ((char *)de <= kaddr) {
675 if (de->inode != 0) {
676 /* check for . and .. */
677 if (de->name[0] != '.')
679 if (de->name_len > 2)
681 if (de->name_len < 2) {
683 cpu_to_le32(inode->i_ino))
685 } else if (de->name[1] != '.')
688 de = ext2_next_entry(de);
699 struct file_operations ll_dir_operations = {
700 read: generic_read_dir,
701 readdir: new_ll_readdir