1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/dir.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * ext2 directory handling functions
18 * Big-endian to little-endian byte-swapping/bitmaps by
19 * David S. Miller (davem@caip.rutgers.edu), 1995
21 * All code that works with directory layout had been switched to pagecache
24 * Adapted for Lustre Light
25 * Copyright (C) 2002, Cluster File Systems, Inc.
30 #include <linux/ext2_fs.h>
31 #include <linux/pagemap.h>
33 #include <linux/version.h>
34 #include <linux/smp_lock.h>
35 #include <asm/uaccess.h>
36 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
37 #include <linux/locks.h> // for wait_on_buffer
39 #include <linux/buffer_head.h> // for wait_on_buffer
42 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <linux/obd_support.h>
45 #include <linux/obd_class.h>
46 #include <linux/lustre_lib.h>
47 #include <linux/lustre_idl.h>
48 #include <linux/lustre_mds.h>
49 #include <linux/lustre_lite.h>
50 #include <linux/lustre_dlm.h>
52 typedef struct ext2_dir_entry_2 ext2_dirent;
54 #define PageChecked(page) test_bit(PG_checked, &(page)->flags)
55 #define SetPageChecked(page) set_bit(PG_checked, &(page)->flags)
58 static int ll_dir_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
63 /* returns the page unlocked, but with a reference */
64 static int ll_dir_readpage(struct file *file, struct page *page)
66 struct inode *inode = page->mapping->host;
67 struct ll_sb_info *sbi = ll_i2sbi(inode);
71 struct ptlrpc_request *request;
72 struct lustre_handle lockh;
73 struct mds_body *body;
74 struct lookup_intent it = {IT_READDIR};
78 if ((inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_SHIFT <= page->index){
79 memset(kmap(page), 0, PAGE_CACHE_SIZE);
81 GOTO(readpage_out, rc);
84 rc = ll_lock(inode, NULL, &it, &lockh);
85 request = (struct ptlrpc_request *)it.it_data;
86 ptlrpc_req_finished(request);
88 CERROR("lock enqueue: err: %d\n", rc);
92 ldlm_lock_dump((void *)(unsigned long)lockh.addr);
94 if (PageUptodate(page)) {
95 CERROR("Explain this please?\n");
96 GOTO(readpage_out, rc);
99 offset = page->index << PAGE_SHIFT;
101 rc = mdc_readpage(&sbi->ll_mdc_conn, inode->i_ino,
102 S_IFDIR, offset, buf, &request);
105 body = lustre_msg_buf(request->rq_repmsg, 0);
109 inode->i_size = body->size;
111 ptlrpc_req_finished(request);
116 SetPageUptodate(page);
119 rc = ll_unlock(LCK_PR, &lockh);
121 CERROR("ll_unlock: err: %d\n", rc);
123 } /* ll_dir_readpage */
125 struct address_space_operations ll_dir_aops = {
126 readpage: ll_dir_readpage,
127 prepare_write: ll_dir_prepare_write
130 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
131 int waitfor_one_page(struct page *page)
134 struct buffer_head *bh, *head = page->buffers;
139 if (buffer_req(bh) && !buffer_uptodate(bh))
141 } while ((bh = bh->b_this_page) != head);
145 int waitfor_one_page(struct page *page)
147 wait_on_page_locked(page);
153 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
154 * more robust, but we have what we have
156 static inline unsigned ext2_chunk_size(struct inode *inode)
158 return inode->i_sb->s_blocksize;
161 static inline void ext2_put_page(struct page *page)
164 page_cache_release(page);
167 static inline unsigned long dir_pages(struct inode *inode)
169 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
172 extern void set_page_clean(struct page *page);
174 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
176 struct inode *dir = page->mapping->host;
177 loff_t new_size = (page->index << PAGE_CACHE_SHIFT) + to;
180 dir->i_version = ++event;
181 if (new_size > dir->i_size)
182 dir->i_size = new_size;
183 SetPageUptodate(page);
184 set_page_clean(page);
186 //page->mapping->a_ops->commit_write(NULL, page, from, to);
188 // err = waitfor_one_page(page);
192 static void ext2_check_page(struct page *page)
194 struct inode *dir = page->mapping->host;
195 unsigned chunk_size = ext2_chunk_size(dir);
196 char *kaddr = page_address(page);
197 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
198 unsigned offs, rec_len;
199 unsigned limit = PAGE_CACHE_SIZE;
203 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
204 limit = dir->i_size & ~PAGE_CACHE_MASK;
205 if (limit & (chunk_size - 1)) {
206 CERROR("limit %d dir size %lld index %ld\n",
207 limit, dir->i_size, page->index);
210 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
211 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
212 p->rec_len = cpu_to_le16(chunk_size);
219 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
220 p = (ext2_dirent *)(kaddr + offs);
221 rec_len = le16_to_cpu(p->rec_len);
223 if (rec_len < EXT2_DIR_REC_LEN(1))
227 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
229 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
231 // if (le32_to_cpu(p->inode) > max_inumber)
237 SetPageChecked(page);
240 /* Too bad, we had an error */
243 CERROR("ext2_check_page"
244 "size of directory #%lu is not a multiple of chunk size\n",
249 error = "rec_len is smaller than minimal";
252 error = "unaligned directory entry";
255 error = "rec_len is too small for name_len";
258 error = "directory entry across blocks";
261 // error = "inode out of bounds";
263 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
264 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
265 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
266 (unsigned long) le32_to_cpu(p->inode),
267 rec_len, p->name_len);
270 p = (ext2_dirent *)(kaddr + offs);
271 CERROR("ext2_check_page"
272 "entry in directory #%lu spans the page boundary"
273 "offset=%lu, inode=%lu",
274 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
275 (unsigned long) le32_to_cpu(p->inode));
277 SetPageChecked(page);
282 static struct page * ext2_get_page(struct inode *dir, unsigned long n)
284 struct address_space *mapping = dir->i_mapping;
285 struct page *page = read_cache_page(mapping, n,
286 (filler_t*)mapping->a_ops->readpage, NULL);
290 if (!PageUptodate(page))
292 if (!PageChecked(page))
293 ext2_check_page(page);
301 return ERR_PTR(-EIO);
305 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
307 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
309 static inline int ext2_match (int len, const char * const name,
310 struct ext2_dir_entry_2 * de)
312 if (len != de->name_len)
316 return !memcmp(name, de->name, len);
320 * p is at least 6 bytes before the end of page
322 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
324 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
327 static inline unsigned
328 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
330 ext2_dirent *de = (ext2_dirent*)(base + offset);
331 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
332 while ((char*)p < (char*)de)
333 p = ext2_next_entry(p);
334 return (char *)p - base;
337 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
338 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
339 [EXT2_FT_REG_FILE] DT_REG,
340 [EXT2_FT_DIR] DT_DIR,
341 [EXT2_FT_CHRDEV] DT_CHR,
342 [EXT2_FT_BLKDEV] DT_BLK,
343 [EXT2_FT_FIFO] DT_FIFO,
344 [EXT2_FT_SOCK] DT_SOCK,
345 [EXT2_FT_SYMLINK] DT_LNK,
348 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
350 [EXT2_FT_REG_FILE] S_IFREG,
351 [EXT2_FT_DIR] S_IFDIR,
352 [EXT2_FT_CHRDEV] S_IFCHR,
353 [EXT2_FT_BLKDEV] S_IFBLK,
354 [EXT2_FT_FIFO] S_IFIFO,
355 [EXT2_FT_SOCK] S_IFSOCK,
356 [EXT2_FT_SYMLINK] S_IFLNK
360 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
361 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
362 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
363 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
364 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
365 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
366 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
367 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
370 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
372 mode_t mode = inode->i_mode;
373 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
376 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
378 loff_t pos = filp->f_pos;
379 struct inode *inode = filp->f_dentry->d_inode;
380 // XXX struct super_block *sb = inode->i_sb;
381 unsigned offset = pos & ~PAGE_CACHE_MASK;
382 unsigned long n = pos >> PAGE_CACHE_SHIFT;
383 unsigned long npages = dir_pages(inode);
384 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
385 unsigned char *types = NULL;
386 int need_revalidate = (filp->f_version != inode->i_version);
389 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
392 types = ext2_filetype_table;
394 for ( ; n < npages; n++, offset = 0) {
397 struct page *page = ext2_get_page(inode, n);
399 /* size might have been updated by mdc_readpage */
400 npages = dir_pages(inode);
404 kaddr = page_address(page);
405 if (need_revalidate) {
406 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
409 de = (ext2_dirent *)(kaddr+offset);
410 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
411 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
414 unsigned char d_type = DT_UNKNOWN;
416 if (types && de->file_type < EXT2_FT_MAX)
417 d_type = types[de->file_type];
419 offset = (char *)de - kaddr;
420 over = filldir(dirent, de->name, de->name_len,
421 (n<<PAGE_CACHE_SHIFT) | offset,
422 le32_to_cpu(de->inode), d_type);
432 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
433 filp->f_version = inode->i_version;
441 * finds an entry in the specified directory with the wanted name. It
442 * returns the page in which the entry was found, and the entry itself
443 * (as a parameter - res_dir). Page is returned mapped and unlocked.
444 * Entry is guaranteed to be valid.
446 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
447 struct dentry *dentry, struct page ** res_page)
449 const char *name = dentry->d_name.name;
450 int namelen = dentry->d_name.len;
451 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
452 unsigned long start, n;
453 unsigned long npages = dir_pages(dir);
454 struct page *page = NULL;
460 // start = dir->u.ext2_i.i_dir_start_lookup;
467 page = ext2_get_page(dir, n);
469 kaddr = page_address(page);
470 de = (ext2_dirent *) kaddr;
471 kaddr += PAGE_CACHE_SIZE - reclen;
472 while ((char *) de <= kaddr) {
473 if (ext2_match (namelen, name, de))
475 de = ext2_next_entry(de);
481 } while (n != start);
486 // dir->u.ext2_i.i_dir_start_lookup = n;
490 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
492 struct page *page = ext2_get_page(dir, 0);
493 ext2_dirent *de = NULL;
496 de = ext2_next_entry((ext2_dirent *) page_address(page));
502 obd_id ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
505 struct ext2_dir_entry_2 * de;
508 de = ext2_find_entry (dir, dentry, &page);
510 res = le32_to_cpu(de->inode);
511 *type = ll_dt2fmt[de->file_type];
513 page_cache_release(page);
518 /* Releases the page */
519 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
520 struct page *page, struct inode *inode)
522 unsigned from = (char *) de - (char *) page_address(page);
523 unsigned to = from + le16_to_cpu(de->rec_len);
527 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
530 de->inode = cpu_to_le32(inode->i_ino);
531 ext2_set_de_type (de, inode);
532 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
533 err = ext2_commit_chunk(page, from, to);
541 int ll_add_link (struct dentry *dentry, struct inode *inode)
543 struct inode *dir = dentry->d_parent->d_inode;
544 const char *name = dentry->d_name.name;
545 int namelen = dentry->d_name.len;
546 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
547 unsigned short rec_len, name_len;
548 struct page *page = NULL;
550 unsigned long npages = dir_pages(dir);
556 /* We take care of directory expansion in the same loop */
557 for (n = 0; n <= npages; n++) {
558 page = ext2_get_page(dir, n);
562 kaddr = page_address(page);
563 de = (ext2_dirent *)kaddr;
564 kaddr += PAGE_CACHE_SIZE - reclen;
565 while ((char *)de <= kaddr) {
567 if (ext2_match (namelen, name, de))
569 name_len = EXT2_DIR_REC_LEN(de->name_len);
570 rec_len = le16_to_cpu(de->rec_len);
571 if ( n==npages && rec_len == 0) {
572 CERROR("Fatal dir behaviour\n");
575 if (!de->inode && rec_len >= reclen)
577 if (rec_len >= name_len + reclen)
579 de = (ext2_dirent *) ((char *) de + rec_len);
587 from = (char*)de - (char*)page_address(page);
590 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
594 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
595 de1->rec_len = cpu_to_le16(rec_len - name_len);
596 de->rec_len = cpu_to_le16(name_len);
599 de->name_len = namelen;
600 memcpy (de->name, name, namelen);
601 de->inode = cpu_to_le32(inode->i_ino);
602 ext2_set_de_type (de, inode);
603 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
604 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
605 err = ext2_commit_chunk(page, from, to);
607 // change_inode happens with the commit_chunk
608 /* XXX OFFSET_CACHE */
619 * ext2_delete_entry deletes a directory entry by merging it with the
620 * previous entry. Page is up-to-date. Releases the page.
622 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
624 struct address_space *mapping = page->mapping;
625 struct inode *inode = mapping->host;
626 char *kaddr = page_address(page);
627 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
628 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
629 ext2_dirent * pde = NULL;
630 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
633 while ((char*)de < (char*)dir) {
635 de = ext2_next_entry(de);
638 from = (char*)pde - (char*)page_address(page);
640 err = mapping->a_ops->prepare_write(NULL, page, from, to);
644 pde->rec_len = cpu_to_le16(to-from);
646 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
647 err = ext2_commit_chunk(page, from, to);
654 * Set the first fragment of directory.
656 int ext2_make_empty(struct inode *inode, struct inode *parent)
658 struct address_space *mapping = inode->i_mapping;
659 struct page *page = grab_cache_page(mapping, 0);
660 unsigned chunk_size = ext2_chunk_size(inode);
661 struct ext2_dir_entry_2 * de;
672 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
676 de = (struct ext2_dir_entry_2 *) base;
678 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
679 memcpy (de->name, ".\0\0", 4);
680 de->inode = cpu_to_le32(inode->i_ino);
681 ext2_set_de_type (de, inode);
683 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
685 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
686 de->inode = cpu_to_le32(parent->i_ino);
687 memcpy (de->name, "..\0", 4);
688 ext2_set_de_type (de, inode);
690 err = ext2_commit_chunk(page, 0, chunk_size);
694 page_cache_release(page);
700 * routine to check that the specified directory is empty (for rmdir)
702 int ext2_empty_dir (struct inode * inode)
704 struct page *page = NULL;
705 unsigned long i, npages = dir_pages(inode);
707 for (i = 0; i < npages; i++) {
710 page = ext2_get_page(inode, i);
715 kaddr = page_address(page);
716 de = (ext2_dirent *)kaddr;
717 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
719 while ((char *)de <= kaddr) {
720 if (de->inode != 0) {
721 /* check for . and .. */
722 if (de->name[0] != '.')
724 if (de->name_len > 2)
726 if (de->name_len < 2) {
728 cpu_to_le32(inode->i_ino))
730 } else if (de->name[1] != '.')
733 de = ext2_next_entry(de);
744 struct file_operations ll_dir_operations = {
745 read: generic_read_dir,