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/locks.h>
34 #include <asm/uaccess.h>
36 #define DEBUG_SUBSYSTEM S_LLITE
38 #include <linux/obd_support.h>
39 #include <linux/obd_class.h>
40 #include <linux/lustre_lib.h>
41 #include <linux/lustre_idl.h>
42 #include <linux/lustre_mds.h>
43 #include <linux/lustre_lite.h>
44 #include <linux/lustre_dlm.h>
46 typedef struct ext2_dir_entry_2 ext2_dirent;
48 #define PageChecked(page) test_bit(PG_checked, &(page)->flags)
49 #define SetPageChecked(page) set_bit(PG_checked, &(page)->flags)
52 static int ll_dir_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
57 /* returns the page unlocked, but with a reference */
58 static int ll_dir_readpage(struct file *file, struct page *page)
60 struct inode *inode = page->mapping->host;
61 struct ll_sb_info *sbi = ll_i2sbi(inode);
65 struct ptlrpc_request *request;
66 struct lustre_handle lockh;
67 struct mds_body *body;
68 struct lookup_intent it = {IT_READDIR};
72 if ((inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_SHIFT <= page->index){
73 memset(kmap(page), 0, PAGE_CACHE_SIZE);
75 GOTO(readpage_out, rc);
78 rc = ll_lock(inode, NULL, &it, &lockh);
79 request = (struct ptlrpc_request *)it.it_data;
80 ptlrpc_free_req(request);
82 CERROR("lock enqueue: err: %d\n", rc);
86 ldlm_lock_dump((void *)(unsigned long)lockh.addr);
88 if (Page_Uptodate(page)) {
89 CERROR("Explain this please?\n");
90 GOTO(readpage_out, rc);
93 offset = page->index << PAGE_SHIFT;
95 rc = mdc_readpage(&sbi->ll_mdc_conn, inode->i_ino,
96 S_IFDIR, offset, buf, &request);
98 body = lustre_msg_buf(request->rq_repmsg, 0);
102 inode->i_size = body->size;
103 ptlrpc_free_req(request);
108 SetPageUptodate(page);
111 rc = ll_unlock(LCK_PR, &lockh);
113 CERROR("ll_unlock: err: %d\n", rc);
115 } /* ll_dir_readpage */
117 struct address_space_operations ll_dir_aops = {
118 readpage: ll_dir_readpage,
119 prepare_write: ll_dir_prepare_write
122 int waitfor_one_page(struct page *page)
125 struct buffer_head *bh, *head = page->buffers;
130 if (buffer_req(bh) && !buffer_uptodate(bh))
132 } while ((bh = bh->b_this_page) != head);
137 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
138 * more robust, but we have what we have
140 static inline unsigned ext2_chunk_size(struct inode *inode)
142 return inode->i_sb->s_blocksize;
145 static inline void ext2_put_page(struct page *page)
148 page_cache_release(page);
151 static inline unsigned long dir_pages(struct inode *inode)
153 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
156 extern void set_page_clean(struct page *page);
158 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
160 struct inode *dir = page->mapping->host;
161 loff_t new_size = (page->index << PAGE_CACHE_SHIFT) + to;
164 dir->i_version = ++event;
165 if (new_size > dir->i_size)
166 dir->i_size = new_size;
167 SetPageUptodate(page);
168 set_page_clean(page);
170 //page->mapping->a_ops->commit_write(NULL, page, from, to);
172 // err = waitfor_one_page(page);
176 static void ext2_check_page(struct page *page)
178 struct inode *dir = page->mapping->host;
179 unsigned chunk_size = ext2_chunk_size(dir);
180 char *kaddr = page_address(page);
181 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
182 unsigned offs, rec_len;
183 unsigned limit = PAGE_CACHE_SIZE;
187 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
188 limit = dir->i_size & ~PAGE_CACHE_MASK;
189 if (limit & (chunk_size - 1)) {
190 CERROR("limit %d dir size %lld index %ld\n",
191 limit, dir->i_size, page->index);
194 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
195 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
196 p->rec_len = cpu_to_le16(chunk_size);
203 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
204 p = (ext2_dirent *)(kaddr + offs);
205 rec_len = le16_to_cpu(p->rec_len);
207 if (rec_len < EXT2_DIR_REC_LEN(1))
211 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
213 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
215 // if (le32_to_cpu(p->inode) > max_inumber)
221 SetPageChecked(page);
224 /* Too bad, we had an error */
227 CERROR("ext2_check_page"
228 "size of directory #%lu is not a multiple of chunk size\n",
233 error = "rec_len is smaller than minimal";
236 error = "unaligned directory entry";
239 error = "rec_len is too small for name_len";
242 error = "directory entry across blocks";
245 // error = "inode out of bounds";
247 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
248 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
249 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
250 (unsigned long) le32_to_cpu(p->inode),
251 rec_len, p->name_len);
254 p = (ext2_dirent *)(kaddr + offs);
255 CERROR("ext2_check_page"
256 "entry in directory #%lu spans the page boundary"
257 "offset=%lu, inode=%lu",
258 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
259 (unsigned long) le32_to_cpu(p->inode));
261 SetPageChecked(page);
266 static struct page * ext2_get_page(struct inode *dir, unsigned long n)
268 struct address_space *mapping = dir->i_mapping;
269 struct page *page = read_cache_page(mapping, n,
270 (filler_t*)mapping->a_ops->readpage, NULL);
274 if (!Page_Uptodate(page))
276 if (!PageChecked(page))
277 ext2_check_page(page);
285 return ERR_PTR(-EIO);
289 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
291 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
293 static inline int ext2_match (int len, const char * const name,
294 struct ext2_dir_entry_2 * de)
296 if (len != de->name_len)
300 return !memcmp(name, de->name, len);
304 * p is at least 6 bytes before the end of page
306 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
308 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
311 static inline unsigned
312 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
314 ext2_dirent *de = (ext2_dirent*)(base + offset);
315 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
316 while ((char*)p < (char*)de)
317 p = ext2_next_entry(p);
318 return (char *)p - base;
321 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
322 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
323 [EXT2_FT_REG_FILE] DT_REG,
324 [EXT2_FT_DIR] DT_DIR,
325 [EXT2_FT_CHRDEV] DT_CHR,
326 [EXT2_FT_BLKDEV] DT_BLK,
327 [EXT2_FT_FIFO] DT_FIFO,
328 [EXT2_FT_SOCK] DT_SOCK,
329 [EXT2_FT_SYMLINK] DT_LNK,
332 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
334 [EXT2_FT_REG_FILE] S_IFREG,
335 [EXT2_FT_DIR] S_IFDIR,
336 [EXT2_FT_CHRDEV] S_IFCHR,
337 [EXT2_FT_BLKDEV] S_IFBLK,
338 [EXT2_FT_FIFO] S_IFIFO,
339 [EXT2_FT_SOCK] S_IFSOCK,
340 [EXT2_FT_SYMLINK] S_IFLNK
344 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
345 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
346 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
347 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
348 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
349 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
350 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
351 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
354 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
356 mode_t mode = inode->i_mode;
357 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
360 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
362 loff_t pos = filp->f_pos;
363 struct inode *inode = filp->f_dentry->d_inode;
364 // XXX struct super_block *sb = inode->i_sb;
365 unsigned offset = pos & ~PAGE_CACHE_MASK;
366 unsigned long n = pos >> PAGE_CACHE_SHIFT;
367 unsigned long npages = dir_pages(inode);
368 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
369 unsigned char *types = NULL;
370 int need_revalidate = (filp->f_version != inode->i_version);
373 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
376 types = ext2_filetype_table;
378 for ( ; n < npages; n++, offset = 0) {
381 struct page *page = ext2_get_page(inode, n);
383 /* size might have been updated by mdc_readpage */
384 npages = dir_pages(inode);
388 kaddr = page_address(page);
389 if (need_revalidate) {
390 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
393 de = (ext2_dirent *)(kaddr+offset);
394 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
395 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
398 unsigned char d_type = DT_UNKNOWN;
400 if (types && de->file_type < EXT2_FT_MAX)
401 d_type = types[de->file_type];
403 offset = (char *)de - kaddr;
404 over = filldir(dirent, de->name, de->name_len,
405 (n<<PAGE_CACHE_SHIFT) | offset,
406 le32_to_cpu(de->inode), d_type);
416 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
417 filp->f_version = inode->i_version;
425 * finds an entry in the specified directory with the wanted name. It
426 * returns the page in which the entry was found, and the entry itself
427 * (as a parameter - res_dir). Page is returned mapped and unlocked.
428 * Entry is guaranteed to be valid.
430 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
431 struct dentry *dentry, struct page ** res_page)
433 const char *name = dentry->d_name.name;
434 int namelen = dentry->d_name.len;
435 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
436 unsigned long start, n;
437 unsigned long npages = dir_pages(dir);
438 struct page *page = NULL;
444 // start = dir->u.ext2_i.i_dir_start_lookup;
451 page = ext2_get_page(dir, n);
453 kaddr = page_address(page);
454 de = (ext2_dirent *) kaddr;
455 kaddr += PAGE_CACHE_SIZE - reclen;
456 while ((char *) de <= kaddr) {
457 if (ext2_match (namelen, name, de))
459 de = ext2_next_entry(de);
465 } while (n != start);
470 // dir->u.ext2_i.i_dir_start_lookup = n;
474 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
476 struct page *page = ext2_get_page(dir, 0);
477 ext2_dirent *de = NULL;
480 de = ext2_next_entry((ext2_dirent *) page_address(page));
486 obd_id ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
489 struct ext2_dir_entry_2 * de;
492 de = ext2_find_entry (dir, dentry, &page);
494 res = le32_to_cpu(de->inode);
495 *type = ll_dt2fmt[de->file_type];
497 page_cache_release(page);
502 /* Releases the page */
503 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
504 struct page *page, struct inode *inode)
506 unsigned from = (char *) de - (char *) page_address(page);
507 unsigned to = from + le16_to_cpu(de->rec_len);
511 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
514 de->inode = cpu_to_le32(inode->i_ino);
515 ext2_set_de_type (de, inode);
516 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
517 err = ext2_commit_chunk(page, from, to);
525 int ll_add_link (struct dentry *dentry, struct inode *inode)
527 struct inode *dir = dentry->d_parent->d_inode;
528 const char *name = dentry->d_name.name;
529 int namelen = dentry->d_name.len;
530 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
531 unsigned short rec_len, name_len;
532 struct page *page = NULL;
534 unsigned long npages = dir_pages(dir);
540 /* We take care of directory expansion in the same loop */
541 for (n = 0; n <= npages; n++) {
542 page = ext2_get_page(dir, n);
546 kaddr = page_address(page);
547 de = (ext2_dirent *)kaddr;
548 kaddr += PAGE_CACHE_SIZE - reclen;
549 while ((char *)de <= kaddr) {
551 if (ext2_match (namelen, name, de))
553 name_len = EXT2_DIR_REC_LEN(de->name_len);
554 rec_len = le16_to_cpu(de->rec_len);
555 if ( n==npages && rec_len == 0) {
556 CERROR("Fatal dir behaviour\n");
559 if (!de->inode && rec_len >= reclen)
561 if (rec_len >= name_len + reclen)
563 de = (ext2_dirent *) ((char *) de + rec_len);
571 from = (char*)de - (char*)page_address(page);
574 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
578 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
579 de1->rec_len = cpu_to_le16(rec_len - name_len);
580 de->rec_len = cpu_to_le16(name_len);
583 de->name_len = namelen;
584 memcpy (de->name, name, namelen);
585 de->inode = cpu_to_le32(inode->i_ino);
586 ext2_set_de_type (de, inode);
587 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
588 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
589 err = ext2_commit_chunk(page, from, to);
591 // change_inode happens with the commit_chunk
592 /* XXX OFFSET_CACHE */
603 * ext2_delete_entry deletes a directory entry by merging it with the
604 * previous entry. Page is up-to-date. Releases the page.
606 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
608 struct address_space *mapping = page->mapping;
609 struct inode *inode = mapping->host;
610 char *kaddr = page_address(page);
611 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
612 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
613 ext2_dirent * pde = NULL;
614 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
617 while ((char*)de < (char*)dir) {
619 de = ext2_next_entry(de);
622 from = (char*)pde - (char*)page_address(page);
624 err = mapping->a_ops->prepare_write(NULL, page, from, to);
628 pde->rec_len = cpu_to_le16(to-from);
630 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
631 err = ext2_commit_chunk(page, from, to);
638 * Set the first fragment of directory.
640 int ext2_make_empty(struct inode *inode, struct inode *parent)
642 struct address_space *mapping = inode->i_mapping;
643 struct page *page = grab_cache_page(mapping, 0);
644 unsigned chunk_size = ext2_chunk_size(inode);
645 struct ext2_dir_entry_2 * de;
656 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
660 de = (struct ext2_dir_entry_2 *) base;
662 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
663 memcpy (de->name, ".\0\0", 4);
664 de->inode = cpu_to_le32(inode->i_ino);
665 ext2_set_de_type (de, inode);
667 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
669 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
670 de->inode = cpu_to_le32(parent->i_ino);
671 memcpy (de->name, "..\0", 4);
672 ext2_set_de_type (de, inode);
674 err = ext2_commit_chunk(page, 0, chunk_size);
678 page_cache_release(page);
684 * routine to check that the specified directory is empty (for rmdir)
686 int ext2_empty_dir (struct inode * inode)
688 struct page *page = NULL;
689 unsigned long i, npages = dir_pages(inode);
691 for (i = 0; i < npages; i++) {
694 page = ext2_get_page(inode, i);
699 kaddr = page_address(page);
700 de = (ext2_dirent *)kaddr;
701 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
703 while ((char *)de <= kaddr) {
704 if (de->inode != 0) {
705 /* check for . and .. */
706 if (de->name[0] != '.')
708 if (de->name_len > 2)
710 if (de->name_len < 2) {
712 cpu_to_le32(inode->i_ino))
714 } else if (de->name[1] != '.')
717 de = ext2_next_entry(de);
728 struct file_operations ll_dir_operations = {
729 read: generic_read_dir,