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/smp_lock.h>
34 #include <asm/uaccess.h>
35 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
36 #include <linux/locks.h> // for wait_on_buffer
38 #include <linux/buffer_head.h> // for wait_on_buffer
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/obd_support.h>
44 #include <linux/obd_class.h>
45 #include <linux/lustre_lib.h>
46 #include <linux/lustre_idl.h>
47 #include <linux/lustre_mds.h>
48 #include <linux/lustre_lite.h>
49 #include <linux/lustre_dlm.h>
51 typedef struct ext2_dir_entry_2 ext2_dirent;
53 #define PageChecked(page) test_bit(PG_checked, &(page)->flags)
54 #define SetPageChecked(page) set_bit(PG_checked, &(page)->flags)
57 static int ll_dir_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
62 /* returns the page unlocked, but with a reference */
63 static int ll_dir_readpage(struct file *file, struct page *page)
65 struct inode *inode = page->mapping->host;
66 struct ll_sb_info *sbi = ll_i2sbi(inode);
70 struct ptlrpc_request *request;
71 struct lustre_handle lockh;
72 struct mds_body *body;
73 struct lookup_intent it = {IT_READDIR};
77 if ((inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_SHIFT <= page->index){
78 memset(kmap(page), 0, PAGE_CACHE_SIZE);
80 GOTO(readpage_out, rc);
83 rc = ll_lock(inode, NULL, &it, &lockh);
84 request = (struct ptlrpc_request *)it.it_data;
85 ptlrpc_req_finished(request);
87 CERROR("lock enqueue: err: %d\n", rc);
91 ldlm_lock_dump((void *)(unsigned long)lockh.addr);
93 if (Page_Uptodate(page)) {
94 CERROR("Explain this please?\n");
95 GOTO(readpage_out, rc);
98 offset = page->index << PAGE_SHIFT;
100 rc = mdc_readpage(&sbi->ll_mdc_conn, inode->i_ino,
101 S_IFDIR, offset, buf, &request);
104 body = lustre_msg_buf(request->rq_repmsg, 0);
108 inode->i_size = body->size;
110 ptlrpc_req_finished(request);
115 SetPageUptodate(page);
118 rc = ll_unlock(LCK_PR, &lockh);
120 CERROR("ll_unlock: err: %d\n", rc);
122 } /* ll_dir_readpage */
124 struct address_space_operations ll_dir_aops = {
125 readpage: ll_dir_readpage,
126 prepare_write: ll_dir_prepare_write
129 int waitfor_one_page(struct page *page)
132 struct buffer_head *bh, *head = page->buffers;
137 if (buffer_req(bh) && !buffer_uptodate(bh))
139 } while ((bh = bh->b_this_page) != head);
144 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
145 * more robust, but we have what we have
147 static inline unsigned ext2_chunk_size(struct inode *inode)
149 return inode->i_sb->s_blocksize;
152 static inline void ext2_put_page(struct page *page)
155 page_cache_release(page);
158 static inline unsigned long dir_pages(struct inode *inode)
160 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
163 extern void set_page_clean(struct page *page);
165 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
167 struct inode *dir = page->mapping->host;
168 loff_t new_size = (page->index << PAGE_CACHE_SHIFT) + to;
171 dir->i_version = ++event;
172 if (new_size > dir->i_size)
173 dir->i_size = new_size;
174 SetPageUptodate(page);
175 set_page_clean(page);
177 //page->mapping->a_ops->commit_write(NULL, page, from, to);
179 // err = waitfor_one_page(page);
183 static void ext2_check_page(struct page *page)
185 struct inode *dir = page->mapping->host;
186 unsigned chunk_size = ext2_chunk_size(dir);
187 char *kaddr = page_address(page);
188 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
189 unsigned offs, rec_len;
190 unsigned limit = PAGE_CACHE_SIZE;
194 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
195 limit = dir->i_size & ~PAGE_CACHE_MASK;
196 if (limit & (chunk_size - 1)) {
197 CERROR("limit %d dir size %lld index %ld\n",
198 limit, dir->i_size, page->index);
201 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
202 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
203 p->rec_len = cpu_to_le16(chunk_size);
210 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
211 p = (ext2_dirent *)(kaddr + offs);
212 rec_len = le16_to_cpu(p->rec_len);
214 if (rec_len < EXT2_DIR_REC_LEN(1))
218 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
220 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
222 // if (le32_to_cpu(p->inode) > max_inumber)
228 SetPageChecked(page);
231 /* Too bad, we had an error */
234 CERROR("ext2_check_page"
235 "size of directory #%lu is not a multiple of chunk size\n",
240 error = "rec_len is smaller than minimal";
243 error = "unaligned directory entry";
246 error = "rec_len is too small for name_len";
249 error = "directory entry across blocks";
252 // error = "inode out of bounds";
254 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
255 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
256 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
257 (unsigned long) le32_to_cpu(p->inode),
258 rec_len, p->name_len);
261 p = (ext2_dirent *)(kaddr + offs);
262 CERROR("ext2_check_page"
263 "entry in directory #%lu spans the page boundary"
264 "offset=%lu, inode=%lu",
265 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
266 (unsigned long) le32_to_cpu(p->inode));
268 SetPageChecked(page);
273 static struct page * ext2_get_page(struct inode *dir, unsigned long n)
275 struct address_space *mapping = dir->i_mapping;
276 struct page *page = read_cache_page(mapping, n,
277 (filler_t*)mapping->a_ops->readpage, NULL);
281 if (!Page_Uptodate(page))
283 if (!PageChecked(page))
284 ext2_check_page(page);
292 return ERR_PTR(-EIO);
296 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
298 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
300 static inline int ext2_match (int len, const char * const name,
301 struct ext2_dir_entry_2 * de)
303 if (len != de->name_len)
307 return !memcmp(name, de->name, len);
311 * p is at least 6 bytes before the end of page
313 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
315 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
318 static inline unsigned
319 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
321 ext2_dirent *de = (ext2_dirent*)(base + offset);
322 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
323 while ((char*)p < (char*)de)
324 p = ext2_next_entry(p);
325 return (char *)p - base;
328 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
329 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
330 [EXT2_FT_REG_FILE] DT_REG,
331 [EXT2_FT_DIR] DT_DIR,
332 [EXT2_FT_CHRDEV] DT_CHR,
333 [EXT2_FT_BLKDEV] DT_BLK,
334 [EXT2_FT_FIFO] DT_FIFO,
335 [EXT2_FT_SOCK] DT_SOCK,
336 [EXT2_FT_SYMLINK] DT_LNK,
339 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
341 [EXT2_FT_REG_FILE] S_IFREG,
342 [EXT2_FT_DIR] S_IFDIR,
343 [EXT2_FT_CHRDEV] S_IFCHR,
344 [EXT2_FT_BLKDEV] S_IFBLK,
345 [EXT2_FT_FIFO] S_IFIFO,
346 [EXT2_FT_SOCK] S_IFSOCK,
347 [EXT2_FT_SYMLINK] S_IFLNK
351 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
352 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
353 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
354 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
355 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
356 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
357 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
358 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
361 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
363 mode_t mode = inode->i_mode;
364 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
367 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
369 loff_t pos = filp->f_pos;
370 struct inode *inode = filp->f_dentry->d_inode;
371 // XXX struct super_block *sb = inode->i_sb;
372 unsigned offset = pos & ~PAGE_CACHE_MASK;
373 unsigned long n = pos >> PAGE_CACHE_SHIFT;
374 unsigned long npages = dir_pages(inode);
375 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
376 unsigned char *types = NULL;
377 int need_revalidate = (filp->f_version != inode->i_version);
380 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
383 types = ext2_filetype_table;
385 for ( ; n < npages; n++, offset = 0) {
388 struct page *page = ext2_get_page(inode, n);
390 /* size might have been updated by mdc_readpage */
391 npages = dir_pages(inode);
395 kaddr = page_address(page);
396 if (need_revalidate) {
397 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
400 de = (ext2_dirent *)(kaddr+offset);
401 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
402 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
405 unsigned char d_type = DT_UNKNOWN;
407 if (types && de->file_type < EXT2_FT_MAX)
408 d_type = types[de->file_type];
410 offset = (char *)de - kaddr;
411 over = filldir(dirent, de->name, de->name_len,
412 (n<<PAGE_CACHE_SHIFT) | offset,
413 le32_to_cpu(de->inode), d_type);
423 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
424 filp->f_version = inode->i_version;
432 * finds an entry in the specified directory with the wanted name. It
433 * returns the page in which the entry was found, and the entry itself
434 * (as a parameter - res_dir). Page is returned mapped and unlocked.
435 * Entry is guaranteed to be valid.
437 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
438 struct dentry *dentry, struct page ** res_page)
440 const char *name = dentry->d_name.name;
441 int namelen = dentry->d_name.len;
442 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
443 unsigned long start, n;
444 unsigned long npages = dir_pages(dir);
445 struct page *page = NULL;
451 // start = dir->u.ext2_i.i_dir_start_lookup;
458 page = ext2_get_page(dir, n);
460 kaddr = page_address(page);
461 de = (ext2_dirent *) kaddr;
462 kaddr += PAGE_CACHE_SIZE - reclen;
463 while ((char *) de <= kaddr) {
464 if (ext2_match (namelen, name, de))
466 de = ext2_next_entry(de);
472 } while (n != start);
477 // dir->u.ext2_i.i_dir_start_lookup = n;
481 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
483 struct page *page = ext2_get_page(dir, 0);
484 ext2_dirent *de = NULL;
487 de = ext2_next_entry((ext2_dirent *) page_address(page));
493 obd_id ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
496 struct ext2_dir_entry_2 * de;
499 de = ext2_find_entry (dir, dentry, &page);
501 res = le32_to_cpu(de->inode);
502 *type = ll_dt2fmt[de->file_type];
504 page_cache_release(page);
509 /* Releases the page */
510 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
511 struct page *page, struct inode *inode)
513 unsigned from = (char *) de - (char *) page_address(page);
514 unsigned to = from + le16_to_cpu(de->rec_len);
518 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
521 de->inode = cpu_to_le32(inode->i_ino);
522 ext2_set_de_type (de, inode);
523 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
524 err = ext2_commit_chunk(page, from, to);
532 int ll_add_link (struct dentry *dentry, struct inode *inode)
534 struct inode *dir = dentry->d_parent->d_inode;
535 const char *name = dentry->d_name.name;
536 int namelen = dentry->d_name.len;
537 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
538 unsigned short rec_len, name_len;
539 struct page *page = NULL;
541 unsigned long npages = dir_pages(dir);
547 /* We take care of directory expansion in the same loop */
548 for (n = 0; n <= npages; n++) {
549 page = ext2_get_page(dir, n);
553 kaddr = page_address(page);
554 de = (ext2_dirent *)kaddr;
555 kaddr += PAGE_CACHE_SIZE - reclen;
556 while ((char *)de <= kaddr) {
558 if (ext2_match (namelen, name, de))
560 name_len = EXT2_DIR_REC_LEN(de->name_len);
561 rec_len = le16_to_cpu(de->rec_len);
562 if ( n==npages && rec_len == 0) {
563 CERROR("Fatal dir behaviour\n");
566 if (!de->inode && rec_len >= reclen)
568 if (rec_len >= name_len + reclen)
570 de = (ext2_dirent *) ((char *) de + rec_len);
578 from = (char*)de - (char*)page_address(page);
581 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
585 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
586 de1->rec_len = cpu_to_le16(rec_len - name_len);
587 de->rec_len = cpu_to_le16(name_len);
590 de->name_len = namelen;
591 memcpy (de->name, name, namelen);
592 de->inode = cpu_to_le32(inode->i_ino);
593 ext2_set_de_type (de, inode);
594 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
595 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
596 err = ext2_commit_chunk(page, from, to);
598 // change_inode happens with the commit_chunk
599 /* XXX OFFSET_CACHE */
610 * ext2_delete_entry deletes a directory entry by merging it with the
611 * previous entry. Page is up-to-date. Releases the page.
613 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
615 struct address_space *mapping = page->mapping;
616 struct inode *inode = mapping->host;
617 char *kaddr = page_address(page);
618 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
619 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
620 ext2_dirent * pde = NULL;
621 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
624 while ((char*)de < (char*)dir) {
626 de = ext2_next_entry(de);
629 from = (char*)pde - (char*)page_address(page);
631 err = mapping->a_ops->prepare_write(NULL, page, from, to);
635 pde->rec_len = cpu_to_le16(to-from);
637 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
638 err = ext2_commit_chunk(page, from, to);
645 * Set the first fragment of directory.
647 int ext2_make_empty(struct inode *inode, struct inode *parent)
649 struct address_space *mapping = inode->i_mapping;
650 struct page *page = grab_cache_page(mapping, 0);
651 unsigned chunk_size = ext2_chunk_size(inode);
652 struct ext2_dir_entry_2 * de;
663 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
667 de = (struct ext2_dir_entry_2 *) base;
669 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
670 memcpy (de->name, ".\0\0", 4);
671 de->inode = cpu_to_le32(inode->i_ino);
672 ext2_set_de_type (de, inode);
674 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
676 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
677 de->inode = cpu_to_le32(parent->i_ino);
678 memcpy (de->name, "..\0", 4);
679 ext2_set_de_type (de, inode);
681 err = ext2_commit_chunk(page, 0, chunk_size);
685 page_cache_release(page);
691 * routine to check that the specified directory is empty (for rmdir)
693 int ext2_empty_dir (struct inode * inode)
695 struct page *page = NULL;
696 unsigned long i, npages = dir_pages(inode);
698 for (i = 0; i < npages; i++) {
701 page = ext2_get_page(inode, i);
706 kaddr = page_address(page);
707 de = (ext2_dirent *)kaddr;
708 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
710 while ((char *)de <= kaddr) {
711 if (de->inode != 0) {
712 /* check for . and .. */
713 if (de->name[0] != '.')
715 if (de->name_len > 2)
717 if (de->name_len < 2) {
719 cpu_to_le32(inode->i_ino))
721 } else if (de->name[1] != '.')
724 de = ext2_next_entry(de);
735 struct file_operations ll_dir_operations = {
736 read: generic_read_dir,