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) 2002-2003, 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,
59 unsigned from, unsigned to)
61 CDEBUG(D_VFSTRACE, "VFS Op\n");
65 /* returns the page unlocked, but with a reference */
66 static int ll_dir_readpage(struct file *file, struct page *page)
68 struct inode *inode = page->mapping->host;
69 struct ll_sb_info *sbi = ll_i2sbi(inode);
73 struct ptlrpc_request *request;
74 struct lustre_handle lockh;
75 struct mds_body *body;
76 struct lookup_intent it = { .it_op = IT_READDIR };
80 CDEBUG(D_VFSTRACE, "VFS Op\n");
81 if ((inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_SHIFT <= page->index){
82 /* XXX why do we need this exactly, and why do we think that
83 * an all-zero directory page is useful?
85 CERROR("memsetting dir page %lu to zero (size %lld)\n",
86 page->index, inode->i_size);
87 memset(kmap(page), 0, PAGE_CACHE_SIZE);
89 GOTO(readpage_out, rc);
92 rc = mdc_enqueue(&sbi->ll_mdc_conn, LDLM_PLAIN, &it, LCK_PR, inode,
93 NULL, &lockh, NULL, 0, inode, sizeof(*inode));
94 request = (struct ptlrpc_request *)it.it_data;
96 ptlrpc_req_finished(request);
98 CERROR("lock enqueue: err: %d\n", rc);
102 ldlm_lock_dump_handle(D_OTHER, &lockh);
104 if (PageUptodate(page)) {
105 CERROR("Explain this please?\n");
106 GOTO(readpage_out, rc);
109 offset = page->index << PAGE_SHIFT;
111 rc = mdc_readpage(&sbi->ll_mdc_conn, inode->i_ino,
112 S_IFDIR, offset, buf, &request);
115 body = lustre_msg_buf(request->rq_repmsg, 0);
119 inode->i_size = body->size;
121 ptlrpc_req_finished(request);
126 SetPageUptodate(page);
129 ll_unlock(LCK_PR, &lockh);
131 CERROR("ll_unlock: err: %d\n", rc);
135 struct address_space_operations ll_dir_aops = {
136 readpage: ll_dir_readpage,
137 prepare_write: ll_dir_prepare_write
140 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,4,3))
141 int waitfor_one_page(struct page *page)
144 struct buffer_head *bh, *head = page->buffers;
149 if (buffer_req(bh) && !buffer_uptodate(bh))
151 } while ((bh = bh->b_this_page) != head);
154 #elif (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
155 int waitfor_one_page(struct page *page)
157 wait_on_page_locked(page);
163 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
164 * more robust, but we have what we have
166 static inline unsigned ext2_chunk_size(struct inode *inode)
168 return inode->i_sb->s_blocksize;
171 static inline void ext2_put_page(struct page *page)
174 page_cache_release(page);
177 static inline unsigned long dir_pages(struct inode *inode)
179 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
182 extern void set_page_clean(struct page *page);
184 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
186 struct inode *dir = page->mapping->host;
187 loff_t new_size = (page->index << PAGE_CACHE_SHIFT) + to;
190 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
191 dir->i_version = ++event;
193 if (new_size > dir->i_size)
194 dir->i_size = new_size;
195 SetPageUptodate(page);
196 set_page_clean(page);
198 //page->mapping->a_ops->commit_write(NULL, page, from, to);
200 // err = waitfor_one_page(page);
204 static void ext2_check_page(struct page *page)
206 struct inode *dir = page->mapping->host;
207 unsigned chunk_size = ext2_chunk_size(dir);
208 char *kaddr = page_address(page);
209 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
210 unsigned offs, rec_len;
211 unsigned limit = PAGE_CACHE_SIZE;
215 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
216 limit = dir->i_size & ~PAGE_CACHE_MASK;
217 if (limit & (chunk_size - 1)) {
218 CERROR("limit %d dir size %lld index %ld\n",
219 limit, dir->i_size, page->index);
222 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
223 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
224 p->rec_len = cpu_to_le16(chunk_size);
231 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
232 p = (ext2_dirent *)(kaddr + offs);
233 rec_len = le16_to_cpu(p->rec_len);
235 if (rec_len < EXT2_DIR_REC_LEN(1))
239 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
241 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
243 // if (le32_to_cpu(p->inode) > max_inumber)
249 SetPageChecked(page);
252 /* Too bad, we had an error */
255 CERROR("ext2_check_page"
256 "size of directory #%lu is not a multiple of chunk size\n",
261 error = "rec_len is smaller than minimal";
264 error = "unaligned directory entry";
267 error = "rec_len is too small for name_len";
270 error = "directory entry across blocks";
273 // error = "inode out of bounds";
275 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
276 "offset=%lu+%u, inode=%lu, rec_len=%d, name_len=%d",
277 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT), offs,
278 (unsigned long) le32_to_cpu(p->inode),
279 rec_len, p->name_len);
282 p = (ext2_dirent *)(kaddr + offs);
283 CERROR("ext2_check_page"
284 "entry in directory #%lu spans the page boundary"
285 "offset=%lu, inode=%lu",
286 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
287 (unsigned long) le32_to_cpu(p->inode));
289 SetPageChecked(page);
294 static struct page *ll_get_dir_page(struct inode *dir, unsigned long n)
296 struct address_space *mapping = dir->i_mapping;
297 struct page *page = read_cache_page(mapping, n,
298 (filler_t*)mapping->a_ops->readpage, NULL);
302 if (!PageUptodate(page))
304 if (!PageChecked(page))
305 ext2_check_page(page);
313 return ERR_PTR(-EIO);
317 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
319 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
321 static inline int ext2_match (int len, const char * const name,
322 struct ext2_dir_entry_2 * de)
324 if (len != de->name_len)
328 return !memcmp(name, de->name, len);
332 * p is at least 6 bytes before the end of page
334 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
336 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
339 static inline unsigned
340 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
342 ext2_dirent *de = (ext2_dirent*)(base + offset);
343 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
344 while ((char*)p < (char*)de)
345 p = ext2_next_entry(p);
346 return (char *)p - base;
349 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
350 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
351 [EXT2_FT_REG_FILE] DT_REG,
352 [EXT2_FT_DIR] DT_DIR,
353 [EXT2_FT_CHRDEV] DT_CHR,
354 [EXT2_FT_BLKDEV] DT_BLK,
355 [EXT2_FT_FIFO] DT_FIFO,
356 [EXT2_FT_SOCK] DT_SOCK,
357 [EXT2_FT_SYMLINK] DT_LNK,
360 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
362 [EXT2_FT_REG_FILE] S_IFREG,
363 [EXT2_FT_DIR] S_IFDIR,
364 [EXT2_FT_CHRDEV] S_IFCHR,
365 [EXT2_FT_BLKDEV] S_IFBLK,
366 [EXT2_FT_FIFO] S_IFIFO,
367 [EXT2_FT_SOCK] S_IFSOCK,
368 [EXT2_FT_SYMLINK] S_IFLNK
372 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
373 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
374 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
375 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
376 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
377 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
378 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
379 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
382 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
384 mode_t mode = inode->i_mode;
385 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
388 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
390 loff_t pos = filp->f_pos;
391 struct inode *inode = filp->f_dentry->d_inode;
392 // XXX struct super_block *sb = inode->i_sb;
393 unsigned offset = pos & ~PAGE_CACHE_MASK;
394 unsigned long n = pos >> PAGE_CACHE_SHIFT;
395 unsigned long npages = dir_pages(inode);
396 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
397 unsigned char *types = NULL;
398 int need_revalidate = (filp->f_version != inode->i_version);
401 CDEBUG(D_VFSTRACE, "VFS Op\n");
402 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
405 types = ext2_filetype_table;
407 for ( ; n < npages; n++, offset = 0) {
412 CDEBUG(D_EXT2, "reading %lu of dir %lu page %lu, size %llu\n",
413 PAGE_CACHE_SIZE, inode->i_ino, n, inode->i_size);
414 page = ll_get_dir_page(inode, n);
416 /* size might have been updated by mdc_readpage */
417 npages = dir_pages(inode);
421 kaddr = page_address(page);
422 if (need_revalidate) {
423 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
426 de = (ext2_dirent *)(kaddr+offset);
427 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
428 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
431 unsigned char d_type = DT_UNKNOWN;
433 if (types && de->file_type < EXT2_FT_MAX)
434 d_type = types[de->file_type];
436 offset = (char *)de - kaddr;
437 over = filldir(dirent, de->name, de->name_len,
438 (n<<PAGE_CACHE_SHIFT) | offset,
439 le32_to_cpu(de->inode), d_type);
449 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
450 filp->f_version = inode->i_version;
458 * finds an entry in the specified directory with the wanted name. It
459 * returns the page in which the entry was found, and the entry itself
460 * (as a parameter - res_dir). Page is returned mapped and unlocked.
461 * Entry is guaranteed to be valid.
463 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
464 struct dentry *dentry, struct page ** res_page)
466 const char *name = dentry->d_name.name;
467 int namelen = dentry->d_name.len;
468 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
469 unsigned long start, n;
470 unsigned long npages = dir_pages(dir);
471 struct page *page = NULL;
477 // start = dir->u.ext2_i.i_dir_start_lookup;
484 page = ll_get_dir_page(dir, n);
486 kaddr = page_address(page);
487 de = (ext2_dirent *) kaddr;
488 kaddr += PAGE_CACHE_SIZE - reclen;
489 while ((char *) de <= kaddr) {
490 if (ext2_match (namelen, name, de))
492 de = ext2_next_entry(de);
498 } while (n != start);
503 // dir->u.ext2_i.i_dir_start_lookup = n;
507 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
509 struct page *page = ll_get_dir_page(dir, 0);
510 ext2_dirent *de = NULL;
513 de = ext2_next_entry((ext2_dirent *) page_address(page));
519 obd_id ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
522 struct ext2_dir_entry_2 * de;
525 de = ext2_find_entry (dir, dentry, &page);
527 res = le32_to_cpu(de->inode);
528 *type = ll_dt2fmt[de->file_type];
530 page_cache_release(page);
535 /* Releases the page */
536 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
537 struct page *page, struct inode *inode)
539 unsigned from = (char *) de - (char *) page_address(page);
540 unsigned to = from + le16_to_cpu(de->rec_len);
544 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
547 de->inode = cpu_to_le32(inode->i_ino);
548 ext2_set_de_type (de, inode);
549 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
550 err = ext2_commit_chunk(page, from, to);
558 int ll_add_link (struct dentry *dentry, struct inode *inode)
560 struct inode *dir = dentry->d_parent->d_inode;
561 const char *name = dentry->d_name.name;
562 int namelen = dentry->d_name.len;
563 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
564 unsigned short rec_len, name_len;
565 struct page *page = NULL;
567 unsigned long npages = dir_pages(dir);
573 /* We take care of directory expansion in the same loop */
574 for (n = 0; n <= npages; n++) {
575 page = ll_get_dir_page(dir, n);
579 kaddr = page_address(page);
580 de = (ext2_dirent *)kaddr;
581 kaddr += PAGE_CACHE_SIZE - reclen;
582 while ((char *)de <= kaddr) {
584 if (ext2_match (namelen, name, de))
586 name_len = EXT2_DIR_REC_LEN(de->name_len);
587 rec_len = le16_to_cpu(de->rec_len);
588 if ( n==npages && rec_len == 0) {
589 CERROR("Fatal dir behaviour\n");
592 if (!de->inode && rec_len >= reclen)
594 if (rec_len >= name_len + reclen)
596 de = (ext2_dirent *) ((char *) de + rec_len);
604 from = (char*)de - (char*)page_address(page);
607 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
611 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
612 de1->rec_len = cpu_to_le16(rec_len - name_len);
613 de->rec_len = cpu_to_le16(name_len);
616 de->name_len = namelen;
617 memcpy (de->name, name, namelen);
618 de->inode = cpu_to_le32(inode->i_ino);
619 ext2_set_de_type (de, inode);
620 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
621 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
622 err = ext2_commit_chunk(page, from, to);
624 // change_inode happens with the commit_chunk
625 /* XXX OFFSET_CACHE */
636 * ext2_delete_entry deletes a directory entry by merging it with the
637 * previous entry. Page is up-to-date. Releases the page.
639 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
641 struct address_space *mapping = page->mapping;
642 struct inode *inode = mapping->host;
643 char *kaddr = page_address(page);
644 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
645 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
646 ext2_dirent * pde = NULL;
647 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
650 while ((char*)de < (char*)dir) {
652 de = ext2_next_entry(de);
655 from = (char*)pde - (char*)page_address(page);
657 err = mapping->a_ops->prepare_write(NULL, page, from, to);
661 pde->rec_len = cpu_to_le16(to-from);
663 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
664 err = ext2_commit_chunk(page, from, to);
671 * Set the first fragment of directory.
673 int ext2_make_empty(struct inode *inode, struct inode *parent)
675 struct address_space *mapping = inode->i_mapping;
676 struct page *page = grab_cache_page(mapping, 0);
677 unsigned chunk_size = ext2_chunk_size(inode);
678 struct ext2_dir_entry_2 * de;
689 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
693 de = (struct ext2_dir_entry_2 *) base;
695 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
696 memcpy (de->name, ".\0\0", 4);
697 de->inode = cpu_to_le32(inode->i_ino);
698 ext2_set_de_type (de, inode);
700 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
702 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
703 de->inode = cpu_to_le32(parent->i_ino);
704 memcpy (de->name, "..\0", 4);
705 ext2_set_de_type (de, inode);
707 err = ext2_commit_chunk(page, 0, chunk_size);
711 page_cache_release(page);
717 * routine to check that the specified directory is empty (for rmdir)
719 int ext2_empty_dir (struct inode * inode)
721 struct page *page = NULL;
722 unsigned long i, npages = dir_pages(inode);
724 for (i = 0; i < npages; i++) {
727 page = ll_get_dir_page(inode, i);
732 kaddr = page_address(page);
733 de = (ext2_dirent *)kaddr;
734 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
736 while ((char *)de <= kaddr) {
737 if (de->inode != 0) {
738 /* check for . and .. */
739 if (de->name[0] != '.')
741 if (de->name_len > 2)
743 if (de->name_len < 2) {
745 cpu_to_le32(inode->i_ino))
747 } else if (de->name[1] != '.')
750 de = ext2_next_entry(de);
761 static int ll_dir_ioctl(struct inode *inode, struct file *file,
762 unsigned int cmd, unsigned long arg)
764 struct ll_sb_info *sbi = ll_i2sbi(inode);
765 struct obd_ioctl_data *data;
767 CDEBUG(D_VFSTRACE, "VFS Op\n");
770 case IOC_MDC_LOOKUP: {
771 struct ptlrpc_request *request = NULL;
774 int namelen, rc, err, len = 0;
775 int ea_size = 0; // obd_size_wiremd(&sbi->ll_osc_conn, NULL);
778 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
783 filename = data->ioc_inlbuf1;
784 namelen = data->ioc_inllen1;
787 CERROR("IOC_MDC_LOOKUP missing filename\n");
788 GOTO(out, rc = -EINVAL);
791 valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE;
792 rc = mdc_getattr_name(&sbi->ll_mdc_conn, inode, filename,
793 namelen, valid, ea_size, &request);
795 CERROR("mdc_getattr_name: %d\n", rc);
798 struct mds_body *body;
799 body = lustre_msg_buf(request->rq_repmsg, 0);
800 /* surely there's a better way -phik */
801 data->ioc_obdo1.o_mode = body->mode;
802 data->ioc_obdo1.o_uid = body->uid;
803 data->ioc_obdo1.o_gid = body->gid;
806 err = copy_to_user((void *)arg, buf, len);
808 GOTO(out_req, rc = -EFAULT);
812 ptlrpc_req_finished(request);
818 CERROR("unrecognized ioctl %#x\n", cmd);
823 struct file_operations ll_dir_operations = {
824 read: generic_read_dir,