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);
72 struct ptlrpc_request *request;
73 struct lustre_handle lockh;
74 struct mds_body *body;
75 struct lookup_intent it = { .it_op = IT_READDIR };
76 struct mdc_op_data data;
77 struct obd_device *obddev = class_conn2obd(&sbi->ll_mdc_conn);
78 struct ldlm_res_id res_id =
79 { .name = {inode->i_ino, (__u64)inode->i_generation} };
80 int flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_MATCH_DATA;
83 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
84 inode->i_generation, inode);
85 if ((inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_SHIFT <= page->index){
86 /* XXX why do we need this exactly, and why do we think that
87 * an all-zero directory page is useful?
89 CERROR("memsetting dir page %lu to zero (size %lld)\n",
90 page->index, inode->i_size);
91 memset(kmap(page), 0, PAGE_CACHE_SIZE);
93 GOTO(readpage_out, rc);
96 rc = ldlm_lock_match(obddev->obd_namespace, flags, &res_id,
97 LDLM_PLAIN, NULL, 0, LCK_PR, inode,
100 ll_prepare_mdc_op_data(&data, inode, NULL, NULL, 0, 0);
102 rc = mdc_enqueue(&sbi->ll_mdc_conn, LDLM_PLAIN, &it, LCK_PR,
103 &data, &lockh, NULL, 0,
104 ldlm_completion_ast, ll_mdc_blocking_ast,
106 request = (struct ptlrpc_request *)it.it_data;
108 ptlrpc_req_finished(request);
110 CERROR("lock enqueue: err: %d\n", rc);
115 ldlm_lock_dump_handle(D_OTHER, &lockh);
117 if (PageUptodate(page)) {
118 CERROR("Explain this please?\n");
119 GOTO(readpage_out, rc);
122 offset = page->index << PAGE_SHIFT;
123 rc = mdc_readpage(&sbi->ll_mdc_conn, inode->i_ino,
124 S_IFDIR, offset, page, &request);
126 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
127 LASSERT (body != NULL); /* checked by mdc_readpage() */
128 LASSERT_REPSWABBED (request, 0); /* swabbed by mdc_readpage() */
130 inode->i_size = body->size;
132 ptlrpc_req_finished(request);
137 SetPageUptodate(page);
140 ll_unlock(LCK_PR, &lockh);
142 CERROR("ll_unlock: err: %d\n", rc);
146 struct address_space_operations ll_dir_aops = {
147 readpage: ll_dir_readpage,
148 prepare_write: ll_dir_prepare_write
151 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,4,3))
152 int waitfor_one_page(struct page *page)
155 struct buffer_head *bh, *head = page->buffers;
160 if (buffer_req(bh) && !buffer_uptodate(bh))
162 } while ((bh = bh->b_this_page) != head);
165 #elif (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
166 int waitfor_one_page(struct page *page)
168 wait_on_page_locked(page);
174 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
175 * more robust, but we have what we have
177 static inline unsigned ext2_chunk_size(struct inode *inode)
179 return inode->i_sb->s_blocksize;
182 static inline void ext2_put_page(struct page *page)
185 page_cache_release(page);
188 static inline unsigned long dir_pages(struct inode *inode)
190 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
193 extern void set_page_clean(struct page *page);
195 static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
197 struct inode *dir = page->mapping->host;
198 loff_t new_size = (page->index << PAGE_CACHE_SHIFT) + to;
201 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
202 dir->i_version = ++event;
204 if (new_size > dir->i_size)
205 dir->i_size = new_size;
206 SetPageUptodate(page);
207 set_page_clean(page);
209 //page->mapping->a_ops->commit_write(NULL, page, from, to);
211 // err = waitfor_one_page(page);
215 static void ext2_check_page(struct page *page)
217 struct inode *dir = page->mapping->host;
218 unsigned chunk_size = ext2_chunk_size(dir);
219 char *kaddr = page_address(page);
220 // u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
221 unsigned offs, rec_len;
222 unsigned limit = PAGE_CACHE_SIZE;
226 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
227 limit = dir->i_size & ~PAGE_CACHE_MASK;
228 if (limit & (chunk_size - 1)) {
229 CERROR("limit %d dir size %lld index %ld\n",
230 limit, dir->i_size, page->index);
233 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
234 ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
235 p->rec_len = cpu_to_le16(chunk_size);
242 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
243 p = (ext2_dirent *)(kaddr + offs);
244 rec_len = le16_to_cpu(p->rec_len);
246 if (rec_len < EXT2_DIR_REC_LEN(1))
250 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
252 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
254 // if (le32_to_cpu(p->inode) > max_inumber)
260 SetPageChecked(page);
263 /* Too bad, we had an error */
266 CERROR("ext2_check_page"
267 "size of directory #%lu is not a multiple of chunk size\n",
272 error = "rec_len is smaller than minimal";
275 error = "unaligned directory entry";
278 error = "rec_len is too small for name_len";
281 error = "directory entry across blocks";
284 // error = "inode out of bounds";
286 CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
287 "offset=%lu+%u, inode=%lu, rec_len=%d, name_len=%d",
288 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT), offs,
289 (unsigned long) le32_to_cpu(p->inode),
290 rec_len, p->name_len);
293 p = (ext2_dirent *)(kaddr + offs);
294 CERROR("ext2_check_page"
295 "entry in directory #%lu spans the page boundary"
296 "offset=%lu, inode=%lu",
297 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
298 (unsigned long) le32_to_cpu(p->inode));
300 SetPageChecked(page);
305 static struct page *ll_get_dir_page(struct inode *dir, unsigned long n)
307 struct address_space *mapping = dir->i_mapping;
308 struct page *page = read_cache_page(mapping, n,
309 (filler_t*)mapping->a_ops->readpage, NULL);
313 if (!PageUptodate(page))
315 if (!PageChecked(page))
316 ext2_check_page(page);
324 return ERR_PTR(-EIO);
328 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
330 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
332 static inline int ext2_match (int len, const char * const name,
333 struct ext2_dir_entry_2 * de)
335 if (len != de->name_len)
339 return !memcmp(name, de->name, len);
343 * p is at least 6 bytes before the end of page
345 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
347 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
350 static inline unsigned
351 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
353 ext2_dirent *de = (ext2_dirent*)(base + offset);
354 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
355 while ((char*)p < (char*)de)
356 p = ext2_next_entry(p);
357 return (char *)p - base;
360 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
361 [EXT2_FT_UNKNOWN] DT_UNKNOWN,
362 [EXT2_FT_REG_FILE] DT_REG,
363 [EXT2_FT_DIR] DT_DIR,
364 [EXT2_FT_CHRDEV] DT_CHR,
365 [EXT2_FT_BLKDEV] DT_BLK,
366 [EXT2_FT_FIFO] DT_FIFO,
367 [EXT2_FT_SOCK] DT_SOCK,
368 [EXT2_FT_SYMLINK] DT_LNK,
371 static unsigned int ll_dt2fmt[DT_WHT + 1] = {
373 [EXT2_FT_REG_FILE] S_IFREG,
374 [EXT2_FT_DIR] S_IFDIR,
375 [EXT2_FT_CHRDEV] S_IFCHR,
376 [EXT2_FT_BLKDEV] S_IFBLK,
377 [EXT2_FT_FIFO] S_IFIFO,
378 [EXT2_FT_SOCK] S_IFSOCK,
379 [EXT2_FT_SYMLINK] S_IFLNK
383 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
384 [S_IFREG >> S_SHIFT] EXT2_FT_REG_FILE,
385 [S_IFDIR >> S_SHIFT] EXT2_FT_DIR,
386 [S_IFCHR >> S_SHIFT] EXT2_FT_CHRDEV,
387 [S_IFBLK >> S_SHIFT] EXT2_FT_BLKDEV,
388 [S_IFIFO >> S_SHIFT] EXT2_FT_FIFO,
389 [S_IFSOCK >> S_SHIFT] EXT2_FT_SOCK,
390 [S_IFLNK >> S_SHIFT] EXT2_FT_SYMLINK,
393 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
395 mode_t mode = inode->i_mode;
396 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
399 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
401 loff_t pos = filp->f_pos;
402 struct inode *inode = filp->f_dentry->d_inode;
403 // XXX struct super_block *sb = inode->i_sb;
404 unsigned offset = pos & ~PAGE_CACHE_MASK;
405 unsigned long n = pos >> PAGE_CACHE_SHIFT;
406 unsigned long npages = dir_pages(inode);
407 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
408 unsigned char *types = NULL;
409 int need_revalidate = (filp->f_version != inode->i_version);
412 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
413 inode->i_generation, inode);
414 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
417 types = ext2_filetype_table;
419 for ( ; n < npages; n++, offset = 0) {
424 CDEBUG(D_EXT2, "reading %lu of dir %lu page %lu, size %llu\n",
425 PAGE_CACHE_SIZE, inode->i_ino, n, inode->i_size);
426 page = ll_get_dir_page(inode, n);
428 /* size might have been updated by mdc_readpage */
429 npages = dir_pages(inode);
433 kaddr = page_address(page);
434 if (need_revalidate) {
435 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
438 de = (ext2_dirent *)(kaddr+offset);
439 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
440 for ( ;(char*)de <= limit; de = ext2_next_entry(de))
443 unsigned char d_type = DT_UNKNOWN;
445 if (types && de->file_type < EXT2_FT_MAX)
446 d_type = types[de->file_type];
448 offset = (char *)de - kaddr;
449 over = filldir(dirent, de->name, de->name_len,
450 (n<<PAGE_CACHE_SHIFT) | offset,
451 le32_to_cpu(de->inode), d_type);
461 filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
462 filp->f_version = inode->i_version;
470 * finds an entry in the specified directory with the wanted name. It
471 * returns the page in which the entry was found, and the entry itself
472 * (as a parameter - res_dir). Page is returned mapped and unlocked.
473 * Entry is guaranteed to be valid.
475 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
476 struct dentry *dentry, struct page ** res_page)
478 const char *name = dentry->d_name.name;
479 int namelen = dentry->d_name.len;
480 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
481 unsigned long start, n;
482 unsigned long npages = dir_pages(dir);
483 struct page *page = NULL;
489 // start = dir->u.ext2_i.i_dir_start_lookup;
496 page = ll_get_dir_page(dir, n);
498 kaddr = page_address(page);
499 de = (ext2_dirent *) kaddr;
500 kaddr += PAGE_CACHE_SIZE - reclen;
501 while ((char *) de <= kaddr) {
502 if (ext2_match (namelen, name, de))
504 de = ext2_next_entry(de);
510 } while (n != start);
515 // dir->u.ext2_i.i_dir_start_lookup = n;
519 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
521 struct page *page = ll_get_dir_page(dir, 0);
522 ext2_dirent *de = NULL;
525 de = ext2_next_entry((ext2_dirent *) page_address(page));
531 obd_id ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *type)
534 struct ext2_dir_entry_2 * de;
537 de = ext2_find_entry (dir, dentry, &page);
539 res = le32_to_cpu(de->inode);
540 *type = ll_dt2fmt[de->file_type];
542 page_cache_release(page);
547 /* Releases the page */
548 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
549 struct page *page, struct inode *inode)
551 unsigned from = (char *) de - (char *) page_address(page);
552 unsigned to = from + le16_to_cpu(de->rec_len);
556 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
559 de->inode = cpu_to_le32(inode->i_ino);
560 ext2_set_de_type (de, inode);
561 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
562 err = ext2_commit_chunk(page, from, to);
570 int ll_add_link (struct dentry *dentry, struct inode *inode)
572 struct inode *dir = dentry->d_parent->d_inode;
573 const char *name = dentry->d_name.name;
574 int namelen = dentry->d_name.len;
575 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
576 unsigned short rec_len, name_len;
577 struct page *page = NULL;
579 unsigned long npages = dir_pages(dir);
585 /* We take care of directory expansion in the same loop */
586 for (n = 0; n <= npages; n++) {
587 page = ll_get_dir_page(dir, n);
591 kaddr = page_address(page);
592 de = (ext2_dirent *)kaddr;
593 kaddr += PAGE_CACHE_SIZE - reclen;
594 while ((char *)de <= kaddr) {
596 if (ext2_match (namelen, name, de))
598 name_len = EXT2_DIR_REC_LEN(de->name_len);
599 rec_len = le16_to_cpu(de->rec_len);
600 if ( n==npages && rec_len == 0) {
601 CERROR("Fatal dir behaviour\n");
604 if (!de->inode && rec_len >= reclen)
606 if (rec_len >= name_len + reclen)
608 de = (ext2_dirent *) ((char *) de + rec_len);
616 from = (char*)de - (char*)page_address(page);
619 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
623 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
624 de1->rec_len = cpu_to_le16(rec_len - name_len);
625 de->rec_len = cpu_to_le16(name_len);
628 de->name_len = namelen;
629 memcpy (de->name, name, namelen);
630 de->inode = cpu_to_le32(inode->i_ino);
631 ext2_set_de_type (de, inode);
632 CDEBUG(D_INODE, "type set to %o\n", de->file_type);
633 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
634 err = ext2_commit_chunk(page, from, to);
636 // change_inode happens with the commit_chunk
637 /* XXX OFFSET_CACHE */
648 * ext2_delete_entry deletes a directory entry by merging it with the
649 * previous entry. Page is up-to-date. Releases the page.
651 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
653 struct address_space *mapping = page->mapping;
654 struct inode *inode = mapping->host;
655 char *kaddr = page_address(page);
656 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
657 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
658 ext2_dirent * pde = NULL;
659 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
662 while ((char*)de < (char*)dir) {
664 de = ext2_next_entry(de);
667 from = (char*)pde - (char*)page_address(page);
669 err = mapping->a_ops->prepare_write(NULL, page, from, to);
673 pde->rec_len = cpu_to_le16(to-from);
675 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
676 err = ext2_commit_chunk(page, from, to);
683 * Set the first fragment of directory.
685 int ext2_make_empty(struct inode *inode, struct inode *parent)
687 struct address_space *mapping = inode->i_mapping;
688 struct page *page = grab_cache_page(mapping, 0);
689 unsigned chunk_size = ext2_chunk_size(inode);
690 struct ext2_dir_entry_2 * de;
701 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
705 de = (struct ext2_dir_entry_2 *) base;
707 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
708 memcpy (de->name, ".\0\0", 4);
709 de->inode = cpu_to_le32(inode->i_ino);
710 ext2_set_de_type (de, inode);
712 de = (struct ext2_dir_entry_2 *) (base + EXT2_DIR_REC_LEN(1));
714 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
715 de->inode = cpu_to_le32(parent->i_ino);
716 memcpy (de->name, "..\0", 4);
717 ext2_set_de_type (de, inode);
719 err = ext2_commit_chunk(page, 0, chunk_size);
723 page_cache_release(page);
729 * routine to check that the specified directory is empty (for rmdir)
731 int ext2_empty_dir (struct inode * inode)
733 struct page *page = NULL;
734 unsigned long i, npages = dir_pages(inode);
736 for (i = 0; i < npages; i++) {
739 page = ll_get_dir_page(inode, i);
744 kaddr = page_address(page);
745 de = (ext2_dirent *)kaddr;
746 kaddr += PAGE_CACHE_SIZE-EXT2_DIR_REC_LEN(1);
748 while ((char *)de <= kaddr) {
749 if (de->inode != 0) {
750 /* check for . and .. */
751 if (de->name[0] != '.')
753 if (de->name_len > 2)
755 if (de->name_len < 2) {
757 cpu_to_le32(inode->i_ino))
759 } else if (de->name[1] != '.')
762 de = ext2_next_entry(de);
773 static int ll_dir_ioctl(struct inode *inode, struct file *file,
774 unsigned int cmd, unsigned long arg)
776 struct ll_sb_info *sbi = ll_i2sbi(inode);
777 struct obd_ioctl_data *data;
779 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%u\n", inode->i_ino,
780 inode->i_generation, inode, cmd);
782 if (_IOC_TYPE(cmd) == 'T') /* tty ioctls */
786 case IOC_MDC_LOOKUP: {
787 struct ptlrpc_request *request = NULL;
790 struct mds_body *body;
792 int namelen, rc, err, len = 0;
795 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
800 filename = data->ioc_inlbuf1;
801 namelen = data->ioc_inllen1;
804 CERROR("IOC_MDC_LOOKUP missing filename\n");
805 GOTO(out, rc = -EINVAL);
808 valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE;
809 ll_inode2fid(&fid, inode);
810 rc = mdc_getattr_name(&sbi->ll_mdc_conn, &fid,
811 filename, namelen, valid, 0, &request);
813 CERROR("mdc_getattr_name: %d\n", rc);
817 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
818 LASSERT(body != NULL); /* checked by mdc_getattr_name */
819 LASSERT_REPSWABBED(request, 0);/* swabbed by mdc_getattr_name */
821 /* surely there's a better way -phik */
822 data->ioc_obdo1.o_mode = body->mode;
823 data->ioc_obdo1.o_uid = body->uid;
824 data->ioc_obdo1.o_gid = body->gid;
826 ptlrpc_req_finished(request);
828 err = copy_to_user((void *)arg, buf, len);
830 GOTO(out, rc = -EFAULT);
834 obd_ioctl_freedata(buf, len);
838 CERROR("unrecognized ioctl %#x\n", cmd);
843 int ll_dir_open(struct inode *inode, struct file *file)
845 return ll_file_open(inode, file);
848 int ll_dir_release(struct inode *inode, struct file *file)
850 return ll_file_release(inode, file);
853 struct file_operations ll_dir_operations = {
855 release: ll_dir_release,
856 read: generic_read_dir,