2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License v.2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/posix_acl.h>
16 #include <linux/sort.h>
17 #include <linux/gfs2_ondisk.h>
18 #include <linux/crc32.h>
19 #include <asm/semaphore.h>
22 #include "lm_interface.h"
33 #include "ops_address.h"
35 #include "ops_inode.h"
43 * inode_attr_in - Copy attributes from the dinode into the VFS inode
44 * @ip: The GFS2 inode (with embedded disk inode data)
45 * @inode: The Linux VFS inode
49 static void inode_attr_in(struct gfs2_inode *ip, struct inode *inode)
51 inode->i_ino = ip->i_num.no_formal_ino;
53 switch (ip->i_di.di_mode & S_IFMT) {
56 inode->i_rdev = MKDEV(ip->i_di.di_major, ip->i_di.di_minor);
63 inode->i_mode = ip->i_di.di_mode;
64 inode->i_nlink = ip->i_di.di_nlink;
65 inode->i_uid = ip->i_di.di_uid;
66 inode->i_gid = ip->i_di.di_gid;
67 i_size_write(inode, ip->i_di.di_size);
68 inode->i_atime.tv_sec = ip->i_di.di_atime;
69 inode->i_mtime.tv_sec = ip->i_di.di_mtime;
70 inode->i_ctime.tv_sec = ip->i_di.di_ctime;
71 inode->i_atime.tv_nsec = 0;
72 inode->i_mtime.tv_nsec = 0;
73 inode->i_ctime.tv_nsec = 0;
74 inode->i_blksize = PAGE_SIZE;
75 inode->i_blocks = ip->i_di.di_blocks <<
76 (ip->i_sbd->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT);
78 if (ip->i_di.di_flags & GFS2_DIF_IMMUTABLE)
79 inode->i_flags |= S_IMMUTABLE;
81 inode->i_flags &= ~S_IMMUTABLE;
83 if (ip->i_di.di_flags & GFS2_DIF_APPENDONLY)
84 inode->i_flags |= S_APPEND;
86 inode->i_flags &= ~S_APPEND;
90 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode
91 * @ip: The GFS2 inode (with embedded disk inode data)
95 void gfs2_inode_attr_in(struct gfs2_inode *ip)
99 inode = gfs2_ip2v_lookup(ip);
101 inode_attr_in(ip, inode);
107 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode
108 * @ip: The GFS2 inode
110 * Only copy out the attributes that we want the VFS layer
111 * to be able to modify.
114 void gfs2_inode_attr_out(struct gfs2_inode *ip)
116 struct inode *inode = ip->i_vnode;
118 gfs2_assert_withdraw(ip->i_sbd,
119 (ip->i_di.di_mode & S_IFMT) == (inode->i_mode & S_IFMT));
120 ip->i_di.di_mode = inode->i_mode;
121 ip->i_di.di_uid = inode->i_uid;
122 ip->i_di.di_gid = inode->i_gid;
123 ip->i_di.di_atime = inode->i_atime.tv_sec;
124 ip->i_di.di_mtime = inode->i_mtime.tv_sec;
125 ip->i_di.di_ctime = inode->i_ctime.tv_sec;
129 * gfs2_ip2v_lookup - Get the struct inode for a struct gfs2_inode
130 * @ip: the struct gfs2_inode to get the struct inode for
132 * Returns: A VFS inode, or NULL if none
135 struct inode *gfs2_ip2v_lookup(struct gfs2_inode *ip)
137 struct inode *inode = NULL;
139 gfs2_assert_warn(ip->i_sbd, test_bit(GIF_MIN_INIT, &ip->i_flags));
141 spin_lock(&ip->i_spin);
143 inode = igrab(ip->i_vnode);
144 spin_unlock(&ip->i_spin);
150 * gfs2_ip2v - Get/Create a struct inode for a struct gfs2_inode
151 * @ip: the struct gfs2_inode to get the struct inode for
153 * Returns: A VFS inode, or NULL if no mem
156 struct inode *gfs2_ip2v(struct gfs2_inode *ip)
158 struct inode *inode, *tmp;
160 inode = gfs2_ip2v_lookup(ip);
164 tmp = new_inode(ip->i_sbd->sd_vfs);
168 inode_attr_in(ip, tmp);
170 if (S_ISREG(ip->i_di.di_mode)) {
171 tmp->i_op = &gfs2_file_iops;
172 tmp->i_fop = &gfs2_file_fops;
173 tmp->i_mapping->a_ops = &gfs2_file_aops;
174 } else if (S_ISDIR(ip->i_di.di_mode)) {
175 tmp->i_op = &gfs2_dir_iops;
176 tmp->i_fop = &gfs2_dir_fops;
177 } else if (S_ISLNK(ip->i_di.di_mode)) {
178 tmp->i_op = &gfs2_symlink_iops;
180 tmp->i_op = &gfs2_dev_iops;
181 init_special_inode(tmp, tmp->i_mode, tmp->i_rdev);
184 tmp->u.generic_ip = NULL;
187 spin_lock(&ip->i_spin);
190 inode = igrab(ip->i_vnode);
191 spin_unlock(&ip->i_spin);
204 inode->u.generic_ip = ip;
206 spin_unlock(&ip->i_spin);
208 insert_inode_hash(inode);
213 static int iget_test(struct inode *inode, void *opaque)
215 struct gfs2_inode *ip = inode->u.generic_ip;
216 struct gfs2_inum *inum = (struct gfs2_inum *)opaque;
218 if (ip && ip->i_num.no_addr == inum->no_addr)
224 struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum *inum)
226 return ilookup5(sb, (unsigned long)inum->no_formal_ino,
230 void gfs2_inode_min_init(struct gfs2_inode *ip, unsigned int type)
232 if (!test_and_set_bit(GIF_MIN_INIT, &ip->i_flags)) {
233 ip->i_di.di_nlink = 1;
234 ip->i_di.di_mode = DT2IF(type);
239 * gfs2_inode_refresh - Refresh the incore copy of the dinode
240 * @ip: The GFS2 inode
245 int gfs2_inode_refresh(struct gfs2_inode *ip)
247 struct buffer_head *dibh;
250 error = gfs2_meta_inode_buffer(ip, &dibh);
254 if (gfs2_metatype_check(ip->i_sbd, dibh, GFS2_METATYPE_DI)) {
259 gfs2_dinode_in(&ip->i_di, dibh->b_data);
260 set_bit(GIF_MIN_INIT, &ip->i_flags);
264 if (ip->i_num.no_addr != ip->i_di.di_num.no_addr) {
265 if (gfs2_consist_inode(ip))
266 gfs2_dinode_print(&ip->i_di);
269 if (ip->i_num.no_formal_ino != ip->i_di.di_num.no_formal_ino)
272 ip->i_vn = ip->i_gl->gl_vn;
278 * inode_create - create a struct gfs2_inode
279 * @i_gl: The glock covering the inode
280 * @inum: The inode number
281 * @io_gl: the iopen glock to acquire/hold (using holder in new gfs2_inode)
282 * @io_state: the state the iopen glock should be acquired in
283 * @ipp: pointer to put the returned inode in
288 static int inode_create(struct gfs2_glock *i_gl, const struct gfs2_inum *inum,
289 struct gfs2_glock *io_gl, unsigned int io_state,
290 struct gfs2_inode **ipp, int need_lock)
292 struct gfs2_sbd *sdp = i_gl->gl_sbd;
293 struct gfs2_inode *ip;
296 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
299 memset(ip, 0, sizeof(struct gfs2_inode));
301 atomic_set(&ip->i_count, 1);
302 ip->i_vn = i_gl->gl_vn - 1;
305 spin_lock_init(&ip->i_spin);
306 init_rwsem(&ip->i_rw_mutex);
307 ip->i_greedy = gfs2_tune_get(sdp, gt_greedy_default);
310 error = gfs2_glock_nq_init(io_gl,
311 io_state, GL_LOCAL_EXCL | GL_EXACT,
316 spin_lock(&io_gl->gl_spin);
317 gfs2_glock_hold(i_gl);
318 io_gl->gl_object = i_gl;
319 spin_unlock(&io_gl->gl_spin);
322 gfs2_glock_hold(i_gl);
323 i_gl->gl_object = ip;
324 atomic_inc(&sdp->sd_inode_count);
329 gfs2_meta_cache_flush(ip);
330 kmem_cache_free(gfs2_inode_cachep, ip);
336 * gfs2_inode_get - Create or get a reference on an inode
337 * @i_gl: The glock covering the inode
338 * @inum: The inode number
340 * @ipp: pointer to put the returned inode in
345 int gfs2_inode_get(struct gfs2_glock *i_gl, const struct gfs2_inum *inum,
346 int create, struct gfs2_inode **ipp)
348 struct gfs2_sbd *sdp = i_gl->gl_sbd;
349 struct gfs2_glock *io_gl;
352 gfs2_glmutex_lock(i_gl);
354 *ipp = i_gl->gl_object;
357 if ((*ipp)->i_num.no_formal_ino != inum->no_formal_ino)
359 atomic_inc(&(*ipp)->i_count);
367 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops,
370 error = inode_create(i_gl, inum, io_gl, LM_ST_SHARED, ipp, 1);
371 gfs2_glock_put(io_gl);
375 gfs2_glmutex_unlock(i_gl);
380 void gfs2_inode_hold(struct gfs2_inode *ip)
382 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
383 atomic_inc(&ip->i_count);
386 void gfs2_inode_put(struct gfs2_inode *ip)
388 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
389 atomic_dec(&ip->i_count);
392 void gfs2_inode_destroy(struct gfs2_inode *ip, int unlock)
394 struct gfs2_sbd *sdp = ip->i_sbd;
395 struct gfs2_glock *i_gl = ip->i_gl;
397 gfs2_assert_warn(sdp, !atomic_read(&ip->i_count));
399 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl;
400 gfs2_assert(sdp, io_gl->gl_object == i_gl);
402 spin_lock(&io_gl->gl_spin);
403 io_gl->gl_object = NULL;
404 spin_unlock(&io_gl->gl_spin);
405 gfs2_glock_put(i_gl);
407 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
410 gfs2_meta_cache_flush(ip);
411 kmem_cache_free(gfs2_inode_cachep, ip);
413 i_gl->gl_object = NULL;
414 gfs2_glock_put(i_gl);
416 atomic_dec(&sdp->sd_inode_count);
419 static int dinode_dealloc(struct gfs2_inode *ip, struct gfs2_unlinked *ul)
421 struct gfs2_sbd *sdp = ip->i_sbd;
422 struct gfs2_alloc *al;
423 struct gfs2_rgrpd *rgd;
426 if (ip->i_di.di_blocks != 1) {
427 if (gfs2_consist_inode(ip))
428 gfs2_dinode_print(&ip->i_di);
432 al = gfs2_alloc_get(ip);
434 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
438 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
442 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
444 gfs2_consist_inode(ip);
446 goto out_rindex_relse;
449 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
452 goto out_rindex_relse;
454 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
455 RES_STATFS + RES_QUOTA, 1);
459 gfs2_trans_add_gl(ip->i_gl);
461 gfs2_free_di(rgd, ip);
463 error = gfs2_unlinked_ondisk_rm(sdp, ul);
466 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags);
469 gfs2_glock_dq_uninit(&al->al_rgd_gh);
472 gfs2_glock_dq_uninit(&al->al_ri_gh);
475 gfs2_quota_unhold(ip);
484 * inode_dealloc - Deallocate all on-disk blocks for an inode (dinode)
485 * @sdp: the filesystem
486 * @inum: the inode number to deallocate
487 * @io_gh: a holder for the iopen glock for this inode
489 * N.B. When we enter this we already hold the iopen glock and getting
490 * the glock for the inode means that we are grabbing the locks in the
491 * "wrong" order so we must only so a try lock operation and fail if we
492 * don't get the lock. Thats ok, since if we fail it means someone else
493 * is using the inode still and thus we shouldn't be deallocating it
499 static int inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul,
500 struct gfs2_holder *io_gh)
502 struct gfs2_inode *ip;
503 struct gfs2_holder i_gh;
506 error = gfs2_glock_nq_num(sdp, ul->ul_ut.ut_inum.no_addr,
507 &gfs2_inode_glops, LM_ST_EXCLUSIVE,
508 LM_FLAG_TRY_1CB, &i_gh);
513 return 1; /* or back off and relock in different order? */
518 gfs2_assert_warn(sdp, !i_gh.gh_gl->gl_object);
519 error = inode_create(i_gh.gh_gl, &ul->ul_ut.ut_inum, io_gh->gh_gl,
520 LM_ST_EXCLUSIVE, &ip, 0);
525 error = gfs2_inode_refresh(ip);
529 if (ip->i_di.di_nlink) {
530 if (gfs2_consist_inode(ip))
531 gfs2_dinode_print(&ip->i_di);
536 if (S_ISDIR(ip->i_di.di_mode) &&
537 (ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
538 error = gfs2_dir_exhash_dealloc(ip);
543 if (ip->i_di.di_eattr) {
544 error = gfs2_ea_dealloc(ip);
549 if (!gfs2_is_stuffed(ip)) {
550 error = gfs2_file_dealloc(ip);
555 error = dinode_dealloc(ip, ul);
560 gfs2_glmutex_lock(i_gh.gh_gl);
562 gfs2_inode_destroy(ip, 0);
563 gfs2_glmutex_unlock(i_gh.gh_gl);
566 gfs2_glock_dq_uninit(&i_gh);
572 * try_inode_dealloc - Try to deallocate an inode and all its blocks
573 * @sdp: the filesystem
575 * Returns: 0 on success, -errno on error, 1 on busy (inode open)
578 static int try_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
581 struct gfs2_holder iogh;
583 gfs2_try_toss_inode(sdp, &ul->ul_ut.ut_inum);
584 error = gfs2_glock_nq_num(sdp, ul->ul_ut.ut_inum.no_addr,
585 &gfs2_iopen_glops, LM_ST_EXCLUSIVE,
586 LM_FLAG_TRY_1CB, &iogh);
596 error = inode_dealloc(sdp, ul, &iogh);
597 gfs2_glock_dq_uninit(&iogh);
602 static int inode_dealloc_uninit(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
604 struct gfs2_rgrpd *rgd;
605 struct gfs2_holder ri_gh, rgd_gh;
608 error = gfs2_rindex_hold(sdp, &ri_gh);
612 rgd = gfs2_blk2rgrpd(sdp, ul->ul_ut.ut_inum.no_addr);
619 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rgd_gh);
623 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED + RES_STATFS, 0);
627 gfs2_free_uninit_di(rgd, ul->ul_ut.ut_inum.no_addr);
628 gfs2_unlinked_ondisk_rm(sdp, ul);
633 gfs2_glock_dq_uninit(&rgd_gh);
635 gfs2_glock_dq_uninit(&ri_gh);
640 int gfs2_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
642 if (ul->ul_ut.ut_flags & GFS2_UTF_UNINIT)
643 return inode_dealloc_uninit(sdp, ul);
645 return try_inode_dealloc(sdp, ul);
649 * gfs2_change_nlink - Change nlink count on inode
650 * @ip: The GFS2 inode
651 * @diff: The change in the nlink count required
656 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
658 struct buffer_head *dibh;
662 nlink = ip->i_di.di_nlink + diff;
664 /* If we are reducing the nlink count, but the new value ends up being
665 bigger than the old one, we must have underflowed. */
666 if (diff < 0 && nlink > ip->i_di.di_nlink) {
667 if (gfs2_consist_inode(ip))
668 gfs2_dinode_print(&ip->i_di);
672 error = gfs2_meta_inode_buffer(ip, &dibh);
676 ip->i_di.di_nlink = nlink;
677 ip->i_di.di_ctime = get_seconds();
679 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
680 gfs2_dinode_out(&ip->i_di, dibh->b_data);
686 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
689 gfs2_str2qstr(&qstr, name);
690 return gfs2_lookupi(dip, &qstr, 1, NULL);
695 * gfs2_lookupi - Look up a filename in a directory and return its inode
696 * @d_gh: An initialized holder for the directory glock
697 * @name: The name of the inode to look for
698 * @is_root: If 1, ignore the caller's permissions
699 * @i_gh: An uninitialized holder for the new inode glock
701 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless
707 struct inode *gfs2_lookupi(struct inode *dir, struct qstr *name, int is_root,
708 struct nameidata *nd)
711 struct super_block *sb = dir->i_sb;
712 struct gfs2_inode *ipp;
713 struct gfs2_inode *dip = dir->u.generic_ip;
714 struct gfs2_sbd *sdp = dip->i_sbd;
715 struct gfs2_holder d_gh;
716 struct gfs2_inum inum;
718 struct gfs2_glock *gl;
720 struct inode *inode = NULL;
722 if (!name->len || name->len > GFS2_FNAMESIZE)
723 return ERR_PTR(-ENAMETOOLONG);
725 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
726 (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
727 dir == sb->s_root->d_inode)) {
728 gfs2_inode_hold(dip);
733 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
735 return ERR_PTR(error);
738 error = gfs2_repermission(dip->i_vnode, MAY_EXEC, NULL);
743 error = gfs2_dir_search(dir, name, &inum, &type);
747 error = gfs2_glock_get(sdp, inum.no_addr, &gfs2_inode_glops,
752 error = gfs2_inode_get(gl, &inum, CREATE, &ipp);
754 gfs2_inode_min_init(ipp, type);
759 gfs2_glock_dq_uninit(&d_gh);
761 if (error == -ENOENT)
764 inode = gfs2_ip2v(ipp);
767 return ERR_PTR(-ENOMEM);
770 return ERR_PTR(error);
773 static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino)
775 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
776 struct buffer_head *bh;
777 struct gfs2_inum_range ir;
780 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
783 mutex_lock(&sdp->sd_inum_mutex);
785 error = gfs2_meta_inode_buffer(ip, &bh);
787 mutex_unlock(&sdp->sd_inum_mutex);
792 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
795 *formal_ino = ir.ir_start++;
797 gfs2_trans_add_bh(ip->i_gl, bh, 1);
798 gfs2_inum_range_out(&ir,
799 bh->b_data + sizeof(struct gfs2_dinode));
801 mutex_unlock(&sdp->sd_inum_mutex);
808 mutex_unlock(&sdp->sd_inum_mutex);
814 static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino)
816 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
817 struct gfs2_inode *m_ip = sdp->sd_inum_inode->u.generic_ip;
818 struct gfs2_holder gh;
819 struct buffer_head *bh;
820 struct gfs2_inum_range ir;
823 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
827 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
830 mutex_lock(&sdp->sd_inum_mutex);
832 error = gfs2_meta_inode_buffer(ip, &bh);
836 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
839 struct buffer_head *m_bh;
842 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
846 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode));
847 x = y = be64_to_cpu(x);
849 ir.ir_length = GFS2_INUM_QUANTUM;
850 x += GFS2_INUM_QUANTUM;
852 gfs2_consist_inode(m_ip);
854 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
855 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x;
860 *formal_ino = ir.ir_start++;
863 gfs2_trans_add_bh(ip->i_gl, bh, 1);
864 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
870 mutex_unlock(&sdp->sd_inum_mutex);
874 gfs2_glock_dq_uninit(&gh);
879 static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum)
883 error = pick_formal_ino_1(sdp, inum);
887 error = pick_formal_ino_2(sdp, inum);
893 * create_ok - OK to create a new on-disk inode here?
894 * @dip: Directory in which dinode is to be created
895 * @name: Name of new dinode
901 static int create_ok(struct gfs2_inode *dip, struct qstr *name,
906 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
910 /* Don't create entries in an unlinked directory */
911 if (!dip->i_di.di_nlink)
914 error = gfs2_dir_search(dip->i_vnode, name, NULL, NULL);
925 if (dip->i_di.di_entries == (uint32_t)-1)
927 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1)
933 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
934 unsigned int *uid, unsigned int *gid)
936 if (dip->i_sbd->sd_args.ar_suiddir &&
937 (dip->i_di.di_mode & S_ISUID) &&
941 else if (dip->i_di.di_uid != current->fsuid)
943 *uid = dip->i_di.di_uid;
945 *uid = current->fsuid;
947 if (dip->i_di.di_mode & S_ISGID) {
950 *gid = dip->i_di.di_gid;
952 *gid = current->fsgid;
955 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_unlinked *ul)
957 struct gfs2_sbd *sdp = dip->i_sbd;
962 dip->i_alloc.al_requested = RES_DINODE;
963 error = gfs2_inplace_reserve(dip);
967 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
972 ul->ul_ut.ut_inum.no_addr = gfs2_alloc_di(dip);
974 ul->ul_ut.ut_flags = GFS2_UTF_UNINIT;
975 error = gfs2_unlinked_ondisk_add(sdp, ul);
980 gfs2_inplace_release(dip);
989 * init_dinode - Fill in a new dinode structure
990 * @dip: the directory this inode is being created in
991 * @gl: The glock covering the new inode
992 * @inum: the inode number
993 * @mode: the file permissions
999 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1000 struct gfs2_inum *inum, unsigned int mode,
1001 unsigned int uid, unsigned int gid)
1003 struct gfs2_sbd *sdp = dip->i_sbd;
1004 struct gfs2_dinode *di;
1005 struct buffer_head *dibh;
1007 dibh = gfs2_meta_new(gl, inum->no_addr);
1008 gfs2_trans_add_bh(gl, dibh, 1);
1009 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
1010 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1011 di = (struct gfs2_dinode *)dibh->b_data;
1013 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
1014 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
1015 di->di_mode = cpu_to_be32(mode);
1016 di->di_uid = cpu_to_be32(uid);
1017 di->di_gid = cpu_to_be32(gid);
1018 di->di_nlink = cpu_to_be32(0);
1019 di->di_size = cpu_to_be64(0);
1020 di->di_blocks = cpu_to_be64(1);
1021 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
1022 di->di_major = di->di_minor = cpu_to_be32(0);
1023 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
1024 di->__pad[0] = di->__pad[1] = 0;
1025 di->di_flags = cpu_to_be32(0);
1027 if (S_ISREG(mode)) {
1028 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
1029 gfs2_tune_get(sdp, gt_new_files_jdata))
1030 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
1031 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
1032 gfs2_tune_get(sdp, gt_new_files_directio))
1033 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
1034 } else if (S_ISDIR(mode)) {
1035 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1036 GFS2_DIF_INHERIT_DIRECTIO);
1037 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1038 GFS2_DIF_INHERIT_JDATA);
1042 di->di_height = cpu_to_be32(0);
1045 di->di_depth = cpu_to_be16(0);
1046 di->di_entries = cpu_to_be32(0);
1047 memset(&di->__pad4, 0, sizeof(di->__pad4));
1048 di->di_eattr = cpu_to_be64(0);
1049 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
1054 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1055 unsigned int mode, struct gfs2_unlinked *ul)
1057 struct gfs2_sbd *sdp = dip->i_sbd;
1058 unsigned int uid, gid;
1061 munge_mode_uid_gid(dip, &mode, &uid, &gid);
1063 gfs2_alloc_get(dip);
1065 error = gfs2_quota_lock(dip, uid, gid);
1069 error = gfs2_quota_check(dip, uid, gid);
1073 error = gfs2_trans_begin(sdp, RES_DINODE + RES_UNLINKED +
1078 ul->ul_ut.ut_flags = 0;
1079 error = gfs2_unlinked_ondisk_munge(sdp, ul);
1081 init_dinode(dip, gl, &ul->ul_ut.ut_inum,
1084 gfs2_quota_change(dip, +1, uid, gid);
1086 gfs2_trans_end(sdp);
1089 gfs2_quota_unlock(dip);
1092 gfs2_alloc_put(dip);
1097 static int link_dinode(struct gfs2_inode *dip, struct qstr *name,
1098 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1100 struct gfs2_sbd *sdp = dip->i_sbd;
1101 struct gfs2_alloc *al;
1103 struct buffer_head *dibh;
1106 al = gfs2_alloc_get(dip);
1108 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1112 error = alloc_required = gfs2_diradd_alloc_required(dip->i_vnode, name);
1113 if (alloc_required < 0)
1115 if (alloc_required) {
1116 error = gfs2_quota_check(dip, dip->i_di.di_uid,
1119 goto fail_quota_locks;
1121 al->al_requested = sdp->sd_max_dirres;
1123 error = gfs2_inplace_reserve(dip);
1125 goto fail_quota_locks;
1127 error = gfs2_trans_begin(sdp,
1128 sdp->sd_max_dirres +
1129 al->al_rgd->rd_ri.ri_length +
1130 2 * RES_DINODE + RES_UNLINKED +
1131 RES_STATFS + RES_QUOTA, 0);
1135 error = gfs2_trans_begin(sdp,
1140 goto fail_quota_locks;
1143 error = gfs2_dir_add(dip->i_vnode, name, &ip->i_num, IF2DT(ip->i_di.di_mode));
1145 goto fail_end_trans;
1147 error = gfs2_meta_inode_buffer(ip, &dibh);
1149 goto fail_end_trans;
1150 ip->i_di.di_nlink = 1;
1151 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1152 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1155 error = gfs2_unlinked_ondisk_rm(sdp, ul);
1157 goto fail_end_trans;
1162 gfs2_trans_end(sdp);
1165 if (dip->i_alloc.al_rgd)
1166 gfs2_inplace_release(dip);
1169 gfs2_quota_unlock(dip);
1172 gfs2_alloc_put(dip);
1178 * gfs2_createi - Create a new inode
1179 * @ghs: An array of two holders
1180 * @name: The name of the new file
1181 * @mode: the permissions on the new inode
1183 * @ghs[0] is an initialized holder for the directory
1184 * @ghs[1] is the holder for the inode lock
1186 * If the return value is not NULL, the glocks on both the directory and the new
1187 * file are held. A transaction has been started and an inplace reservation
1193 struct inode *gfs2_createi(struct gfs2_holder *ghs, struct qstr *name,
1196 struct inode *inode;
1197 struct gfs2_inode *dip = ghs->gh_gl->gl_object;
1198 struct gfs2_sbd *sdp = dip->i_sbd;
1199 struct gfs2_unlinked *ul;
1200 struct gfs2_inode *ip;
1203 if (!name->len || name->len > GFS2_FNAMESIZE)
1204 return ERR_PTR(-ENAMETOOLONG);
1206 error = gfs2_unlinked_get(sdp, &ul);
1208 return ERR_PTR(error);
1210 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1211 error = gfs2_glock_nq(ghs);
1215 error = create_ok(dip, name, mode);
1219 error = pick_formal_ino(sdp, &ul->ul_ut.ut_inum.no_formal_ino);
1223 error = alloc_dinode(dip, ul);
1227 if (ul->ul_ut.ut_inum.no_addr < dip->i_num.no_addr) {
1230 error = gfs2_glock_nq_num(sdp,
1231 ul->ul_ut.ut_inum.no_addr,
1233 LM_ST_EXCLUSIVE, GL_SKIP,
1236 gfs2_unlinked_put(sdp, ul);
1237 return ERR_PTR(error);
1240 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1241 error = gfs2_glock_nq(ghs);
1243 gfs2_glock_dq_uninit(ghs + 1);
1244 gfs2_unlinked_put(sdp, ul);
1245 return ERR_PTR(error);
1248 error = create_ok(dip, name, mode);
1252 error = gfs2_glock_nq_num(sdp,
1253 ul->ul_ut.ut_inum.no_addr,
1255 LM_ST_EXCLUSIVE, GL_SKIP,
1261 error = make_dinode(dip, ghs[1].gh_gl, mode, ul);
1265 error = gfs2_inode_get(ghs[1].gh_gl, &ul->ul_ut.ut_inum, CREATE, &ip);
1269 error = gfs2_inode_refresh(ip);
1273 error = gfs2_acl_create(dip, ip);
1277 error = link_dinode(dip, name, ip, ul);
1281 gfs2_unlinked_put(sdp, ul);
1283 inode = gfs2_ip2v(ip);
1286 return ERR_PTR(-ENOMEM);
1293 gfs2_glock_dq_uninit(ghs + 1);
1299 gfs2_unlinked_put(sdp, ul);
1301 return ERR_PTR(error);
1305 * gfs2_unlinki - Unlink a file
1306 * @dip: The inode of the directory
1307 * @name: The name of the file to be unlinked
1308 * @ip: The inode of the file to be removed
1310 * Assumes Glocks on both dip and ip are held.
1315 int gfs2_unlinki(struct gfs2_inode *dip, struct qstr *name,
1316 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1318 struct gfs2_sbd *sdp = dip->i_sbd;
1321 error = gfs2_dir_del(dip, name);
1325 error = gfs2_change_nlink(ip, -1);
1329 /* If this inode is being unlinked from the directory structure,
1330 we need to mark that in the log so that it isn't lost during
1333 if (!ip->i_di.di_nlink) {
1334 ul->ul_ut.ut_inum = ip->i_num;
1335 error = gfs2_unlinked_ondisk_add(sdp, ul);
1337 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1344 * gfs2_rmdiri - Remove a directory
1345 * @dip: The parent directory of the directory to be removed
1346 * @name: The name of the directory to be removed
1347 * @ip: The GFS2 inode of the directory to be removed
1349 * Assumes Glocks on dip and ip are held
1354 int gfs2_rmdiri(struct gfs2_inode *dip, struct qstr *name,
1355 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1357 struct gfs2_sbd *sdp = dip->i_sbd;
1358 struct qstr dotname;
1361 if (ip->i_di.di_entries != 2) {
1362 if (gfs2_consist_inode(ip))
1363 gfs2_dinode_print(&ip->i_di);
1367 error = gfs2_dir_del(dip, name);
1371 error = gfs2_change_nlink(dip, -1);
1375 gfs2_str2qstr(&dotname, ".");
1376 error = gfs2_dir_del(ip, &dotname);
1381 dotname.name = "..";
1382 dotname.hash = gfs2_disk_hash(dotname.name, dotname.len);
1383 error = gfs2_dir_del(ip, &dotname);
1387 error = gfs2_change_nlink(ip, -2);
1391 /* This inode is being unlinked from the directory structure and
1392 we need to mark that in the log so that it isn't lost during
1395 ul->ul_ut.ut_inum = ip->i_num;
1396 error = gfs2_unlinked_ondisk_add(sdp, ul);
1398 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1404 * gfs2_unlink_ok - check to see that a inode is still in a directory
1405 * @dip: the directory
1406 * @name: the name of the file
1409 * Assumes that the lock on (at least) @dip is held.
1411 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1414 int gfs2_unlink_ok(struct gfs2_inode *dip, struct qstr *name,
1415 struct gfs2_inode *ip)
1417 struct gfs2_inum inum;
1421 if (IS_IMMUTABLE(ip->i_vnode) || IS_APPEND(ip->i_vnode))
1424 if ((dip->i_di.di_mode & S_ISVTX) &&
1425 dip->i_di.di_uid != current->fsuid &&
1426 ip->i_di.di_uid != current->fsuid &&
1427 !capable(CAP_FOWNER))
1430 if (IS_APPEND(dip->i_vnode))
1433 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
1437 error = gfs2_dir_search(dip->i_vnode, name, &inum, &type);
1441 if (!gfs2_inum_equal(&inum, &ip->i_num))
1444 if (IF2DT(ip->i_di.di_mode) != type) {
1445 gfs2_consist_inode(dip);
1453 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1457 * Follow @to back to the root and make sure we don't encounter @this
1458 * Assumes we already hold the rename lock.
1463 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1465 struct inode *dir = to->i_vnode;
1466 struct super_block *sb = dir->i_sb;
1471 gfs2_str2qstr(&dotdot, "..");
1476 if (dir == this->i_vnode) {
1480 if (dir == sb->s_root->d_inode) {
1485 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL);
1487 error = PTR_ERR(tmp);
1501 * gfs2_readlinki - return the contents of a symlink
1502 * @ip: the symlink's inode
1503 * @buf: a pointer to the buffer to be filled
1504 * @len: a pointer to the length of @buf
1506 * If @buf is too small, a piece of memory is kmalloc()ed and needs
1507 * to be freed by the caller.
1512 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1514 struct gfs2_holder i_gh;
1515 struct buffer_head *dibh;
1519 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1520 error = gfs2_glock_nq_atime(&i_gh);
1522 gfs2_holder_uninit(&i_gh);
1526 if (!ip->i_di.di_size) {
1527 gfs2_consist_inode(ip);
1532 error = gfs2_meta_inode_buffer(ip, &dibh);
1536 x = ip->i_di.di_size + 1;
1538 *buf = kmalloc(x, GFP_KERNEL);
1545 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1552 gfs2_glock_dq_uninit(&i_gh);
1558 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1559 * conditionally update the inode's atime
1560 * @gh: the holder to acquire
1562 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1563 * Update if the difference between the current time and the inode's current
1564 * atime is greater than an interval specified at mount.
1569 int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1571 struct gfs2_glock *gl = gh->gh_gl;
1572 struct gfs2_sbd *sdp = gl->gl_sbd;
1573 struct gfs2_inode *ip = gl->gl_object;
1574 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1579 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1580 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1581 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1584 state = gh->gh_state;
1585 flags = gh->gh_flags;
1587 error = gfs2_glock_nq(gh);
1591 if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1592 (sdp->sd_vfs->s_flags & MS_RDONLY))
1595 curtime = get_seconds();
1596 if (curtime - ip->i_di.di_atime >= quantum) {
1598 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY,
1600 error = gfs2_glock_nq(gh);
1604 /* Verify that atime hasn't been updated while we were
1605 trying to get exclusive lock. */
1607 curtime = get_seconds();
1608 if (curtime - ip->i_di.di_atime >= quantum) {
1609 struct buffer_head *dibh;
1611 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1612 if (error == -EROFS)
1617 error = gfs2_meta_inode_buffer(ip, &dibh);
1619 goto fail_end_trans;
1621 ip->i_di.di_atime = curtime;
1623 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1624 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1627 gfs2_trans_end(sdp);
1630 /* If someone else has asked for the glock,
1631 unlock and let them have it. Then reacquire
1632 in the original state. */
1633 if (gfs2_glock_is_blocking(gl)) {
1635 gfs2_holder_reinit(state, flags, gh);
1636 return gfs2_glock_nq(gh);
1643 gfs2_trans_end(sdp);
1652 * glock_compare_atime - Compare two struct gfs2_glock structures for sort
1653 * @arg_a: the first structure
1654 * @arg_b: the second structure
1656 * Returns: 1 if A > B
1661 static int glock_compare_atime(const void *arg_a, const void *arg_b)
1663 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1664 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1665 struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1666 struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1669 if (a->ln_number > b->ln_number)
1671 else if (a->ln_number < b->ln_number)
1674 if (gh_a->gh_state == LM_ST_SHARED &&
1675 gh_b->gh_state == LM_ST_EXCLUSIVE)
1677 else if (gh_a->gh_state == LM_ST_SHARED &&
1678 (gh_b->gh_flags & GL_ATIME))
1686 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an
1688 * @num_gh: the number of structures
1689 * @ghs: an array of struct gfs2_holder structures
1691 * Returns: 0 on success (all glocks acquired),
1692 * errno on failure (no glocks acquired)
1695 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs)
1697 struct gfs2_holder **p;
1705 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1706 if (ghs->gh_flags & GL_ATIME)
1707 error = gfs2_glock_nq_atime(ghs);
1709 error = gfs2_glock_nq(ghs);
1713 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1717 for (x = 0; x < num_gh; x++)
1720 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL);
1722 for (x = 0; x < num_gh; x++) {
1723 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1725 if (p[x]->gh_flags & GL_ATIME)
1726 error = gfs2_glock_nq_atime(p[x]);
1728 error = gfs2_glock_nq(p[x]);
1732 gfs2_glock_dq(p[x]);
1743 * gfs2_try_toss_vnode - See if we can toss a vnode from memory
1746 * Returns: 1 if the vnode was tossed
1749 void gfs2_try_toss_vnode(struct gfs2_inode *ip)
1751 struct inode *inode;
1753 inode = gfs2_ip2v_lookup(ip);
1757 d_prune_aliases(inode);
1759 if (S_ISDIR(ip->i_di.di_mode)) {
1760 struct list_head *head = &inode->i_dentry;
1761 struct dentry *d = NULL;
1763 spin_lock(&dcache_lock);
1764 if (list_empty(head))
1765 spin_unlock(&dcache_lock);
1767 d = list_entry(head->next, struct dentry, d_alias);
1769 spin_unlock(&dcache_lock);
1771 if (have_submounts(d))
1774 shrink_dcache_parent(d);
1776 d_prune_aliases(inode);
1787 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1789 struct buffer_head *dibh;
1792 error = gfs2_meta_inode_buffer(ip, &dibh);
1794 error = inode_setattr(ip->i_vnode, attr);
1795 gfs2_assert_warn(ip->i_sbd, !error);
1796 gfs2_inode_attr_out(ip);
1798 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1799 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1806 * gfs2_setattr_simple -
1810 * Called with a reference on the vnode.
1815 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1819 if (current->journal_info)
1820 return __gfs2_setattr_simple(ip, attr);
1822 error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0);
1826 error = __gfs2_setattr_simple(ip, attr);
1828 gfs2_trans_end(ip->i_sbd);
1833 int gfs2_repermission(struct inode *inode, int mask, struct nameidata *nd)
1835 return permission(inode, mask, nd);