2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 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>
21 #include "lm_interface.h"
32 #include "ops_address.h"
34 #include "ops_inode.h"
42 * inode_attr_in - Copy attributes from the dinode into the VFS inode
43 * @ip: The GFS2 inode (with embedded disk inode data)
44 * @inode: The Linux VFS inode
48 static void inode_attr_in(struct gfs2_inode *ip, struct inode *inode)
50 inode->i_ino = ip->i_num.no_formal_ino;
52 switch (ip->i_di.di_mode & S_IFMT) {
55 inode->i_rdev = MKDEV(ip->i_di.di_major, ip->i_di.di_minor);
62 inode->i_mode = ip->i_di.di_mode;
63 inode->i_nlink = ip->i_di.di_nlink;
64 inode->i_uid = ip->i_di.di_uid;
65 inode->i_gid = ip->i_di.di_gid;
66 i_size_write(inode, ip->i_di.di_size);
67 inode->i_atime.tv_sec = ip->i_di.di_atime;
68 inode->i_mtime.tv_sec = ip->i_di.di_mtime;
69 inode->i_ctime.tv_sec = ip->i_di.di_ctime;
70 inode->i_atime.tv_nsec = 0;
71 inode->i_mtime.tv_nsec = 0;
72 inode->i_ctime.tv_nsec = 0;
73 inode->i_blksize = PAGE_SIZE;
74 inode->i_blocks = ip->i_di.di_blocks <<
75 (ip->i_sbd->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT);
77 if (ip->i_di.di_flags & GFS2_DIF_IMMUTABLE)
78 inode->i_flags |= S_IMMUTABLE;
80 inode->i_flags &= ~S_IMMUTABLE;
82 if (ip->i_di.di_flags & GFS2_DIF_APPENDONLY)
83 inode->i_flags |= S_APPEND;
85 inode->i_flags &= ~S_APPEND;
89 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode
90 * @ip: The GFS2 inode (with embedded disk inode data)
94 void gfs2_inode_attr_in(struct gfs2_inode *ip)
98 inode = gfs2_ip2v_lookup(ip);
100 inode_attr_in(ip, inode);
106 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode
107 * @ip: The GFS2 inode
109 * Only copy out the attributes that we want the VFS layer
110 * to be able to modify.
113 void gfs2_inode_attr_out(struct gfs2_inode *ip)
115 struct inode *inode = ip->i_vnode;
117 gfs2_assert_withdraw(ip->i_sbd,
118 (ip->i_di.di_mode & S_IFMT) == (inode->i_mode & S_IFMT));
119 ip->i_di.di_mode = inode->i_mode;
120 ip->i_di.di_uid = inode->i_uid;
121 ip->i_di.di_gid = inode->i_gid;
122 ip->i_di.di_atime = inode->i_atime.tv_sec;
123 ip->i_di.di_mtime = inode->i_mtime.tv_sec;
124 ip->i_di.di_ctime = inode->i_ctime.tv_sec;
128 * gfs2_ip2v_lookup - Get the struct inode for a struct gfs2_inode
129 * @ip: the struct gfs2_inode to get the struct inode for
131 * Returns: A VFS inode, or NULL if none
134 struct inode *gfs2_ip2v_lookup(struct gfs2_inode *ip)
136 struct inode *inode = NULL;
138 gfs2_assert_warn(ip->i_sbd, test_bit(GIF_MIN_INIT, &ip->i_flags));
140 spin_lock(&ip->i_spin);
142 inode = igrab(ip->i_vnode);
143 spin_unlock(&ip->i_spin);
149 * gfs2_ip2v - Get/Create a struct inode for a struct gfs2_inode
150 * @ip: the struct gfs2_inode to get the struct inode for
152 * Returns: A VFS inode, or NULL if no mem
155 struct inode *gfs2_ip2v(struct gfs2_inode *ip)
157 struct inode *inode, *tmp;
159 inode = gfs2_ip2v_lookup(ip);
163 tmp = new_inode(ip->i_sbd->sd_vfs);
167 inode_attr_in(ip, tmp);
169 if (S_ISREG(ip->i_di.di_mode)) {
170 tmp->i_op = &gfs2_file_iops;
171 tmp->i_fop = &gfs2_file_fops;
172 tmp->i_mapping->a_ops = &gfs2_file_aops;
173 } else if (S_ISDIR(ip->i_di.di_mode)) {
174 tmp->i_op = &gfs2_dir_iops;
175 tmp->i_fop = &gfs2_dir_fops;
176 } else if (S_ISLNK(ip->i_di.di_mode)) {
177 tmp->i_op = &gfs2_symlink_iops;
179 tmp->i_op = &gfs2_dev_iops;
180 init_special_inode(tmp, tmp->i_mode, tmp->i_rdev);
183 tmp->u.generic_ip = NULL;
186 spin_lock(&ip->i_spin);
189 inode = igrab(ip->i_vnode);
190 spin_unlock(&ip->i_spin);
203 inode->u.generic_ip = ip;
205 spin_unlock(&ip->i_spin);
207 insert_inode_hash(inode);
212 static int iget_test(struct inode *inode, void *opaque)
214 struct gfs2_inode *ip = inode->u.generic_ip;
215 struct gfs2_inum *inum = (struct gfs2_inum *)opaque;
217 if (ip && ip->i_num.no_addr == inum->no_addr)
223 struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum *inum)
225 return ilookup5(sb, (unsigned long)inum->no_formal_ino,
229 void gfs2_inode_min_init(struct gfs2_inode *ip, unsigned int type)
231 if (!test_and_set_bit(GIF_MIN_INIT, &ip->i_flags)) {
232 ip->i_di.di_nlink = 1;
233 ip->i_di.di_mode = DT2IF(type);
238 * gfs2_inode_refresh - Refresh the incore copy of the dinode
239 * @ip: The GFS2 inode
244 int gfs2_inode_refresh(struct gfs2_inode *ip)
246 struct buffer_head *dibh;
249 error = gfs2_meta_inode_buffer(ip, &dibh);
253 if (gfs2_metatype_check(ip->i_sbd, dibh, GFS2_METATYPE_DI)) {
258 gfs2_dinode_in(&ip->i_di, dibh->b_data);
259 set_bit(GIF_MIN_INIT, &ip->i_flags);
263 if (ip->i_num.no_addr != ip->i_di.di_num.no_addr) {
264 if (gfs2_consist_inode(ip))
265 gfs2_dinode_print(&ip->i_di);
268 if (ip->i_num.no_formal_ino != ip->i_di.di_num.no_formal_ino)
271 ip->i_vn = ip->i_gl->gl_vn;
277 * inode_create - create a struct gfs2_inode
278 * @i_gl: The glock covering the inode
279 * @inum: The inode number
280 * @io_gl: the iopen glock to acquire/hold (using holder in new gfs2_inode)
281 * @io_state: the state the iopen glock should be acquired in
282 * @ipp: pointer to put the returned inode in
287 static int inode_create(struct gfs2_glock *i_gl, const struct gfs2_inum *inum,
288 struct gfs2_glock *io_gl, unsigned int io_state,
289 struct gfs2_inode **ipp, int need_lock)
291 struct gfs2_sbd *sdp = i_gl->gl_sbd;
292 struct gfs2_inode *ip;
295 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
298 memset(ip, 0, sizeof(struct gfs2_inode));
300 atomic_set(&ip->i_count, 1);
301 ip->i_vn = i_gl->gl_vn - 1;
304 spin_lock_init(&ip->i_spin);
305 init_rwsem(&ip->i_rw_mutex);
306 ip->i_greedy = gfs2_tune_get(sdp, gt_greedy_default);
309 error = gfs2_glock_nq_init(io_gl,
310 io_state, GL_LOCAL_EXCL | GL_EXACT,
315 spin_lock(&io_gl->gl_spin);
316 gfs2_glock_hold(i_gl);
317 io_gl->gl_object = i_gl;
318 spin_unlock(&io_gl->gl_spin);
321 gfs2_glock_hold(i_gl);
322 i_gl->gl_object = ip;
323 atomic_inc(&sdp->sd_inode_count);
328 gfs2_meta_cache_flush(ip);
329 kmem_cache_free(gfs2_inode_cachep, ip);
335 * gfs2_inode_get - Create or get a reference on an inode
336 * @i_gl: The glock covering the inode
337 * @inum: The inode number
339 * @ipp: pointer to put the returned inode in
344 int gfs2_inode_get(struct gfs2_glock *i_gl, const struct gfs2_inum *inum,
345 int create, struct gfs2_inode **ipp)
347 struct gfs2_sbd *sdp = i_gl->gl_sbd;
348 struct gfs2_glock *io_gl;
351 gfs2_glmutex_lock(i_gl);
353 *ipp = i_gl->gl_object;
356 if ((*ipp)->i_num.no_formal_ino != inum->no_formal_ino)
358 atomic_inc(&(*ipp)->i_count);
366 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops,
369 error = inode_create(i_gl, inum, io_gl, LM_ST_SHARED, ipp, 1);
370 gfs2_glock_put(io_gl);
374 gfs2_glmutex_unlock(i_gl);
379 void gfs2_inode_hold(struct gfs2_inode *ip)
381 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
382 atomic_inc(&ip->i_count);
385 void gfs2_inode_put(struct gfs2_inode *ip)
387 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
388 atomic_dec(&ip->i_count);
391 void gfs2_inode_destroy(struct gfs2_inode *ip, int unlock)
393 struct gfs2_sbd *sdp = ip->i_sbd;
394 struct gfs2_glock *i_gl = ip->i_gl;
396 gfs2_assert_warn(sdp, !atomic_read(&ip->i_count));
398 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl;
399 gfs2_assert(sdp, io_gl->gl_object == i_gl);
401 spin_lock(&io_gl->gl_spin);
402 io_gl->gl_object = NULL;
403 spin_unlock(&io_gl->gl_spin);
404 gfs2_glock_put(i_gl);
406 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
409 gfs2_meta_cache_flush(ip);
410 kmem_cache_free(gfs2_inode_cachep, ip);
412 i_gl->gl_object = NULL;
413 gfs2_glock_put(i_gl);
415 atomic_dec(&sdp->sd_inode_count);
418 static int dinode_dealloc(struct gfs2_inode *ip, struct gfs2_unlinked *ul)
420 struct gfs2_sbd *sdp = ip->i_sbd;
421 struct gfs2_alloc *al;
422 struct gfs2_rgrpd *rgd;
425 if (ip->i_di.di_blocks != 1) {
426 if (gfs2_consist_inode(ip))
427 gfs2_dinode_print(&ip->i_di);
431 al = gfs2_alloc_get(ip);
433 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
437 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
441 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
443 gfs2_consist_inode(ip);
445 goto out_rindex_relse;
448 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
451 goto out_rindex_relse;
453 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
454 RES_STATFS + RES_QUOTA, 1);
458 gfs2_trans_add_gl(ip->i_gl);
460 gfs2_free_di(rgd, ip);
462 error = gfs2_unlinked_ondisk_rm(sdp, ul);
465 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags);
468 gfs2_glock_dq_uninit(&al->al_rgd_gh);
471 gfs2_glock_dq_uninit(&al->al_ri_gh);
474 gfs2_quota_unhold(ip);
483 * inode_dealloc - Deallocate all on-disk blocks for an inode (dinode)
484 * @sdp: the filesystem
485 * @inum: the inode number to deallocate
486 * @io_gh: a holder for the iopen glock for this inode
488 * N.B. When we enter this we already hold the iopen glock and getting
489 * the glock for the inode means that we are grabbing the locks in the
490 * "wrong" order so we must only so a try lock operation and fail if we
491 * don't get the lock. Thats ok, since if we fail it means someone else
492 * is using the inode still and thus we shouldn't be deallocating it
498 static int inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul,
499 struct gfs2_holder *io_gh)
501 struct gfs2_inode *ip;
502 struct gfs2_holder i_gh;
505 error = gfs2_glock_nq_num(sdp, ul->ul_ut.ut_inum.no_addr,
506 &gfs2_inode_glops, LM_ST_EXCLUSIVE,
507 LM_FLAG_TRY_1CB|GL_DUMP, &i_gh);
512 return 1; /* or back off and relock in different order? */
517 gfs2_assert_warn(sdp, !i_gh.gh_gl->gl_object);
518 error = inode_create(i_gh.gh_gl, &ul->ul_ut.ut_inum, io_gh->gh_gl,
519 LM_ST_EXCLUSIVE, &ip, 0);
524 error = gfs2_inode_refresh(ip);
528 if (ip->i_di.di_nlink) {
529 if (gfs2_consist_inode(ip))
530 gfs2_dinode_print(&ip->i_di);
535 if (S_ISDIR(ip->i_di.di_mode) &&
536 (ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
537 error = gfs2_dir_exhash_dealloc(ip);
542 if (ip->i_di.di_eattr) {
543 error = gfs2_ea_dealloc(ip);
548 if (!gfs2_is_stuffed(ip)) {
549 error = gfs2_file_dealloc(ip);
554 error = dinode_dealloc(ip, ul);
559 gfs2_glmutex_lock(i_gh.gh_gl);
561 gfs2_inode_destroy(ip, 0);
562 gfs2_glmutex_unlock(i_gh.gh_gl);
565 gfs2_glock_dq_uninit(&i_gh);
571 * try_inode_dealloc - Try to deallocate an inode and all its blocks
572 * @sdp: the filesystem
574 * Returns: 0 on success, -errno on error, 1 on busy (inode open)
577 static int try_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
580 struct gfs2_holder iogh;
582 gfs2_try_toss_inode(sdp, &ul->ul_ut.ut_inum);
583 error = gfs2_glock_nq_num(sdp, ul->ul_ut.ut_inum.no_addr,
584 &gfs2_iopen_glops, LM_ST_EXCLUSIVE,
585 LM_FLAG_TRY_1CB, &iogh);
595 error = inode_dealloc(sdp, ul, &iogh);
596 gfs2_glock_dq_uninit(&iogh);
601 static int inode_dealloc_uninit(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
603 struct gfs2_rgrpd *rgd;
604 struct gfs2_holder ri_gh, rgd_gh;
607 error = gfs2_rindex_hold(sdp, &ri_gh);
611 rgd = gfs2_blk2rgrpd(sdp, ul->ul_ut.ut_inum.no_addr);
618 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rgd_gh);
622 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED + RES_STATFS, 0);
626 gfs2_free_uninit_di(rgd, ul->ul_ut.ut_inum.no_addr);
627 gfs2_unlinked_ondisk_rm(sdp, ul);
632 gfs2_glock_dq_uninit(&rgd_gh);
634 gfs2_glock_dq_uninit(&ri_gh);
639 int gfs2_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
641 if (ul->ul_ut.ut_flags & GFS2_UTF_UNINIT)
642 return inode_dealloc_uninit(sdp, ul);
644 return try_inode_dealloc(sdp, ul);
648 * gfs2_change_nlink - Change nlink count on inode
649 * @ip: The GFS2 inode
650 * @diff: The change in the nlink count required
655 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
657 struct buffer_head *dibh;
661 nlink = ip->i_di.di_nlink + diff;
663 /* If we are reducing the nlink count, but the new value ends up being
664 bigger than the old one, we must have underflowed. */
665 if (diff < 0 && nlink > ip->i_di.di_nlink) {
666 if (gfs2_consist_inode(ip))
667 gfs2_dinode_print(&ip->i_di);
671 error = gfs2_meta_inode_buffer(ip, &dibh);
675 ip->i_di.di_nlink = nlink;
676 ip->i_di.di_ctime = get_seconds();
678 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
679 gfs2_dinode_out(&ip->i_di, dibh->b_data);
685 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
688 gfs2_str2qstr(&qstr, name);
689 return gfs2_lookupi(dip, &qstr, 1, NULL);
694 * gfs2_lookupi - Look up a filename in a directory and return its inode
695 * @d_gh: An initialized holder for the directory glock
696 * @name: The name of the inode to look for
697 * @is_root: If 1, ignore the caller's permissions
698 * @i_gh: An uninitialized holder for the new inode glock
700 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless
706 struct inode *gfs2_lookupi(struct inode *dir, struct qstr *name, int is_root,
707 struct nameidata *nd)
710 struct super_block *sb = dir->i_sb;
711 struct gfs2_inode *ipp;
712 struct gfs2_inode *dip = dir->u.generic_ip;
713 struct gfs2_sbd *sdp = dip->i_sbd;
714 struct gfs2_holder d_gh;
715 struct gfs2_inum inum;
717 struct gfs2_glock *gl;
719 struct inode *inode = NULL;
721 if (!name->len || name->len > GFS2_FNAMESIZE)
722 return ERR_PTR(-ENAMETOOLONG);
724 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
725 (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
726 dir == sb->s_root->d_inode)) {
731 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
733 return ERR_PTR(error);
736 error = gfs2_repermission(dir, MAY_EXEC, NULL);
741 error = gfs2_dir_search(dir, name, &inum, &type);
745 error = gfs2_glock_get(sdp, inum.no_addr, &gfs2_inode_glops,
750 error = gfs2_inode_get(gl, &inum, CREATE, &ipp);
752 gfs2_inode_min_init(ipp, type);
757 gfs2_glock_dq_uninit(&d_gh);
758 if (error == -ENOENT)
761 inode = gfs2_ip2v(ipp);
764 return ERR_PTR(-ENOMEM);
767 return ERR_PTR(error);
770 static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino)
772 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
773 struct buffer_head *bh;
774 struct gfs2_inum_range ir;
777 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
780 mutex_lock(&sdp->sd_inum_mutex);
782 error = gfs2_meta_inode_buffer(ip, &bh);
784 mutex_unlock(&sdp->sd_inum_mutex);
789 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
792 *formal_ino = ir.ir_start++;
794 gfs2_trans_add_bh(ip->i_gl, bh, 1);
795 gfs2_inum_range_out(&ir,
796 bh->b_data + sizeof(struct gfs2_dinode));
798 mutex_unlock(&sdp->sd_inum_mutex);
805 mutex_unlock(&sdp->sd_inum_mutex);
811 static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino)
813 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
814 struct gfs2_inode *m_ip = sdp->sd_inum_inode->u.generic_ip;
815 struct gfs2_holder gh;
816 struct buffer_head *bh;
817 struct gfs2_inum_range ir;
820 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
824 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
827 mutex_lock(&sdp->sd_inum_mutex);
829 error = gfs2_meta_inode_buffer(ip, &bh);
833 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
836 struct buffer_head *m_bh;
839 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
843 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode));
844 x = y = be64_to_cpu(x);
846 ir.ir_length = GFS2_INUM_QUANTUM;
847 x += GFS2_INUM_QUANTUM;
849 gfs2_consist_inode(m_ip);
851 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
852 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x;
857 *formal_ino = ir.ir_start++;
860 gfs2_trans_add_bh(ip->i_gl, bh, 1);
861 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
867 mutex_unlock(&sdp->sd_inum_mutex);
871 gfs2_glock_dq_uninit(&gh);
876 static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum)
880 error = pick_formal_ino_1(sdp, inum);
884 error = pick_formal_ino_2(sdp, inum);
890 * create_ok - OK to create a new on-disk inode here?
891 * @dip: Directory in which dinode is to be created
892 * @name: Name of new dinode
898 static int create_ok(struct gfs2_inode *dip, struct qstr *name,
903 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
907 /* Don't create entries in an unlinked directory */
908 if (!dip->i_di.di_nlink)
911 error = gfs2_dir_search(dip->i_vnode, name, NULL, NULL);
922 if (dip->i_di.di_entries == (uint32_t)-1)
924 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1)
930 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
931 unsigned int *uid, unsigned int *gid)
933 if (dip->i_sbd->sd_args.ar_suiddir &&
934 (dip->i_di.di_mode & S_ISUID) &&
938 else if (dip->i_di.di_uid != current->fsuid)
940 *uid = dip->i_di.di_uid;
942 *uid = current->fsuid;
944 if (dip->i_di.di_mode & S_ISGID) {
947 *gid = dip->i_di.di_gid;
949 *gid = current->fsgid;
952 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_unlinked *ul)
954 struct gfs2_sbd *sdp = dip->i_sbd;
959 dip->i_alloc.al_requested = RES_DINODE;
960 error = gfs2_inplace_reserve(dip);
964 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
969 ul->ul_ut.ut_inum.no_addr = gfs2_alloc_di(dip);
971 ul->ul_ut.ut_flags = GFS2_UTF_UNINIT;
972 error = gfs2_unlinked_ondisk_add(sdp, ul);
977 gfs2_inplace_release(dip);
986 * init_dinode - Fill in a new dinode structure
987 * @dip: the directory this inode is being created in
988 * @gl: The glock covering the new inode
989 * @inum: the inode number
990 * @mode: the file permissions
996 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
997 struct gfs2_inum *inum, unsigned int mode,
998 unsigned int uid, unsigned int gid)
1000 struct gfs2_sbd *sdp = dip->i_sbd;
1001 struct gfs2_dinode *di;
1002 struct buffer_head *dibh;
1004 dibh = gfs2_meta_new(gl, inum->no_addr);
1005 gfs2_trans_add_bh(gl, dibh, 1);
1006 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
1007 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1008 di = (struct gfs2_dinode *)dibh->b_data;
1010 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
1011 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
1012 di->di_mode = cpu_to_be32(mode);
1013 di->di_uid = cpu_to_be32(uid);
1014 di->di_gid = cpu_to_be32(gid);
1015 di->di_nlink = cpu_to_be32(0);
1016 di->di_size = cpu_to_be64(0);
1017 di->di_blocks = cpu_to_be64(1);
1018 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
1019 di->di_major = di->di_minor = cpu_to_be32(0);
1020 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
1021 di->__pad[0] = di->__pad[1] = 0;
1022 di->di_flags = cpu_to_be32(0);
1024 if (S_ISREG(mode)) {
1025 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
1026 gfs2_tune_get(sdp, gt_new_files_jdata))
1027 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
1028 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
1029 gfs2_tune_get(sdp, gt_new_files_directio))
1030 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
1031 } else if (S_ISDIR(mode)) {
1032 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1033 GFS2_DIF_INHERIT_DIRECTIO);
1034 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1035 GFS2_DIF_INHERIT_JDATA);
1039 di->di_height = cpu_to_be32(0);
1042 di->di_depth = cpu_to_be16(0);
1043 di->di_entries = cpu_to_be32(0);
1044 memset(&di->__pad4, 0, sizeof(di->__pad4));
1045 di->di_eattr = cpu_to_be64(0);
1046 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
1051 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1052 unsigned int mode, struct gfs2_unlinked *ul)
1054 struct gfs2_sbd *sdp = dip->i_sbd;
1055 unsigned int uid, gid;
1058 munge_mode_uid_gid(dip, &mode, &uid, &gid);
1059 gfs2_alloc_get(dip);
1061 error = gfs2_quota_lock(dip, uid, gid);
1065 error = gfs2_quota_check(dip, uid, gid);
1069 error = gfs2_trans_begin(sdp, RES_DINODE + RES_UNLINKED + RES_QUOTA, 0);
1073 ul->ul_ut.ut_flags = 0;
1074 error = gfs2_unlinked_ondisk_munge(sdp, ul);
1075 init_dinode(dip, gl, &ul->ul_ut.ut_inum, mode, uid, gid);
1076 gfs2_quota_change(dip, +1, uid, gid);
1077 gfs2_trans_end(sdp);
1080 gfs2_quota_unlock(dip);
1083 gfs2_alloc_put(dip);
1087 static int link_dinode(struct gfs2_inode *dip, struct qstr *name,
1088 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1090 struct gfs2_sbd *sdp = dip->i_sbd;
1091 struct gfs2_alloc *al;
1093 struct buffer_head *dibh;
1096 al = gfs2_alloc_get(dip);
1098 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1102 error = alloc_required = gfs2_diradd_alloc_required(dip->i_vnode, name);
1103 if (alloc_required < 0)
1105 if (alloc_required) {
1106 error = gfs2_quota_check(dip, dip->i_di.di_uid,
1109 goto fail_quota_locks;
1111 al->al_requested = sdp->sd_max_dirres;
1113 error = gfs2_inplace_reserve(dip);
1115 goto fail_quota_locks;
1117 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1118 al->al_rgd->rd_ri.ri_length +
1119 2 * RES_DINODE + RES_UNLINKED +
1120 RES_STATFS + RES_QUOTA, 0);
1124 error = gfs2_trans_begin(sdp,
1129 goto fail_quota_locks;
1132 error = gfs2_dir_add(dip->i_vnode, name, &ip->i_num, IF2DT(ip->i_di.di_mode));
1134 goto fail_end_trans;
1136 error = gfs2_meta_inode_buffer(ip, &dibh);
1138 goto fail_end_trans;
1139 ip->i_di.di_nlink = 1;
1140 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1141 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1144 error = gfs2_unlinked_ondisk_rm(sdp, ul);
1146 goto fail_end_trans;
1151 gfs2_trans_end(sdp);
1154 if (dip->i_alloc.al_rgd)
1155 gfs2_inplace_release(dip);
1158 gfs2_quota_unlock(dip);
1161 gfs2_alloc_put(dip);
1166 * gfs2_createi - Create a new inode
1167 * @ghs: An array of two holders
1168 * @name: The name of the new file
1169 * @mode: the permissions on the new inode
1171 * @ghs[0] is an initialized holder for the directory
1172 * @ghs[1] is the holder for the inode lock
1174 * If the return value is not NULL, the glocks on both the directory and the new
1175 * file are held. A transaction has been started and an inplace reservation
1181 struct inode *gfs2_createi(struct gfs2_holder *ghs, struct qstr *name,
1184 struct inode *inode;
1185 struct gfs2_inode *dip = ghs->gh_gl->gl_object;
1186 struct gfs2_sbd *sdp = dip->i_sbd;
1187 struct gfs2_unlinked *ul;
1188 struct gfs2_inode *ip;
1191 if (!name->len || name->len > GFS2_FNAMESIZE)
1192 return ERR_PTR(-ENAMETOOLONG);
1194 error = gfs2_unlinked_get(sdp, &ul);
1196 return ERR_PTR(error);
1198 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1199 error = gfs2_glock_nq(ghs);
1203 error = create_ok(dip, name, mode);
1207 error = pick_formal_ino(sdp, &ul->ul_ut.ut_inum.no_formal_ino);
1211 error = alloc_dinode(dip, ul);
1215 if (ul->ul_ut.ut_inum.no_addr < dip->i_num.no_addr) {
1218 error = gfs2_glock_nq_num(sdp, ul->ul_ut.ut_inum.no_addr,
1219 &gfs2_inode_glops, LM_ST_EXCLUSIVE,
1222 gfs2_unlinked_put(sdp, ul);
1223 return ERR_PTR(error);
1226 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1227 error = gfs2_glock_nq(ghs);
1229 gfs2_glock_dq_uninit(ghs + 1);
1230 gfs2_unlinked_put(sdp, ul);
1231 return ERR_PTR(error);
1234 error = create_ok(dip, name, mode);
1238 error = gfs2_glock_nq_num(sdp, ul->ul_ut.ut_inum.no_addr,
1239 &gfs2_inode_glops, LM_ST_EXCLUSIVE,
1245 error = make_dinode(dip, ghs[1].gh_gl, mode, ul);
1249 error = gfs2_inode_get(ghs[1].gh_gl, &ul->ul_ut.ut_inum, CREATE, &ip);
1253 error = gfs2_inode_refresh(ip);
1257 error = gfs2_acl_create(dip, ip);
1261 error = link_dinode(dip, name, ip, ul);
1265 gfs2_unlinked_put(sdp, ul);
1267 inode = gfs2_ip2v(ip);
1270 return ERR_PTR(-ENOMEM);
1277 gfs2_glock_dq_uninit(ghs + 1);
1283 gfs2_unlinked_put(sdp, ul);
1284 return ERR_PTR(error);
1288 * gfs2_unlinki - Unlink a file
1289 * @dip: The inode of the directory
1290 * @name: The name of the file to be unlinked
1291 * @ip: The inode of the file to be removed
1293 * Assumes Glocks on both dip and ip are held.
1298 int gfs2_unlinki(struct gfs2_inode *dip, struct qstr *name,
1299 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1301 struct gfs2_sbd *sdp = dip->i_sbd;
1304 error = gfs2_dir_del(dip, name);
1308 error = gfs2_change_nlink(ip, -1);
1312 /* If this inode is being unlinked from the directory structure,
1313 we need to mark that in the log so that it isn't lost during
1316 if (!ip->i_di.di_nlink) {
1317 ul->ul_ut.ut_inum = ip->i_num;
1318 error = gfs2_unlinked_ondisk_add(sdp, ul);
1320 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1327 * gfs2_rmdiri - Remove a directory
1328 * @dip: The parent directory of the directory to be removed
1329 * @name: The name of the directory to be removed
1330 * @ip: The GFS2 inode of the directory to be removed
1332 * Assumes Glocks on dip and ip are held
1337 int gfs2_rmdiri(struct gfs2_inode *dip, struct qstr *name,
1338 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1340 struct gfs2_sbd *sdp = dip->i_sbd;
1341 struct qstr dotname;
1344 if (ip->i_di.di_entries != 2) {
1345 if (gfs2_consist_inode(ip))
1346 gfs2_dinode_print(&ip->i_di);
1350 error = gfs2_dir_del(dip, name);
1354 error = gfs2_change_nlink(dip, -1);
1358 gfs2_str2qstr(&dotname, ".");
1359 error = gfs2_dir_del(ip, &dotname);
1364 dotname.name = "..";
1365 dotname.hash = gfs2_disk_hash(dotname.name, dotname.len);
1366 error = gfs2_dir_del(ip, &dotname);
1370 error = gfs2_change_nlink(ip, -2);
1374 /* This inode is being unlinked from the directory structure and
1375 we need to mark that in the log so that it isn't lost during
1378 ul->ul_ut.ut_inum = ip->i_num;
1379 error = gfs2_unlinked_ondisk_add(sdp, ul);
1381 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1387 * gfs2_unlink_ok - check to see that a inode is still in a directory
1388 * @dip: the directory
1389 * @name: the name of the file
1392 * Assumes that the lock on (at least) @dip is held.
1394 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1397 int gfs2_unlink_ok(struct gfs2_inode *dip, struct qstr *name,
1398 struct gfs2_inode *ip)
1400 struct gfs2_inum inum;
1404 if (IS_IMMUTABLE(ip->i_vnode) || IS_APPEND(ip->i_vnode))
1407 if ((dip->i_di.di_mode & S_ISVTX) &&
1408 dip->i_di.di_uid != current->fsuid &&
1409 ip->i_di.di_uid != current->fsuid &&
1410 !capable(CAP_FOWNER))
1413 if (IS_APPEND(dip->i_vnode))
1416 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
1420 error = gfs2_dir_search(dip->i_vnode, name, &inum, &type);
1424 if (!gfs2_inum_equal(&inum, &ip->i_num))
1427 if (IF2DT(ip->i_di.di_mode) != type) {
1428 gfs2_consist_inode(dip);
1436 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1440 * Follow @to back to the root and make sure we don't encounter @this
1441 * Assumes we already hold the rename lock.
1446 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1448 struct inode *dir = to->i_vnode;
1449 struct super_block *sb = dir->i_sb;
1454 gfs2_str2qstr(&dotdot, "..");
1459 if (dir == this->i_vnode) {
1463 if (dir == sb->s_root->d_inode) {
1468 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL);
1470 error = PTR_ERR(tmp);
1484 * gfs2_readlinki - return the contents of a symlink
1485 * @ip: the symlink's inode
1486 * @buf: a pointer to the buffer to be filled
1487 * @len: a pointer to the length of @buf
1489 * If @buf is too small, a piece of memory is kmalloc()ed and needs
1490 * to be freed by the caller.
1495 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1497 struct gfs2_holder i_gh;
1498 struct buffer_head *dibh;
1502 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1503 error = gfs2_glock_nq_atime(&i_gh);
1505 gfs2_holder_uninit(&i_gh);
1509 if (!ip->i_di.di_size) {
1510 gfs2_consist_inode(ip);
1515 error = gfs2_meta_inode_buffer(ip, &dibh);
1519 x = ip->i_di.di_size + 1;
1521 *buf = kmalloc(x, GFP_KERNEL);
1528 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1535 gfs2_glock_dq_uninit(&i_gh);
1541 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1542 * conditionally update the inode's atime
1543 * @gh: the holder to acquire
1545 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1546 * Update if the difference between the current time and the inode's current
1547 * atime is greater than an interval specified at mount.
1552 int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1554 struct gfs2_glock *gl = gh->gh_gl;
1555 struct gfs2_sbd *sdp = gl->gl_sbd;
1556 struct gfs2_inode *ip = gl->gl_object;
1557 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1562 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1563 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1564 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1567 state = gh->gh_state;
1568 flags = gh->gh_flags;
1570 error = gfs2_glock_nq(gh);
1574 if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1575 (sdp->sd_vfs->s_flags & MS_RDONLY))
1578 curtime = get_seconds();
1579 if (curtime - ip->i_di.di_atime >= quantum) {
1581 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY,
1583 error = gfs2_glock_nq(gh);
1587 /* Verify that atime hasn't been updated while we were
1588 trying to get exclusive lock. */
1590 curtime = get_seconds();
1591 if (curtime - ip->i_di.di_atime >= quantum) {
1592 struct buffer_head *dibh;
1594 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1595 if (error == -EROFS)
1600 error = gfs2_meta_inode_buffer(ip, &dibh);
1602 goto fail_end_trans;
1604 ip->i_di.di_atime = curtime;
1606 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1607 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1610 gfs2_trans_end(sdp);
1613 /* If someone else has asked for the glock,
1614 unlock and let them have it. Then reacquire
1615 in the original state. */
1616 if (gfs2_glock_is_blocking(gl)) {
1618 gfs2_holder_reinit(state, flags, gh);
1619 return gfs2_glock_nq(gh);
1626 gfs2_trans_end(sdp);
1635 * glock_compare_atime - Compare two struct gfs2_glock structures for sort
1636 * @arg_a: the first structure
1637 * @arg_b: the second structure
1639 * Returns: 1 if A > B
1644 static int glock_compare_atime(const void *arg_a, const void *arg_b)
1646 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1647 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1648 struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1649 struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1652 if (a->ln_number > b->ln_number)
1654 else if (a->ln_number < b->ln_number)
1657 if (gh_a->gh_state == LM_ST_SHARED &&
1658 gh_b->gh_state == LM_ST_EXCLUSIVE)
1660 else if (gh_a->gh_state == LM_ST_SHARED &&
1661 (gh_b->gh_flags & GL_ATIME))
1669 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an
1671 * @num_gh: the number of structures
1672 * @ghs: an array of struct gfs2_holder structures
1674 * Returns: 0 on success (all glocks acquired),
1675 * errno on failure (no glocks acquired)
1678 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs)
1680 struct gfs2_holder **p;
1688 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1689 if (ghs->gh_flags & GL_ATIME)
1690 error = gfs2_glock_nq_atime(ghs);
1692 error = gfs2_glock_nq(ghs);
1696 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1700 for (x = 0; x < num_gh; x++)
1703 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL);
1705 for (x = 0; x < num_gh; x++) {
1706 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1708 if (p[x]->gh_flags & GL_ATIME)
1709 error = gfs2_glock_nq_atime(p[x]);
1711 error = gfs2_glock_nq(p[x]);
1715 gfs2_glock_dq(p[x]);
1726 * gfs2_try_toss_vnode - See if we can toss a vnode from memory
1729 * Returns: 1 if the vnode was tossed
1732 void gfs2_try_toss_vnode(struct gfs2_inode *ip)
1734 struct inode *inode;
1736 inode = gfs2_ip2v_lookup(ip);
1740 d_prune_aliases(inode);
1742 if (S_ISDIR(ip->i_di.di_mode)) {
1743 struct list_head *head = &inode->i_dentry;
1744 struct dentry *d = NULL;
1746 spin_lock(&dcache_lock);
1747 if (list_empty(head))
1748 spin_unlock(&dcache_lock);
1750 d = list_entry(head->next, struct dentry, d_alias);
1752 spin_unlock(&dcache_lock);
1754 if (have_submounts(d))
1757 shrink_dcache_parent(d);
1759 d_prune_aliases(inode);
1770 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1772 struct buffer_head *dibh;
1775 error = gfs2_meta_inode_buffer(ip, &dibh);
1777 error = inode_setattr(ip->i_vnode, attr);
1778 gfs2_assert_warn(ip->i_sbd, !error);
1779 gfs2_inode_attr_out(ip);
1781 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1782 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1789 * gfs2_setattr_simple -
1793 * Called with a reference on the vnode.
1798 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1802 if (current->journal_info)
1803 return __gfs2_setattr_simple(ip, attr);
1805 error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0);
1809 error = __gfs2_setattr_simple(ip, attr);
1811 gfs2_trans_end(ip->i_sbd);
1816 int gfs2_repermission(struct inode *inode, int mask, struct nameidata *nd)
1818 return permission(inode, mask, nd);