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 <asm/semaphore.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)
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));
302 atomic_set(&ip->i_count, 1);
304 ip->i_vn = i_gl->gl_vn - 1;
309 spin_lock_init(&ip->i_spin);
310 init_rwsem(&ip->i_rw_mutex);
312 ip->i_greedy = gfs2_tune_get(sdp, gt_greedy_default);
314 error = gfs2_glock_nq_init(io_gl,
315 io_state, GL_LOCAL_EXCL | GL_EXACT,
319 ip->i_iopen_gh.gh_owner = NULL;
321 spin_lock(&io_gl->gl_spin);
322 gfs2_glock_hold(i_gl);
323 io_gl->gl_object = i_gl;
324 spin_unlock(&io_gl->gl_spin);
326 gfs2_glock_hold(i_gl);
327 i_gl->gl_object = ip;
329 atomic_inc(&sdp->sd_inode_count);
336 gfs2_meta_cache_flush(ip);
337 kmem_cache_free(gfs2_inode_cachep, ip);
344 * gfs2_inode_get - Create or get a reference on an inode
345 * @i_gl: The glock covering the inode
346 * @inum: The inode number
348 * @ipp: pointer to put the returned inode in
353 int gfs2_inode_get(struct gfs2_glock *i_gl, const struct gfs2_inum *inum,
354 int create, struct gfs2_inode **ipp)
356 struct gfs2_sbd *sdp = i_gl->gl_sbd;
357 struct gfs2_glock *io_gl;
360 gfs2_glmutex_lock(i_gl);
362 *ipp = i_gl->gl_object;
365 if ((*ipp)->i_num.no_formal_ino != inum->no_formal_ino)
367 atomic_inc(&(*ipp)->i_count);
375 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops,
378 error = inode_create(i_gl, inum, io_gl, LM_ST_SHARED, ipp);
379 gfs2_glock_put(io_gl);
383 gfs2_glmutex_unlock(i_gl);
388 void gfs2_inode_hold(struct gfs2_inode *ip)
390 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
391 atomic_inc(&ip->i_count);
394 void gfs2_inode_put(struct gfs2_inode *ip)
396 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
397 atomic_dec(&ip->i_count);
400 void gfs2_inode_destroy(struct gfs2_inode *ip)
402 struct gfs2_sbd *sdp = ip->i_sbd;
403 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl;
404 struct gfs2_glock *i_gl = ip->i_gl;
406 gfs2_assert_warn(sdp, !atomic_read(&ip->i_count));
407 gfs2_assert(sdp, io_gl->gl_object == i_gl);
409 spin_lock(&io_gl->gl_spin);
410 io_gl->gl_object = NULL;
411 gfs2_glock_put(i_gl);
412 spin_unlock(&io_gl->gl_spin);
414 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
416 gfs2_meta_cache_flush(ip);
417 kmem_cache_free(gfs2_inode_cachep, ip);
419 i_gl->gl_object = NULL;
420 gfs2_glock_put(i_gl);
422 atomic_dec(&sdp->sd_inode_count);
425 static int dinode_dealloc(struct gfs2_inode *ip, struct gfs2_unlinked *ul)
427 struct gfs2_sbd *sdp = ip->i_sbd;
428 struct gfs2_alloc *al;
429 struct gfs2_rgrpd *rgd;
432 if (ip->i_di.di_blocks != 1) {
433 if (gfs2_consist_inode(ip))
434 gfs2_dinode_print(&ip->i_di);
438 al = gfs2_alloc_get(ip);
440 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
444 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
448 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
450 gfs2_consist_inode(ip);
452 goto out_rindex_relse;
455 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
458 goto out_rindex_relse;
460 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
461 RES_STATFS + RES_QUOTA, 1);
465 gfs2_trans_add_gl(ip->i_gl);
467 gfs2_free_di(rgd, ip);
469 error = gfs2_unlinked_ondisk_rm(sdp, ul);
472 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags);
475 gfs2_glock_dq_uninit(&al->al_rgd_gh);
478 gfs2_glock_dq_uninit(&al->al_ri_gh);
481 gfs2_quota_unhold(ip);
490 * inode_dealloc - Deallocate all on-disk blocks for an inode (dinode)
491 * @sdp: the filesystem
492 * @inum: the inode number to deallocate
493 * @io_gh: a holder for the iopen glock for this inode
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,
506 ul->ul_ut.ut_inum.no_addr, &gfs2_inode_glops,
507 LM_ST_EXCLUSIVE, 0, &i_gh);
511 /* We reacquire the iopen lock here to avoid a race with the NFS server
512 calling gfs2_read_inode() with the inode number of a inode we're in
513 the process of deallocating. And we can't keep our hold on the lock
514 from inode_dealloc_init() for deadlock reasons. */
516 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY, io_gh);
517 error = gfs2_glock_nq(io_gh);
527 gfs2_assert_warn(sdp, !i_gh.gh_gl->gl_object);
528 error = inode_create(i_gh.gh_gl, &ul->ul_ut.ut_inum, io_gh->gh_gl,
529 LM_ST_EXCLUSIVE, &ip);
531 gfs2_glock_dq(io_gh);
536 error = gfs2_inode_refresh(ip);
540 if (ip->i_di.di_nlink) {
541 if (gfs2_consist_inode(ip))
542 gfs2_dinode_print(&ip->i_di);
547 if (S_ISDIR(ip->i_di.di_mode) &&
548 (ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
549 error = gfs2_dir_exhash_dealloc(ip);
554 if (ip->i_di.di_eattr) {
555 error = gfs2_ea_dealloc(ip);
560 if (!gfs2_is_stuffed(ip)) {
561 error = gfs2_file_dealloc(ip);
566 error = dinode_dealloc(ip, ul);
571 gfs2_glmutex_lock(i_gh.gh_gl);
573 gfs2_inode_destroy(ip);
574 gfs2_glmutex_unlock(i_gh.gh_gl);
577 gfs2_glock_dq_uninit(&i_gh);
583 * try_inode_dealloc - Try to deallocate an inode and all its blocks
584 * @sdp: the filesystem
586 * Returns: 0 on success, -errno on error, 1 on busy (inode open)
589 static int try_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
591 struct gfs2_holder io_gh;
594 gfs2_try_toss_inode(sdp, &ul->ul_ut.ut_inum);
596 error = gfs2_glock_nq_num(sdp,
597 ul->ul_ut.ut_inum.no_addr, &gfs2_iopen_glops,
598 LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB, &io_gh);
608 gfs2_glock_dq(&io_gh);
609 error = inode_dealloc(sdp, ul, &io_gh);
610 gfs2_holder_uninit(&io_gh);
615 static int inode_dealloc_uninit(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
617 struct gfs2_rgrpd *rgd;
618 struct gfs2_holder ri_gh, rgd_gh;
621 error = gfs2_rindex_hold(sdp, &ri_gh);
625 rgd = gfs2_blk2rgrpd(sdp, ul->ul_ut.ut_inum.no_addr);
632 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rgd_gh);
636 error = gfs2_trans_begin(sdp,
637 RES_RG_BIT + RES_UNLINKED + RES_STATFS,
642 gfs2_free_uninit_di(rgd, ul->ul_ut.ut_inum.no_addr);
643 gfs2_unlinked_ondisk_rm(sdp, ul);
648 gfs2_glock_dq_uninit(&rgd_gh);
650 gfs2_glock_dq_uninit(&ri_gh);
655 int gfs2_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
657 if (ul->ul_ut.ut_flags & GFS2_UTF_UNINIT)
658 return inode_dealloc_uninit(sdp, ul);
660 return try_inode_dealloc(sdp, ul);
664 * gfs2_change_nlink - Change nlink count on inode
665 * @ip: The GFS2 inode
666 * @diff: The change in the nlink count required
671 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
673 struct buffer_head *dibh;
677 nlink = ip->i_di.di_nlink + diff;
679 /* If we are reducing the nlink count, but the new value ends up being
680 bigger than the old one, we must have underflowed. */
681 if (diff < 0 && nlink > ip->i_di.di_nlink) {
682 if (gfs2_consist_inode(ip))
683 gfs2_dinode_print(&ip->i_di);
687 error = gfs2_meta_inode_buffer(ip, &dibh);
691 ip->i_di.di_nlink = nlink;
692 ip->i_di.di_ctime = get_seconds();
694 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
695 gfs2_dinode_out(&ip->i_di, dibh->b_data);
701 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
705 qstr.len = strlen(name);
706 qstr.hash = gfs2_disk_hash(qstr.name, qstr.len);
707 return gfs2_lookupi(dip, &qstr, 1, NULL);
712 * gfs2_lookupi - Look up a filename in a directory and return its inode
713 * @d_gh: An initialized holder for the directory glock
714 * @name: The name of the inode to look for
715 * @is_root: If 1, ignore the caller's permissions
716 * @i_gh: An uninitialized holder for the new inode glock
718 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless
724 struct inode *gfs2_lookupi(struct inode *dir, struct qstr *name, int is_root,
725 struct nameidata *nd)
728 struct super_block *sb = dir->i_sb;
729 struct gfs2_inode *ipp;
730 struct gfs2_inode *dip = dir->u.generic_ip;
731 struct gfs2_sbd *sdp = dip->i_sbd;
732 struct gfs2_holder d_gh;
733 struct gfs2_inum inum;
735 struct gfs2_glock *gl;
737 struct inode *inode = NULL;
739 if (!name->len || name->len > GFS2_FNAMESIZE)
740 return ERR_PTR(-ENAMETOOLONG);
742 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
743 (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
744 dir == sb->s_root->d_inode)) {
745 gfs2_inode_hold(dip);
750 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
752 return ERR_PTR(error);
755 error = gfs2_repermission(dip->i_vnode, MAY_EXEC, NULL);
760 error = gfs2_dir_search(dir, name, &inum, &type);
764 error = gfs2_glock_get(sdp, inum.no_addr, &gfs2_inode_glops,
769 error = gfs2_inode_get(gl, &inum, CREATE, &ipp);
771 gfs2_inode_min_init(ipp, type);
776 gfs2_glock_dq_uninit(&d_gh);
778 if (error == -ENOENT)
781 inode = gfs2_ip2v(ipp);
784 return ERR_PTR(-ENOMEM);
787 return ERR_PTR(error);
790 static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino)
792 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
793 struct buffer_head *bh;
794 struct gfs2_inum_range ir;
797 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
800 mutex_lock(&sdp->sd_inum_mutex);
802 error = gfs2_meta_inode_buffer(ip, &bh);
804 mutex_unlock(&sdp->sd_inum_mutex);
809 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
812 *formal_ino = ir.ir_start++;
814 gfs2_trans_add_bh(ip->i_gl, bh, 1);
815 gfs2_inum_range_out(&ir,
816 bh->b_data + sizeof(struct gfs2_dinode));
818 mutex_unlock(&sdp->sd_inum_mutex);
825 mutex_unlock(&sdp->sd_inum_mutex);
831 static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino)
833 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
834 struct gfs2_inode *m_ip = sdp->sd_inum_inode->u.generic_ip;
835 struct gfs2_holder gh;
836 struct buffer_head *bh;
837 struct gfs2_inum_range ir;
840 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
844 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
847 mutex_lock(&sdp->sd_inum_mutex);
849 error = gfs2_meta_inode_buffer(ip, &bh);
853 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
856 struct buffer_head *m_bh;
859 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
863 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode));
864 x = y = be64_to_cpu(x);
866 ir.ir_length = GFS2_INUM_QUANTUM;
867 x += GFS2_INUM_QUANTUM;
869 gfs2_consist_inode(m_ip);
871 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
872 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x;
877 *formal_ino = ir.ir_start++;
880 gfs2_trans_add_bh(ip->i_gl, bh, 1);
881 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
887 mutex_unlock(&sdp->sd_inum_mutex);
891 gfs2_glock_dq_uninit(&gh);
896 static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum)
900 error = pick_formal_ino_1(sdp, inum);
904 error = pick_formal_ino_2(sdp, inum);
910 * create_ok - OK to create a new on-disk inode here?
911 * @dip: Directory in which dinode is to be created
912 * @name: Name of new dinode
918 static int create_ok(struct gfs2_inode *dip, struct qstr *name,
923 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
927 /* Don't create entries in an unlinked directory */
928 if (!dip->i_di.di_nlink)
931 error = gfs2_dir_search(dip->i_vnode, name, NULL, NULL);
942 if (dip->i_di.di_entries == (uint32_t)-1)
944 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1)
950 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
951 unsigned int *uid, unsigned int *gid)
953 if (dip->i_sbd->sd_args.ar_suiddir &&
954 (dip->i_di.di_mode & S_ISUID) &&
958 else if (dip->i_di.di_uid != current->fsuid)
960 *uid = dip->i_di.di_uid;
962 *uid = current->fsuid;
964 if (dip->i_di.di_mode & S_ISGID) {
967 *gid = dip->i_di.di_gid;
969 *gid = current->fsgid;
972 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_unlinked *ul)
974 struct gfs2_sbd *sdp = dip->i_sbd;
979 dip->i_alloc.al_requested = RES_DINODE;
980 error = gfs2_inplace_reserve(dip);
984 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
989 ul->ul_ut.ut_inum.no_addr = gfs2_alloc_di(dip);
991 ul->ul_ut.ut_flags = GFS2_UTF_UNINIT;
992 error = gfs2_unlinked_ondisk_add(sdp, ul);
997 gfs2_inplace_release(dip);
1000 gfs2_alloc_put(dip);
1006 * init_dinode - Fill in a new dinode structure
1007 * @dip: the directory this inode is being created in
1008 * @gl: The glock covering the new inode
1009 * @inum: the inode number
1010 * @mode: the file permissions
1016 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1017 struct gfs2_inum *inum, unsigned int mode,
1018 unsigned int uid, unsigned int gid)
1020 struct gfs2_sbd *sdp = dip->i_sbd;
1021 struct gfs2_dinode *di;
1022 struct buffer_head *dibh;
1024 dibh = gfs2_meta_new(gl, inum->no_addr);
1025 gfs2_trans_add_bh(gl, dibh, 1);
1026 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
1027 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1028 di = (struct gfs2_dinode *)dibh->b_data;
1030 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
1031 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
1032 di->di_mode = cpu_to_be32(mode);
1033 di->di_uid = cpu_to_be32(uid);
1034 di->di_gid = cpu_to_be32(gid);
1035 di->di_nlink = cpu_to_be32(0);
1036 di->di_size = cpu_to_be64(0);
1037 di->di_blocks = cpu_to_be64(1);
1038 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
1039 di->di_major = di->di_minor = cpu_to_be32(0);
1040 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
1041 di->__pad[0] = di->__pad[1] = 0;
1042 di->di_flags = cpu_to_be32(0);
1044 if (S_ISREG(mode)) {
1045 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
1046 gfs2_tune_get(sdp, gt_new_files_jdata))
1047 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
1048 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
1049 gfs2_tune_get(sdp, gt_new_files_directio))
1050 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
1051 } else if (S_ISDIR(mode)) {
1052 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1053 GFS2_DIF_INHERIT_DIRECTIO);
1054 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1055 GFS2_DIF_INHERIT_JDATA);
1059 di->di_height = cpu_to_be32(0);
1062 di->di_depth = cpu_to_be16(0);
1063 di->di_entries = cpu_to_be32(0);
1064 memset(&di->__pad4, 0, sizeof(di->__pad4));
1065 di->di_eattr = cpu_to_be64(0);
1066 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
1071 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1072 unsigned int mode, struct gfs2_unlinked *ul)
1074 struct gfs2_sbd *sdp = dip->i_sbd;
1075 unsigned int uid, gid;
1078 munge_mode_uid_gid(dip, &mode, &uid, &gid);
1080 gfs2_alloc_get(dip);
1082 error = gfs2_quota_lock(dip, uid, gid);
1086 error = gfs2_quota_check(dip, uid, gid);
1090 error = gfs2_trans_begin(sdp, RES_DINODE + RES_UNLINKED +
1095 ul->ul_ut.ut_flags = 0;
1096 error = gfs2_unlinked_ondisk_munge(sdp, ul);
1098 init_dinode(dip, gl, &ul->ul_ut.ut_inum,
1101 gfs2_quota_change(dip, +1, uid, gid);
1103 gfs2_trans_end(sdp);
1106 gfs2_quota_unlock(dip);
1109 gfs2_alloc_put(dip);
1114 static int link_dinode(struct gfs2_inode *dip, struct qstr *name,
1115 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1117 struct gfs2_sbd *sdp = dip->i_sbd;
1118 struct gfs2_alloc *al;
1120 struct buffer_head *dibh;
1123 al = gfs2_alloc_get(dip);
1125 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1129 error = alloc_required = gfs2_diradd_alloc_required(dip->i_vnode, name);
1130 if (alloc_required < 0)
1132 if (alloc_required) {
1133 error = gfs2_quota_check(dip, dip->i_di.di_uid,
1136 goto fail_quota_locks;
1138 al->al_requested = sdp->sd_max_dirres;
1140 error = gfs2_inplace_reserve(dip);
1142 goto fail_quota_locks;
1144 error = gfs2_trans_begin(sdp,
1145 sdp->sd_max_dirres +
1146 al->al_rgd->rd_ri.ri_length +
1147 2 * RES_DINODE + RES_UNLINKED +
1148 RES_STATFS + RES_QUOTA, 0);
1152 error = gfs2_trans_begin(sdp,
1157 goto fail_quota_locks;
1160 error = gfs2_dir_add(dip->i_vnode, name, &ip->i_num, IF2DT(ip->i_di.di_mode));
1162 goto fail_end_trans;
1164 error = gfs2_meta_inode_buffer(ip, &dibh);
1166 goto fail_end_trans;
1167 ip->i_di.di_nlink = 1;
1168 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1169 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1172 error = gfs2_unlinked_ondisk_rm(sdp, ul);
1174 goto fail_end_trans;
1179 gfs2_trans_end(sdp);
1182 if (dip->i_alloc.al_rgd)
1183 gfs2_inplace_release(dip);
1186 gfs2_quota_unlock(dip);
1189 gfs2_alloc_put(dip);
1195 * gfs2_createi - Create a new inode
1196 * @ghs: An array of two holders
1197 * @name: The name of the new file
1198 * @mode: the permissions on the new inode
1200 * @ghs[0] is an initialized holder for the directory
1201 * @ghs[1] is the holder for the inode lock
1203 * If the return value is not NULL, the glocks on both the directory and the new
1204 * file are held. A transaction has been started and an inplace reservation
1210 struct inode *gfs2_createi(struct gfs2_holder *ghs, struct qstr *name,
1213 struct inode *inode;
1214 struct gfs2_inode *dip = ghs->gh_gl->gl_object;
1215 struct gfs2_sbd *sdp = dip->i_sbd;
1216 struct gfs2_unlinked *ul;
1217 struct gfs2_inode *ip;
1220 if (!name->len || name->len > GFS2_FNAMESIZE)
1221 return ERR_PTR(-ENAMETOOLONG);
1223 error = gfs2_unlinked_get(sdp, &ul);
1225 return ERR_PTR(error);
1227 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1228 error = gfs2_glock_nq(ghs);
1232 error = create_ok(dip, name, mode);
1236 error = pick_formal_ino(sdp, &ul->ul_ut.ut_inum.no_formal_ino);
1240 error = alloc_dinode(dip, ul);
1244 if (ul->ul_ut.ut_inum.no_addr < dip->i_num.no_addr) {
1247 error = gfs2_glock_nq_num(sdp,
1248 ul->ul_ut.ut_inum.no_addr,
1250 LM_ST_EXCLUSIVE, GL_SKIP,
1253 gfs2_unlinked_put(sdp, ul);
1254 return ERR_PTR(error);
1257 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1258 error = gfs2_glock_nq(ghs);
1260 gfs2_glock_dq_uninit(ghs + 1);
1261 gfs2_unlinked_put(sdp, ul);
1262 return ERR_PTR(error);
1265 error = create_ok(dip, name, mode);
1269 error = gfs2_glock_nq_num(sdp,
1270 ul->ul_ut.ut_inum.no_addr,
1272 LM_ST_EXCLUSIVE, GL_SKIP,
1278 error = make_dinode(dip, ghs[1].gh_gl, mode, ul);
1282 error = gfs2_inode_get(ghs[1].gh_gl, &ul->ul_ut.ut_inum, CREATE, &ip);
1286 error = gfs2_inode_refresh(ip);
1290 error = gfs2_acl_create(dip, ip);
1294 error = link_dinode(dip, name, ip, ul);
1298 gfs2_unlinked_put(sdp, ul);
1300 inode = gfs2_ip2v(ip);
1303 return ERR_PTR(-ENOMEM);
1310 gfs2_glock_dq_uninit(ghs + 1);
1316 gfs2_unlinked_put(sdp, ul);
1318 return ERR_PTR(error);
1322 * gfs2_unlinki - Unlink a file
1323 * @dip: The inode of the directory
1324 * @name: The name of the file to be unlinked
1325 * @ip: The inode of the file to be removed
1327 * Assumes Glocks on both dip and ip are held.
1332 int gfs2_unlinki(struct gfs2_inode *dip, struct qstr *name,
1333 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1335 struct gfs2_sbd *sdp = dip->i_sbd;
1338 error = gfs2_dir_del(dip, name);
1342 error = gfs2_change_nlink(ip, -1);
1346 /* If this inode is being unlinked from the directory structure,
1347 we need to mark that in the log so that it isn't lost during
1350 if (!ip->i_di.di_nlink) {
1351 ul->ul_ut.ut_inum = ip->i_num;
1352 error = gfs2_unlinked_ondisk_add(sdp, ul);
1354 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1361 * gfs2_rmdiri - Remove a directory
1362 * @dip: The parent directory of the directory to be removed
1363 * @name: The name of the directory to be removed
1364 * @ip: The GFS2 inode of the directory to be removed
1366 * Assumes Glocks on dip and ip are held
1371 int gfs2_rmdiri(struct gfs2_inode *dip, struct qstr *name,
1372 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1374 struct gfs2_sbd *sdp = dip->i_sbd;
1375 struct qstr dotname;
1378 if (ip->i_di.di_entries != 2) {
1379 if (gfs2_consist_inode(ip))
1380 gfs2_dinode_print(&ip->i_di);
1384 error = gfs2_dir_del(dip, name);
1388 error = gfs2_change_nlink(dip, -1);
1394 dotname.hash = gfs2_disk_hash(dotname.name, dotname.len);
1395 error = gfs2_dir_del(ip, &dotname);
1400 dotname.name = "..";
1401 dotname.hash = gfs2_disk_hash(dotname.name, dotname.len);
1402 error = gfs2_dir_del(ip, &dotname);
1406 error = gfs2_change_nlink(ip, -2);
1410 /* This inode is being unlinked from the directory structure and
1411 we need to mark that in the log so that it isn't lost during
1414 ul->ul_ut.ut_inum = ip->i_num;
1415 error = gfs2_unlinked_ondisk_add(sdp, ul);
1417 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1423 * gfs2_unlink_ok - check to see that a inode is still in a directory
1424 * @dip: the directory
1425 * @name: the name of the file
1428 * Assumes that the lock on (at least) @dip is held.
1430 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1433 int gfs2_unlink_ok(struct gfs2_inode *dip, struct qstr *name,
1434 struct gfs2_inode *ip)
1436 struct gfs2_inum inum;
1440 if (IS_IMMUTABLE(ip->i_vnode) || IS_APPEND(ip->i_vnode))
1443 if ((dip->i_di.di_mode & S_ISVTX) &&
1444 dip->i_di.di_uid != current->fsuid &&
1445 ip->i_di.di_uid != current->fsuid &&
1446 !capable(CAP_FOWNER))
1449 if (IS_APPEND(dip->i_vnode))
1452 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
1456 error = gfs2_dir_search(dip->i_vnode, name, &inum, &type);
1460 if (!gfs2_inum_equal(&inum, &ip->i_num))
1463 if (IF2DT(ip->i_di.di_mode) != type) {
1464 gfs2_consist_inode(dip);
1472 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1476 * Follow @to back to the root and make sure we don't encounter @this
1477 * Assumes we already hold the rename lock.
1482 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1484 struct inode *dir = to->i_vnode;
1485 struct super_block *sb = dir->i_sb;
1490 memset(&dotdot, 0, sizeof(struct qstr));
1493 dotdot.hash = gfs2_disk_hash(dotdot.name, dotdot.len);
1498 if (dir == this->i_vnode) {
1502 if (dir == sb->s_root->d_inode) {
1507 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL);
1509 error = PTR_ERR(tmp);
1523 * gfs2_readlinki - return the contents of a symlink
1524 * @ip: the symlink's inode
1525 * @buf: a pointer to the buffer to be filled
1526 * @len: a pointer to the length of @buf
1528 * If @buf is too small, a piece of memory is kmalloc()ed and needs
1529 * to be freed by the caller.
1534 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1536 struct gfs2_holder i_gh;
1537 struct buffer_head *dibh;
1541 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1542 error = gfs2_glock_nq_atime(&i_gh);
1544 gfs2_holder_uninit(&i_gh);
1548 if (!ip->i_di.di_size) {
1549 gfs2_consist_inode(ip);
1554 error = gfs2_meta_inode_buffer(ip, &dibh);
1558 x = ip->i_di.di_size + 1;
1560 *buf = kmalloc(x, GFP_KERNEL);
1567 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1574 gfs2_glock_dq_uninit(&i_gh);
1580 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1581 * conditionally update the inode's atime
1582 * @gh: the holder to acquire
1584 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1585 * Update if the difference between the current time and the inode's current
1586 * atime is greater than an interval specified at mount.
1591 int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1593 struct gfs2_glock *gl = gh->gh_gl;
1594 struct gfs2_sbd *sdp = gl->gl_sbd;
1595 struct gfs2_inode *ip = gl->gl_object;
1596 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1601 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1602 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1603 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1606 state = gh->gh_state;
1607 flags = gh->gh_flags;
1609 error = gfs2_glock_nq(gh);
1613 if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1614 (sdp->sd_vfs->s_flags & MS_RDONLY))
1617 curtime = get_seconds();
1618 if (curtime - ip->i_di.di_atime >= quantum) {
1620 gfs2_holder_reinit(LM_ST_EXCLUSIVE,
1621 gh->gh_flags & ~LM_FLAG_ANY,
1623 error = gfs2_glock_nq(gh);
1627 /* Verify that atime hasn't been updated while we were
1628 trying to get exclusive lock. */
1630 curtime = get_seconds();
1631 if (curtime - ip->i_di.di_atime >= quantum) {
1632 struct buffer_head *dibh;
1634 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1635 if (error == -EROFS)
1640 error = gfs2_meta_inode_buffer(ip, &dibh);
1642 goto fail_end_trans;
1644 ip->i_di.di_atime = curtime;
1646 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1647 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1650 gfs2_trans_end(sdp);
1653 /* If someone else has asked for the glock,
1654 unlock and let them have it. Then reacquire
1655 in the original state. */
1656 if (gfs2_glock_is_blocking(gl)) {
1658 gfs2_holder_reinit(state, flags, gh);
1659 return gfs2_glock_nq(gh);
1666 gfs2_trans_end(sdp);
1675 * glock_compare_atime - Compare two struct gfs2_glock structures for sort
1676 * @arg_a: the first structure
1677 * @arg_b: the second structure
1679 * Returns: 1 if A > B
1684 static int glock_compare_atime(const void *arg_a, const void *arg_b)
1686 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1687 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1688 struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1689 struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1692 if (a->ln_number > b->ln_number)
1694 else if (a->ln_number < b->ln_number)
1697 if (gh_a->gh_state == LM_ST_SHARED &&
1698 gh_b->gh_state == LM_ST_EXCLUSIVE)
1700 else if (gh_a->gh_state == LM_ST_SHARED &&
1701 (gh_b->gh_flags & GL_ATIME))
1709 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an
1711 * @num_gh: the number of structures
1712 * @ghs: an array of struct gfs2_holder structures
1714 * Returns: 0 on success (all glocks acquired),
1715 * errno on failure (no glocks acquired)
1718 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs)
1720 struct gfs2_holder **p;
1728 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1729 if (ghs->gh_flags & GL_ATIME)
1730 error = gfs2_glock_nq_atime(ghs);
1732 error = gfs2_glock_nq(ghs);
1736 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1740 for (x = 0; x < num_gh; x++)
1743 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL);
1745 for (x = 0; x < num_gh; x++) {
1746 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1748 if (p[x]->gh_flags & GL_ATIME)
1749 error = gfs2_glock_nq_atime(p[x]);
1751 error = gfs2_glock_nq(p[x]);
1755 gfs2_glock_dq(p[x]);
1766 * gfs2_try_toss_vnode - See if we can toss a vnode from memory
1769 * Returns: 1 if the vnode was tossed
1772 void gfs2_try_toss_vnode(struct gfs2_inode *ip)
1774 struct inode *inode;
1776 inode = gfs2_ip2v_lookup(ip);
1780 d_prune_aliases(inode);
1782 if (S_ISDIR(ip->i_di.di_mode)) {
1783 struct list_head *head = &inode->i_dentry;
1784 struct dentry *d = NULL;
1786 spin_lock(&dcache_lock);
1787 if (list_empty(head))
1788 spin_unlock(&dcache_lock);
1790 d = list_entry(head->next, struct dentry, d_alias);
1792 spin_unlock(&dcache_lock);
1794 if (have_submounts(d))
1797 shrink_dcache_parent(d);
1799 d_prune_aliases(inode);
1810 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1812 struct buffer_head *dibh;
1815 error = gfs2_meta_inode_buffer(ip, &dibh);
1817 error = inode_setattr(ip->i_vnode, attr);
1818 gfs2_assert_warn(ip->i_sbd, !error);
1819 gfs2_inode_attr_out(ip);
1821 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1822 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1829 * gfs2_setattr_simple -
1833 * Called with a reference on the vnode.
1838 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1842 if (current->journal_info)
1843 return __gfs2_setattr_simple(ip, attr);
1845 error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0);
1849 error = __gfs2_setattr_simple(ip, attr);
1851 gfs2_trans_end(ip->i_sbd);
1856 int gfs2_repermission(struct inode *inode, int mask, struct nameidata *nd)
1858 return permission(inode, mask, nd);