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 <asm/semaphore.h>
29 #include "ops_address.h"
31 #include "ops_inode.h"
38 * inode_attr_in - Copy attributes from the dinode into the VFS inode
39 * @ip: The GFS2 inode (with embedded disk inode data)
40 * @inode: The Linux VFS inode
44 static void inode_attr_in(struct gfs2_inode *ip, struct inode *inode)
46 inode->i_ino = ip->i_num.no_formal_ino;
48 switch (ip->i_di.di_mode & S_IFMT) {
51 inode->i_rdev = MKDEV(ip->i_di.di_major, ip->i_di.di_minor);
58 inode->i_mode = ip->i_di.di_mode;
59 inode->i_nlink = ip->i_di.di_nlink;
60 inode->i_uid = ip->i_di.di_uid;
61 inode->i_gid = ip->i_di.di_gid;
62 i_size_write(inode, ip->i_di.di_size);
63 inode->i_atime.tv_sec = ip->i_di.di_atime;
64 inode->i_mtime.tv_sec = ip->i_di.di_mtime;
65 inode->i_ctime.tv_sec = ip->i_di.di_ctime;
66 inode->i_atime.tv_nsec = 0;
67 inode->i_mtime.tv_nsec = 0;
68 inode->i_ctime.tv_nsec = 0;
69 inode->i_blksize = PAGE_SIZE;
70 inode->i_blocks = ip->i_di.di_blocks <<
71 (ip->i_sbd->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT);
73 if (ip->i_di.di_flags & GFS2_DIF_IMMUTABLE)
74 inode->i_flags |= S_IMMUTABLE;
76 inode->i_flags &= ~S_IMMUTABLE;
78 if (ip->i_di.di_flags & GFS2_DIF_APPENDONLY)
79 inode->i_flags |= S_APPEND;
81 inode->i_flags &= ~S_APPEND;
85 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode
86 * @ip: The GFS2 inode (with embedded disk inode data)
90 void gfs2_inode_attr_in(struct gfs2_inode *ip)
94 inode = gfs2_ip2v_lookup(ip);
96 inode_attr_in(ip, inode);
102 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode
103 * @ip: The GFS2 inode
105 * Only copy out the attributes that we want the VFS layer
106 * to be able to modify.
109 void gfs2_inode_attr_out(struct gfs2_inode *ip)
111 struct inode *inode = ip->i_vnode;
113 gfs2_assert_withdraw(ip->i_sbd,
114 (ip->i_di.di_mode & S_IFMT) == (inode->i_mode & S_IFMT));
115 ip->i_di.di_mode = inode->i_mode;
116 ip->i_di.di_uid = inode->i_uid;
117 ip->i_di.di_gid = inode->i_gid;
118 ip->i_di.di_atime = inode->i_atime.tv_sec;
119 ip->i_di.di_mtime = inode->i_mtime.tv_sec;
120 ip->i_di.di_ctime = inode->i_ctime.tv_sec;
124 * gfs2_ip2v_lookup - Get the struct inode for a struct gfs2_inode
125 * @ip: the struct gfs2_inode to get the struct inode for
127 * Returns: A VFS inode, or NULL if none
130 struct inode *gfs2_ip2v_lookup(struct gfs2_inode *ip)
132 struct inode *inode = NULL;
134 gfs2_assert_warn(ip->i_sbd, test_bit(GIF_MIN_INIT, &ip->i_flags));
136 spin_lock(&ip->i_spin);
138 inode = igrab(ip->i_vnode);
139 spin_unlock(&ip->i_spin);
145 * gfs2_ip2v - Get/Create a struct inode for a struct gfs2_inode
146 * @ip: the struct gfs2_inode to get the struct inode for
148 * Returns: A VFS inode, or NULL if no mem
151 struct inode *gfs2_ip2v(struct gfs2_inode *ip)
153 struct inode *inode, *tmp;
155 inode = gfs2_ip2v_lookup(ip);
159 tmp = new_inode(ip->i_sbd->sd_vfs);
163 inode_attr_in(ip, tmp);
165 if (S_ISREG(ip->i_di.di_mode)) {
166 tmp->i_op = &gfs2_file_iops;
167 tmp->i_fop = &gfs2_file_fops;
168 tmp->i_mapping->a_ops = &gfs2_file_aops;
169 } else if (S_ISDIR(ip->i_di.di_mode)) {
170 tmp->i_op = &gfs2_dir_iops;
171 tmp->i_fop = &gfs2_dir_fops;
172 } else if (S_ISLNK(ip->i_di.di_mode)) {
173 tmp->i_op = &gfs2_symlink_iops;
175 tmp->i_op = &gfs2_dev_iops;
176 init_special_inode(tmp, tmp->i_mode, tmp->i_rdev);
182 spin_lock(&ip->i_spin);
185 inode = igrab(ip->i_vnode);
186 spin_unlock(&ip->i_spin);
201 spin_unlock(&ip->i_spin);
203 insert_inode_hash(inode);
208 static int iget_test(struct inode *inode, void *opaque)
210 struct gfs2_inode *ip = get_v2ip(inode);
211 struct gfs2_inum *inum = (struct gfs2_inum *)opaque;
213 if (ip && ip->i_num.no_addr == inum->no_addr)
219 struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum *inum)
221 return ilookup5(sb, (unsigned long)inum->no_formal_ino,
225 void gfs2_inode_min_init(struct gfs2_inode *ip, unsigned int type)
227 spin_lock(&ip->i_spin);
228 if (!test_and_set_bit(GIF_MIN_INIT, &ip->i_flags)) {
229 ip->i_di.di_nlink = 1;
230 ip->i_di.di_mode = DT2IF(type);
232 spin_unlock(&ip->i_spin);
236 * gfs2_inode_refresh - Refresh the incore copy of the dinode
237 * @ip: The GFS2 inode
242 int gfs2_inode_refresh(struct gfs2_inode *ip)
244 struct buffer_head *dibh;
247 error = gfs2_meta_inode_buffer(ip, &dibh);
251 if (gfs2_metatype_check(ip->i_sbd, dibh, GFS2_METATYPE_DI)) {
256 spin_lock(&ip->i_spin);
257 gfs2_dinode_in(&ip->i_di, dibh->b_data);
258 set_bit(GIF_MIN_INIT, &ip->i_flags);
259 spin_unlock(&ip->i_spin);
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, 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 set_gl2gl(io_gl, i_gl);
324 spin_unlock(&io_gl->gl_spin);
326 gfs2_glock_hold(i_gl);
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, struct gfs2_inum *inum, int create,
354 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 = get_gl2ip(i_gl);
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, get_gl2gl(io_gl) == i_gl);
409 spin_lock(&io_gl->gl_spin);
410 set_gl2gl(io_gl, 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 set_gl2ip(i_gl, 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, !get_gl2ip(i_gh.gh_gl));
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);
702 * gfs2_lookupi - Look up a filename in a directory and return its inode
703 * @d_gh: An initialized holder for the directory glock
704 * @name: The name of the inode to look for
705 * @is_root: If 1, ignore the caller's permissions
706 * @i_gh: An uninitialized holder for the new inode glock
708 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless
714 int gfs2_lookupi(struct inode *dir, struct qstr *name, int is_root,
715 struct inode **inodep)
717 struct gfs2_inode *ipp;
718 struct gfs2_inode *dip = get_v2ip(dir);
719 struct gfs2_sbd *sdp = dip->i_sbd;
720 struct gfs2_holder d_gh;
721 struct gfs2_inum inum;
723 struct gfs2_glock *gl;
728 if (!name->len || name->len > GFS2_FNAMESIZE)
729 return -ENAMETOOLONG;
731 if (gfs2_filecmp(name, ".", 1) ||
732 (gfs2_filecmp(name, "..", 2) && dir == sdp->sd_root_dir)) {
733 gfs2_inode_hold(dip);
738 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
743 error = gfs2_repermission(dip->i_vnode, MAY_EXEC, NULL);
748 error = gfs2_dir_search(dip, name, &inum, &type);
752 error = gfs2_glock_get(sdp, inum.no_addr, &gfs2_inode_glops,
757 error = gfs2_inode_get(gl, &inum, CREATE, &ipp);
759 gfs2_inode_min_init(ipp, type);
764 gfs2_glock_dq_uninit(&d_gh);
767 *inodep = gfs2_ip2v(ipp);
775 static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino)
777 struct gfs2_inode *ip = get_v2ip(sdp->sd_ir_inode);
778 struct buffer_head *bh;
779 struct gfs2_inum_range ir;
782 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
785 mutex_lock(&sdp->sd_inum_mutex);
787 error = gfs2_meta_inode_buffer(ip, &bh);
789 mutex_unlock(&sdp->sd_inum_mutex);
794 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
797 *formal_ino = ir.ir_start++;
799 gfs2_trans_add_bh(ip->i_gl, bh, 1);
800 gfs2_inum_range_out(&ir,
801 bh->b_data + sizeof(struct gfs2_dinode));
803 mutex_unlock(&sdp->sd_inum_mutex);
810 mutex_unlock(&sdp->sd_inum_mutex);
816 static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino)
818 struct gfs2_inode *ip = get_v2ip(sdp->sd_ir_inode);
819 struct gfs2_inode *m_ip = get_v2ip(sdp->sd_inum_inode);
820 struct gfs2_holder gh;
821 struct buffer_head *bh;
822 struct gfs2_inum_range ir;
825 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
829 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
832 mutex_lock(&sdp->sd_inum_mutex);
834 error = gfs2_meta_inode_buffer(ip, &bh);
838 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
841 struct buffer_head *m_bh;
844 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
848 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode));
849 x = y = be64_to_cpu(x);
851 ir.ir_length = GFS2_INUM_QUANTUM;
852 x += GFS2_INUM_QUANTUM;
854 gfs2_consist_inode(m_ip);
856 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
857 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x;
862 *formal_ino = ir.ir_start++;
865 gfs2_trans_add_bh(ip->i_gl, bh, 1);
866 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
872 mutex_unlock(&sdp->sd_inum_mutex);
876 gfs2_glock_dq_uninit(&gh);
881 static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum)
885 error = pick_formal_ino_1(sdp, inum);
889 error = pick_formal_ino_2(sdp, inum);
895 * create_ok - OK to create a new on-disk inode here?
896 * @dip: Directory in which dinode is to be created
897 * @name: Name of new dinode
903 static int create_ok(struct gfs2_inode *dip, struct qstr *name,
908 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
912 /* Don't create entries in an unlinked directory */
913 if (!dip->i_di.di_nlink)
916 error = gfs2_dir_search(dip, name, NULL, NULL);
927 if (dip->i_di.di_entries == (uint32_t)-1)
929 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1)
935 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
936 unsigned int *uid, unsigned int *gid)
938 if (dip->i_sbd->sd_args.ar_suiddir &&
939 (dip->i_di.di_mode & S_ISUID) &&
943 else if (dip->i_di.di_uid != current->fsuid)
945 *uid = dip->i_di.di_uid;
947 *uid = current->fsuid;
949 if (dip->i_di.di_mode & S_ISGID) {
952 *gid = dip->i_di.di_gid;
954 *gid = current->fsgid;
957 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_unlinked *ul)
959 struct gfs2_sbd *sdp = dip->i_sbd;
964 dip->i_alloc.al_requested = RES_DINODE;
965 error = gfs2_inplace_reserve(dip);
969 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
974 ul->ul_ut.ut_inum.no_addr = gfs2_alloc_di(dip);
976 ul->ul_ut.ut_flags = GFS2_UTF_UNINIT;
977 error = gfs2_unlinked_ondisk_add(sdp, ul);
982 gfs2_inplace_release(dip);
991 * init_dinode - Fill in a new dinode structure
992 * @dip: the directory this inode is being created in
993 * @gl: The glock covering the new inode
994 * @inum: the inode number
995 * @mode: the file permissions
1001 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1002 struct gfs2_inum *inum, unsigned int mode,
1003 unsigned int uid, unsigned int gid)
1005 struct gfs2_sbd *sdp = dip->i_sbd;
1006 struct gfs2_dinode *di;
1007 struct buffer_head *dibh;
1009 dibh = gfs2_meta_new(gl, inum->no_addr);
1010 gfs2_trans_add_bh(gl, dibh, 1);
1011 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
1012 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1013 di = (struct gfs2_dinode *)dibh->b_data;
1015 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
1016 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
1017 di->di_mode = cpu_to_be32(mode);
1018 di->di_uid = cpu_to_be32(uid);
1019 di->di_gid = cpu_to_be32(gid);
1020 di->di_nlink = cpu_to_be32(0);
1021 di->di_size = cpu_to_be64(0);
1022 di->di_blocks = cpu_to_be64(1);
1023 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
1024 di->di_major = di->di_minor = cpu_to_be32(0);
1025 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
1026 di->__pad[0] = di->__pad[1] = 0;
1027 di->di_flags = cpu_to_be32(0);
1029 if (S_ISREG(mode)) {
1030 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
1031 gfs2_tune_get(sdp, gt_new_files_jdata))
1032 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
1033 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
1034 gfs2_tune_get(sdp, gt_new_files_directio))
1035 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
1036 } else if (S_ISDIR(mode)) {
1037 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO);
1038 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 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 = gfs2_diradd_alloc_required(dip, name, &alloc_required);
1113 if (alloc_required) {
1114 error = gfs2_quota_check(dip, dip->i_di.di_uid,
1117 goto fail_quota_locks;
1119 al->al_requested = sdp->sd_max_dirres;
1121 error = gfs2_inplace_reserve(dip);
1123 goto fail_quota_locks;
1125 error = gfs2_trans_begin(sdp,
1126 sdp->sd_max_dirres +
1127 al->al_rgd->rd_ri.ri_length +
1128 2 * RES_DINODE + RES_UNLINKED +
1129 RES_STATFS + RES_QUOTA, 0);
1133 error = gfs2_trans_begin(sdp,
1138 goto fail_quota_locks;
1141 error = gfs2_dir_add(dip, name, &ip->i_num, IF2DT(ip->i_di.di_mode));
1143 goto fail_end_trans;
1145 error = gfs2_meta_inode_buffer(ip, &dibh);
1147 goto fail_end_trans;
1148 ip->i_di.di_nlink = 1;
1149 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1150 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1153 error = gfs2_unlinked_ondisk_rm(sdp, ul);
1155 goto fail_end_trans;
1160 gfs2_trans_end(sdp);
1163 if (dip->i_alloc.al_rgd)
1164 gfs2_inplace_release(dip);
1167 gfs2_quota_unlock(dip);
1170 gfs2_alloc_put(dip);
1176 * gfs2_createi - Create a new inode
1177 * @ghs: An array of two holders
1178 * @name: The name of the new file
1179 * @mode: the permissions on the new inode
1181 * @ghs[0] is an initialized holder for the directory
1182 * @ghs[1] is the holder for the inode lock
1184 * If the return value is not NULL, the glocks on both the directory and the new
1185 * file are held. A transaction has been started and an inplace reservation
1191 struct inode *gfs2_createi(struct gfs2_holder *ghs, struct qstr *name, unsigned int mode)
1193 struct inode *inode;
1194 struct gfs2_inode *dip = get_gl2ip(ghs->gh_gl);
1195 struct gfs2_sbd *sdp = dip->i_sbd;
1196 struct gfs2_unlinked *ul;
1197 struct gfs2_inode *ip;
1200 if (!name->len || name->len > GFS2_FNAMESIZE)
1201 return ERR_PTR(-ENAMETOOLONG);
1203 error = gfs2_unlinked_get(sdp, &ul);
1205 return ERR_PTR(error);
1207 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1208 error = gfs2_glock_nq(ghs);
1212 error = create_ok(dip, name, mode);
1216 error = pick_formal_ino(sdp, &ul->ul_ut.ut_inum.no_formal_ino);
1220 error = alloc_dinode(dip, ul);
1224 if (ul->ul_ut.ut_inum.no_addr < dip->i_num.no_addr) {
1227 error = gfs2_glock_nq_num(sdp,
1228 ul->ul_ut.ut_inum.no_addr,
1230 LM_ST_EXCLUSIVE, GL_SKIP,
1233 gfs2_unlinked_put(sdp, ul);
1234 return ERR_PTR(error);
1237 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1238 error = gfs2_glock_nq(ghs);
1240 gfs2_glock_dq_uninit(ghs + 1);
1241 gfs2_unlinked_put(sdp, ul);
1242 return ERR_PTR(error);
1245 error = create_ok(dip, name, mode);
1249 error = gfs2_glock_nq_num(sdp,
1250 ul->ul_ut.ut_inum.no_addr,
1252 LM_ST_EXCLUSIVE, GL_SKIP,
1258 error = make_dinode(dip, ghs[1].gh_gl, mode, ul);
1262 error = gfs2_inode_get(ghs[1].gh_gl, &ul->ul_ut.ut_inum, CREATE, &ip);
1266 error = gfs2_inode_refresh(ip);
1270 error = gfs2_acl_create(dip, ip);
1274 error = link_dinode(dip, name, ip, ul);
1278 gfs2_unlinked_put(sdp, ul);
1280 inode = gfs2_ip2v(ip);
1283 return ERR_PTR(-ENOMEM);
1290 gfs2_glock_dq_uninit(ghs + 1);
1296 gfs2_unlinked_put(sdp, ul);
1298 return ERR_PTR(error);
1302 * gfs2_unlinki - Unlink a file
1303 * @dip: The inode of the directory
1304 * @name: The name of the file to be unlinked
1305 * @ip: The inode of the file to be removed
1307 * Assumes Glocks on both dip and ip are held.
1312 int gfs2_unlinki(struct gfs2_inode *dip, struct qstr *name,
1313 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1315 struct gfs2_sbd *sdp = dip->i_sbd;
1318 error = gfs2_dir_del(dip, name);
1322 error = gfs2_change_nlink(ip, -1);
1326 /* If this inode is being unlinked from the directory structure,
1327 we need to mark that in the log so that it isn't lost during
1330 if (!ip->i_di.di_nlink) {
1331 ul->ul_ut.ut_inum = ip->i_num;
1332 error = gfs2_unlinked_ondisk_add(sdp, ul);
1334 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1341 * gfs2_rmdiri - Remove a directory
1342 * @dip: The parent directory of the directory to be removed
1343 * @name: The name of the directory to be removed
1344 * @ip: The GFS2 inode of the directory to be removed
1346 * Assumes Glocks on dip and ip are held
1351 int gfs2_rmdiri(struct gfs2_inode *dip, struct qstr *name,
1352 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1354 struct gfs2_sbd *sdp = dip->i_sbd;
1355 struct qstr dotname;
1358 if (ip->i_di.di_entries != 2) {
1359 if (gfs2_consist_inode(ip))
1360 gfs2_dinode_print(&ip->i_di);
1364 error = gfs2_dir_del(dip, name);
1368 error = gfs2_change_nlink(dip, -1);
1374 error = gfs2_dir_del(ip, &dotname);
1379 dotname.name = "..";
1380 error = gfs2_dir_del(ip, &dotname);
1384 error = gfs2_change_nlink(ip, -2);
1388 /* This inode is being unlinked from the directory structure and
1389 we need to mark that in the log so that it isn't lost during
1392 ul->ul_ut.ut_inum = ip->i_num;
1393 error = gfs2_unlinked_ondisk_add(sdp, ul);
1395 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1401 * gfs2_unlink_ok - check to see that a inode is still in a directory
1402 * @dip: the directory
1403 * @name: the name of the file
1406 * Assumes that the lock on (at least) @dip is held.
1408 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1411 int gfs2_unlink_ok(struct gfs2_inode *dip, struct qstr *name,
1412 struct gfs2_inode *ip)
1414 struct gfs2_inum inum;
1418 if (IS_IMMUTABLE(ip->i_vnode) || IS_APPEND(ip->i_vnode))
1421 if ((dip->i_di.di_mode & S_ISVTX) &&
1422 dip->i_di.di_uid != current->fsuid &&
1423 ip->i_di.di_uid != current->fsuid &&
1424 !capable(CAP_FOWNER))
1427 if (IS_APPEND(dip->i_vnode))
1430 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
1434 error = gfs2_dir_search(dip, name, &inum, &type);
1438 if (!gfs2_inum_equal(&inum, &ip->i_num))
1441 if (IF2DT(ip->i_di.di_mode) != type) {
1442 gfs2_consist_inode(dip);
1450 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1454 * Follow @to back to the root and make sure we don't encounter @this
1455 * Assumes we already hold the rename lock.
1460 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1462 struct gfs2_sbd *sdp = this->i_sbd;
1463 struct inode *dir = to->i_vnode;
1468 memset(&dotdot, 0, sizeof(struct qstr));
1475 if (dir == this->i_vnode) {
1479 if (dir == sdp->sd_root_dir) {
1484 error = gfs2_lookupi(dir, &dotdot, 1, &tmp);
1498 * gfs2_readlinki - return the contents of a symlink
1499 * @ip: the symlink's inode
1500 * @buf: a pointer to the buffer to be filled
1501 * @len: a pointer to the length of @buf
1503 * If @buf is too small, a piece of memory is kmalloc()ed and needs
1504 * to be freed by the caller.
1509 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1511 struct gfs2_holder i_gh;
1512 struct buffer_head *dibh;
1516 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1517 error = gfs2_glock_nq_atime(&i_gh);
1519 gfs2_holder_uninit(&i_gh);
1523 if (!ip->i_di.di_size) {
1524 gfs2_consist_inode(ip);
1529 error = gfs2_meta_inode_buffer(ip, &dibh);
1533 x = ip->i_di.di_size + 1;
1535 *buf = kmalloc(x, GFP_KERNEL);
1542 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1549 gfs2_glock_dq_uninit(&i_gh);
1555 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1556 * conditionally update the inode's atime
1557 * @gh: the holder to acquire
1559 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1560 * Update if the difference between the current time and the inode's current
1561 * atime is greater than an interval specified at mount.
1566 int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1568 struct gfs2_glock *gl = gh->gh_gl;
1569 struct gfs2_sbd *sdp = gl->gl_sbd;
1570 struct gfs2_inode *ip = get_gl2ip(gl);
1571 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1576 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1577 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1578 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1581 state = gh->gh_state;
1582 flags = gh->gh_flags;
1584 error = gfs2_glock_nq(gh);
1588 if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1589 (sdp->sd_vfs->s_flags & MS_RDONLY))
1592 curtime = get_seconds();
1593 if (curtime - ip->i_di.di_atime >= quantum) {
1595 gfs2_holder_reinit(LM_ST_EXCLUSIVE,
1596 gh->gh_flags & ~LM_FLAG_ANY,
1598 error = gfs2_glock_nq(gh);
1602 /* Verify that atime hasn't been updated while we were
1603 trying to get exclusive lock. */
1605 curtime = get_seconds();
1606 if (curtime - ip->i_di.di_atime >= quantum) {
1607 struct buffer_head *dibh;
1609 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1610 if (error == -EROFS)
1615 error = gfs2_meta_inode_buffer(ip, &dibh);
1617 goto fail_end_trans;
1619 ip->i_di.di_atime = curtime;
1621 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1622 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1625 gfs2_trans_end(sdp);
1628 /* If someone else has asked for the glock,
1629 unlock and let them have it. Then reacquire
1630 in the original state. */
1631 if (gfs2_glock_is_blocking(gl)) {
1633 gfs2_holder_reinit(state, flags, gh);
1634 return gfs2_glock_nq(gh);
1641 gfs2_trans_end(sdp);
1650 * glock_compare_atime - Compare two struct gfs2_glock structures for sort
1651 * @arg_a: the first structure
1652 * @arg_b: the second structure
1654 * Returns: 1 if A > B
1659 static int glock_compare_atime(const void *arg_a, const void *arg_b)
1661 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1662 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1663 struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1664 struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1667 if (a->ln_number > b->ln_number)
1669 else if (a->ln_number < b->ln_number)
1672 if (gh_a->gh_state == LM_ST_SHARED &&
1673 gh_b->gh_state == LM_ST_EXCLUSIVE)
1675 else if (gh_a->gh_state == LM_ST_SHARED &&
1676 (gh_b->gh_flags & GL_ATIME))
1684 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an
1686 * @num_gh: the number of structures
1687 * @ghs: an array of struct gfs2_holder structures
1689 * Returns: 0 on success (all glocks acquired),
1690 * errno on failure (no glocks acquired)
1693 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs)
1695 struct gfs2_holder **p;
1703 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1704 if (ghs->gh_flags & GL_ATIME)
1705 error = gfs2_glock_nq_atime(ghs);
1707 error = gfs2_glock_nq(ghs);
1711 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1715 for (x = 0; x < num_gh; x++)
1718 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL);
1720 for (x = 0; x < num_gh; x++) {
1721 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1723 if (p[x]->gh_flags & GL_ATIME)
1724 error = gfs2_glock_nq_atime(p[x]);
1726 error = gfs2_glock_nq(p[x]);
1730 gfs2_glock_dq(p[x]);
1741 * gfs2_try_toss_vnode - See if we can toss a vnode from memory
1744 * Returns: 1 if the vnode was tossed
1747 void gfs2_try_toss_vnode(struct gfs2_inode *ip)
1749 struct inode *inode;
1751 inode = gfs2_ip2v_lookup(ip);
1755 d_prune_aliases(inode);
1757 if (S_ISDIR(ip->i_di.di_mode)) {
1758 struct list_head *head = &inode->i_dentry;
1759 struct dentry *d = NULL;
1761 spin_lock(&dcache_lock);
1762 if (list_empty(head))
1763 spin_unlock(&dcache_lock);
1765 d = list_entry(head->next, struct dentry, d_alias);
1767 spin_unlock(&dcache_lock);
1769 if (have_submounts(d))
1772 shrink_dcache_parent(d);
1774 d_prune_aliases(inode);
1785 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1787 struct buffer_head *dibh;
1790 error = gfs2_meta_inode_buffer(ip, &dibh);
1792 error = inode_setattr(ip->i_vnode, attr);
1793 gfs2_assert_warn(ip->i_sbd, !error);
1794 gfs2_inode_attr_out(ip);
1796 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1797 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1804 * gfs2_setattr_simple -
1808 * Called with a reference on the vnode.
1813 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1817 if (get_transaction)
1818 return __gfs2_setattr_simple(ip, attr);
1820 error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0);
1824 error = __gfs2_setattr_simple(ip, attr);
1826 gfs2_trans_end(ip->i_sbd);
1831 int gfs2_repermission(struct inode *inode, int mask, struct nameidata *nd)
1833 return permission(inode, mask, nd);