2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License v.2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/posix_acl.h>
16 #include <linux/sort.h>
17 #include <linux/gfs2_ondisk.h>
18 #include <linux/crc32.h>
19 #include <asm/semaphore.h>
22 #include "lm_interface.h"
33 #include "ops_address.h"
35 #include "ops_inode.h"
43 * inode_attr_in - Copy attributes from the dinode into the VFS inode
44 * @ip: The GFS2 inode (with embedded disk inode data)
45 * @inode: The Linux VFS inode
49 static void inode_attr_in(struct gfs2_inode *ip, struct inode *inode)
51 inode->i_ino = ip->i_num.no_formal_ino;
53 switch (ip->i_di.di_mode & S_IFMT) {
56 inode->i_rdev = MKDEV(ip->i_di.di_major, ip->i_di.di_minor);
63 inode->i_mode = ip->i_di.di_mode;
64 inode->i_nlink = ip->i_di.di_nlink;
65 inode->i_uid = ip->i_di.di_uid;
66 inode->i_gid = ip->i_di.di_gid;
67 i_size_write(inode, ip->i_di.di_size);
68 inode->i_atime.tv_sec = ip->i_di.di_atime;
69 inode->i_mtime.tv_sec = ip->i_di.di_mtime;
70 inode->i_ctime.tv_sec = ip->i_di.di_ctime;
71 inode->i_atime.tv_nsec = 0;
72 inode->i_mtime.tv_nsec = 0;
73 inode->i_ctime.tv_nsec = 0;
74 inode->i_blksize = PAGE_SIZE;
75 inode->i_blocks = ip->i_di.di_blocks <<
76 (ip->i_sbd->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT);
78 if (ip->i_di.di_flags & GFS2_DIF_IMMUTABLE)
79 inode->i_flags |= S_IMMUTABLE;
81 inode->i_flags &= ~S_IMMUTABLE;
83 if (ip->i_di.di_flags & GFS2_DIF_APPENDONLY)
84 inode->i_flags |= S_APPEND;
86 inode->i_flags &= ~S_APPEND;
90 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode
91 * @ip: The GFS2 inode (with embedded disk inode data)
95 void gfs2_inode_attr_in(struct gfs2_inode *ip)
99 inode = gfs2_ip2v_lookup(ip);
101 inode_attr_in(ip, inode);
107 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode
108 * @ip: The GFS2 inode
110 * Only copy out the attributes that we want the VFS layer
111 * to be able to modify.
114 void gfs2_inode_attr_out(struct gfs2_inode *ip)
116 struct inode *inode = ip->i_vnode;
118 gfs2_assert_withdraw(ip->i_sbd,
119 (ip->i_di.di_mode & S_IFMT) == (inode->i_mode & S_IFMT));
120 ip->i_di.di_mode = inode->i_mode;
121 ip->i_di.di_uid = inode->i_uid;
122 ip->i_di.di_gid = inode->i_gid;
123 ip->i_di.di_atime = inode->i_atime.tv_sec;
124 ip->i_di.di_mtime = inode->i_mtime.tv_sec;
125 ip->i_di.di_ctime = inode->i_ctime.tv_sec;
129 * gfs2_ip2v_lookup - Get the struct inode for a struct gfs2_inode
130 * @ip: the struct gfs2_inode to get the struct inode for
132 * Returns: A VFS inode, or NULL if none
135 struct inode *gfs2_ip2v_lookup(struct gfs2_inode *ip)
137 struct inode *inode = NULL;
139 gfs2_assert_warn(ip->i_sbd, test_bit(GIF_MIN_INIT, &ip->i_flags));
141 spin_lock(&ip->i_spin);
143 inode = igrab(ip->i_vnode);
144 spin_unlock(&ip->i_spin);
150 * gfs2_ip2v - Get/Create a struct inode for a struct gfs2_inode
151 * @ip: the struct gfs2_inode to get the struct inode for
153 * Returns: A VFS inode, or NULL if no mem
156 struct inode *gfs2_ip2v(struct gfs2_inode *ip)
158 struct inode *inode, *tmp;
160 inode = gfs2_ip2v_lookup(ip);
164 tmp = new_inode(ip->i_sbd->sd_vfs);
168 inode_attr_in(ip, tmp);
170 if (S_ISREG(ip->i_di.di_mode)) {
171 tmp->i_op = &gfs2_file_iops;
172 tmp->i_fop = &gfs2_file_fops;
173 tmp->i_mapping->a_ops = &gfs2_file_aops;
174 } else if (S_ISDIR(ip->i_di.di_mode)) {
175 tmp->i_op = &gfs2_dir_iops;
176 tmp->i_fop = &gfs2_dir_fops;
177 } else if (S_ISLNK(ip->i_di.di_mode)) {
178 tmp->i_op = &gfs2_symlink_iops;
180 tmp->i_op = &gfs2_dev_iops;
181 init_special_inode(tmp, tmp->i_mode, tmp->i_rdev);
184 tmp->u.generic_ip = NULL;
187 spin_lock(&ip->i_spin);
190 inode = igrab(ip->i_vnode);
191 spin_unlock(&ip->i_spin);
204 inode->u.generic_ip = ip;
206 spin_unlock(&ip->i_spin);
208 insert_inode_hash(inode);
213 static int iget_test(struct inode *inode, void *opaque)
215 struct gfs2_inode *ip = inode->u.generic_ip;
216 struct gfs2_inum *inum = (struct gfs2_inum *)opaque;
218 if (ip && ip->i_num.no_addr == inum->no_addr)
224 struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum *inum)
226 return ilookup5(sb, (unsigned long)inum->no_formal_ino,
230 void gfs2_inode_min_init(struct gfs2_inode *ip, unsigned int type)
232 if (!test_and_set_bit(GIF_MIN_INIT, &ip->i_flags)) {
233 ip->i_di.di_nlink = 1;
234 ip->i_di.di_mode = DT2IF(type);
239 * gfs2_inode_refresh - Refresh the incore copy of the dinode
240 * @ip: The GFS2 inode
245 int gfs2_inode_refresh(struct gfs2_inode *ip)
247 struct buffer_head *dibh;
250 error = gfs2_meta_inode_buffer(ip, &dibh);
254 if (gfs2_metatype_check(ip->i_sbd, dibh, GFS2_METATYPE_DI)) {
259 gfs2_dinode_in(&ip->i_di, dibh->b_data);
260 set_bit(GIF_MIN_INIT, &ip->i_flags);
264 if (ip->i_num.no_addr != ip->i_di.di_num.no_addr) {
265 if (gfs2_consist_inode(ip))
266 gfs2_dinode_print(&ip->i_di);
269 if (ip->i_num.no_formal_ino != ip->i_di.di_num.no_formal_ino)
272 ip->i_vn = ip->i_gl->gl_vn;
278 * inode_create - create a struct gfs2_inode
279 * @i_gl: The glock covering the inode
280 * @inum: The inode number
281 * @io_gl: the iopen glock to acquire/hold (using holder in new gfs2_inode)
282 * @io_state: the state the iopen glock should be acquired in
283 * @ipp: pointer to put the returned inode in
288 static int inode_create(struct gfs2_glock *i_gl, const struct gfs2_inum *inum,
289 struct gfs2_glock *io_gl, unsigned int io_state,
290 struct gfs2_inode **ipp, int need_lock)
292 struct gfs2_sbd *sdp = i_gl->gl_sbd;
293 struct gfs2_inode *ip;
296 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
299 memset(ip, 0, sizeof(struct gfs2_inode));
303 atomic_set(&ip->i_count, 1);
305 ip->i_vn = i_gl->gl_vn - 1;
310 spin_lock_init(&ip->i_spin);
311 init_rwsem(&ip->i_rw_mutex);
313 ip->i_greedy = gfs2_tune_get(sdp, gt_greedy_default);
316 error = gfs2_glock_nq_init(io_gl,
317 io_state, GL_LOCAL_EXCL | GL_EXACT,
323 spin_lock(&io_gl->gl_spin);
324 gfs2_glock_hold(i_gl);
325 io_gl->gl_object = i_gl;
326 spin_unlock(&io_gl->gl_spin);
328 gfs2_glock_hold(i_gl);
329 i_gl->gl_object = ip;
331 atomic_inc(&sdp->sd_inode_count);
338 gfs2_meta_cache_flush(ip);
339 kmem_cache_free(gfs2_inode_cachep, ip);
346 * gfs2_inode_get - Create or get a reference on an inode
347 * @i_gl: The glock covering the inode
348 * @inum: The inode number
350 * @ipp: pointer to put the returned inode in
355 int gfs2_inode_get(struct gfs2_glock *i_gl, const struct gfs2_inum *inum,
356 int create, struct gfs2_inode **ipp)
358 struct gfs2_sbd *sdp = i_gl->gl_sbd;
359 struct gfs2_glock *io_gl;
362 gfs2_glmutex_lock(i_gl);
364 *ipp = i_gl->gl_object;
367 if ((*ipp)->i_num.no_formal_ino != inum->no_formal_ino)
369 atomic_inc(&(*ipp)->i_count);
377 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops,
380 error = inode_create(i_gl, inum, io_gl, LM_ST_SHARED, ipp, 1);
381 gfs2_glock_put(io_gl);
385 gfs2_glmutex_unlock(i_gl);
390 void gfs2_inode_hold(struct gfs2_inode *ip)
392 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
393 atomic_inc(&ip->i_count);
396 void gfs2_inode_put(struct gfs2_inode *ip)
398 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
399 atomic_dec(&ip->i_count);
402 void gfs2_inode_destroy(struct gfs2_inode *ip, int unlock)
404 struct gfs2_sbd *sdp = ip->i_sbd;
405 struct gfs2_glock *i_gl = ip->i_gl;
407 gfs2_assert_warn(sdp, !atomic_read(&ip->i_count));
409 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl;
410 gfs2_assert(sdp, io_gl->gl_object == i_gl);
412 spin_lock(&io_gl->gl_spin);
413 io_gl->gl_object = NULL;
414 spin_unlock(&io_gl->gl_spin);
415 gfs2_glock_put(i_gl);
417 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
420 gfs2_meta_cache_flush(ip);
421 kmem_cache_free(gfs2_inode_cachep, ip);
423 i_gl->gl_object = NULL;
424 gfs2_glock_put(i_gl);
426 atomic_dec(&sdp->sd_inode_count);
429 static int dinode_dealloc(struct gfs2_inode *ip, struct gfs2_unlinked *ul)
431 struct gfs2_sbd *sdp = ip->i_sbd;
432 struct gfs2_alloc *al;
433 struct gfs2_rgrpd *rgd;
436 if (ip->i_di.di_blocks != 1) {
437 if (gfs2_consist_inode(ip))
438 gfs2_dinode_print(&ip->i_di);
442 al = gfs2_alloc_get(ip);
444 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
448 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
452 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
454 gfs2_consist_inode(ip);
456 goto out_rindex_relse;
459 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
462 goto out_rindex_relse;
464 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
465 RES_STATFS + RES_QUOTA, 1);
469 gfs2_trans_add_gl(ip->i_gl);
471 gfs2_free_di(rgd, ip);
473 error = gfs2_unlinked_ondisk_rm(sdp, ul);
476 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags);
479 gfs2_glock_dq_uninit(&al->al_rgd_gh);
482 gfs2_glock_dq_uninit(&al->al_ri_gh);
485 gfs2_quota_unhold(ip);
494 * inode_dealloc - Deallocate all on-disk blocks for an inode (dinode)
495 * @sdp: the filesystem
496 * @inum: the inode number to deallocate
497 * @io_gh: a holder for the iopen glock for this inode
499 * N.B. When we enter this we already hold the iopen glock and getting
500 * the glock for the inode means that we are grabbing the locks in the
501 * "wrong" order so we must only so a try lock operation and fail if we
502 * don't get the lock. Thats ok, since if we fail it means someone else
503 * is using the inode still and thus we shouldn't be deallocating it
509 static int inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul,
510 struct gfs2_holder *io_gh)
512 struct gfs2_inode *ip;
513 struct gfs2_holder i_gh;
516 error = gfs2_glock_nq_num(sdp, ul->ul_ut.ut_inum.no_addr,
517 &gfs2_inode_glops, LM_ST_EXCLUSIVE,
518 LM_FLAG_TRY_1CB, &i_gh);
523 return 1; /* or back off and relock in different order? */
528 gfs2_assert_warn(sdp, !i_gh.gh_gl->gl_object);
529 error = inode_create(i_gh.gh_gl, &ul->ul_ut.ut_inum, io_gh->gh_gl,
530 LM_ST_EXCLUSIVE, &ip, 0);
535 error = gfs2_inode_refresh(ip);
539 if (ip->i_di.di_nlink) {
540 if (gfs2_consist_inode(ip))
541 gfs2_dinode_print(&ip->i_di);
546 if (S_ISDIR(ip->i_di.di_mode) &&
547 (ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
548 error = gfs2_dir_exhash_dealloc(ip);
553 if (ip->i_di.di_eattr) {
554 error = gfs2_ea_dealloc(ip);
559 if (!gfs2_is_stuffed(ip)) {
560 error = gfs2_file_dealloc(ip);
565 error = dinode_dealloc(ip, ul);
570 gfs2_glmutex_lock(i_gh.gh_gl);
572 gfs2_inode_destroy(ip, 0);
573 gfs2_glmutex_unlock(i_gh.gh_gl);
576 gfs2_glock_dq_uninit(&i_gh);
582 * try_inode_dealloc - Try to deallocate an inode and all its blocks
583 * @sdp: the filesystem
585 * Returns: 0 on success, -errno on error, 1 on busy (inode open)
588 static int try_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
591 struct gfs2_holder iogh;
593 gfs2_try_toss_inode(sdp, &ul->ul_ut.ut_inum);
594 error = gfs2_glock_nq_num(sdp, ul->ul_ut.ut_inum.no_addr,
595 &gfs2_iopen_glops, LM_ST_EXCLUSIVE,
596 LM_FLAG_TRY_1CB, &iogh);
606 error = inode_dealloc(sdp, ul, &iogh);
607 gfs2_glock_dq_uninit(&iogh);
612 static int inode_dealloc_uninit(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
614 struct gfs2_rgrpd *rgd;
615 struct gfs2_holder ri_gh, rgd_gh;
618 error = gfs2_rindex_hold(sdp, &ri_gh);
622 rgd = gfs2_blk2rgrpd(sdp, ul->ul_ut.ut_inum.no_addr);
629 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rgd_gh);
633 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED + RES_STATFS, 0);
637 gfs2_free_uninit_di(rgd, ul->ul_ut.ut_inum.no_addr);
638 gfs2_unlinked_ondisk_rm(sdp, ul);
643 gfs2_glock_dq_uninit(&rgd_gh);
645 gfs2_glock_dq_uninit(&ri_gh);
650 int gfs2_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
652 if (ul->ul_ut.ut_flags & GFS2_UTF_UNINIT)
653 return inode_dealloc_uninit(sdp, ul);
655 return try_inode_dealloc(sdp, ul);
659 * gfs2_change_nlink - Change nlink count on inode
660 * @ip: The GFS2 inode
661 * @diff: The change in the nlink count required
666 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
668 struct buffer_head *dibh;
672 nlink = ip->i_di.di_nlink + diff;
674 /* If we are reducing the nlink count, but the new value ends up being
675 bigger than the old one, we must have underflowed. */
676 if (diff < 0 && nlink > ip->i_di.di_nlink) {
677 if (gfs2_consist_inode(ip))
678 gfs2_dinode_print(&ip->i_di);
682 error = gfs2_meta_inode_buffer(ip, &dibh);
686 ip->i_di.di_nlink = nlink;
687 ip->i_di.di_ctime = get_seconds();
689 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
690 gfs2_dinode_out(&ip->i_di, dibh->b_data);
696 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
699 gfs2_str2qstr(&qstr, name);
700 return gfs2_lookupi(dip, &qstr, 1, NULL);
705 * gfs2_lookupi - Look up a filename in a directory and return its inode
706 * @d_gh: An initialized holder for the directory glock
707 * @name: The name of the inode to look for
708 * @is_root: If 1, ignore the caller's permissions
709 * @i_gh: An uninitialized holder for the new inode glock
711 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless
717 struct inode *gfs2_lookupi(struct inode *dir, struct qstr *name, int is_root,
718 struct nameidata *nd)
721 struct super_block *sb = dir->i_sb;
722 struct gfs2_inode *ipp;
723 struct gfs2_inode *dip = dir->u.generic_ip;
724 struct gfs2_sbd *sdp = dip->i_sbd;
725 struct gfs2_holder d_gh;
726 struct gfs2_inum inum;
728 struct gfs2_glock *gl;
730 struct inode *inode = NULL;
732 if (!name->len || name->len > GFS2_FNAMESIZE)
733 return ERR_PTR(-ENAMETOOLONG);
735 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
736 (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
737 dir == sb->s_root->d_inode)) {
738 gfs2_inode_hold(dip);
743 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
745 return ERR_PTR(error);
748 error = gfs2_repermission(dip->i_vnode, MAY_EXEC, NULL);
753 error = gfs2_dir_search(dir, name, &inum, &type);
757 error = gfs2_glock_get(sdp, inum.no_addr, &gfs2_inode_glops,
762 error = gfs2_inode_get(gl, &inum, CREATE, &ipp);
764 gfs2_inode_min_init(ipp, type);
769 gfs2_glock_dq_uninit(&d_gh);
771 if (error == -ENOENT)
774 inode = gfs2_ip2v(ipp);
777 return ERR_PTR(-ENOMEM);
780 return ERR_PTR(error);
783 static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino)
785 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
786 struct buffer_head *bh;
787 struct gfs2_inum_range ir;
790 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
793 mutex_lock(&sdp->sd_inum_mutex);
795 error = gfs2_meta_inode_buffer(ip, &bh);
797 mutex_unlock(&sdp->sd_inum_mutex);
802 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
805 *formal_ino = ir.ir_start++;
807 gfs2_trans_add_bh(ip->i_gl, bh, 1);
808 gfs2_inum_range_out(&ir,
809 bh->b_data + sizeof(struct gfs2_dinode));
811 mutex_unlock(&sdp->sd_inum_mutex);
818 mutex_unlock(&sdp->sd_inum_mutex);
824 static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino)
826 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
827 struct gfs2_inode *m_ip = sdp->sd_inum_inode->u.generic_ip;
828 struct gfs2_holder gh;
829 struct buffer_head *bh;
830 struct gfs2_inum_range ir;
833 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
837 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
840 mutex_lock(&sdp->sd_inum_mutex);
842 error = gfs2_meta_inode_buffer(ip, &bh);
846 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
849 struct buffer_head *m_bh;
852 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
856 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode));
857 x = y = be64_to_cpu(x);
859 ir.ir_length = GFS2_INUM_QUANTUM;
860 x += GFS2_INUM_QUANTUM;
862 gfs2_consist_inode(m_ip);
864 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
865 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x;
870 *formal_ino = ir.ir_start++;
873 gfs2_trans_add_bh(ip->i_gl, bh, 1);
874 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
880 mutex_unlock(&sdp->sd_inum_mutex);
884 gfs2_glock_dq_uninit(&gh);
889 static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum)
893 error = pick_formal_ino_1(sdp, inum);
897 error = pick_formal_ino_2(sdp, inum);
903 * create_ok - OK to create a new on-disk inode here?
904 * @dip: Directory in which dinode is to be created
905 * @name: Name of new dinode
911 static int create_ok(struct gfs2_inode *dip, struct qstr *name,
916 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
920 /* Don't create entries in an unlinked directory */
921 if (!dip->i_di.di_nlink)
924 error = gfs2_dir_search(dip->i_vnode, name, NULL, NULL);
935 if (dip->i_di.di_entries == (uint32_t)-1)
937 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1)
943 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
944 unsigned int *uid, unsigned int *gid)
946 if (dip->i_sbd->sd_args.ar_suiddir &&
947 (dip->i_di.di_mode & S_ISUID) &&
951 else if (dip->i_di.di_uid != current->fsuid)
953 *uid = dip->i_di.di_uid;
955 *uid = current->fsuid;
957 if (dip->i_di.di_mode & S_ISGID) {
960 *gid = dip->i_di.di_gid;
962 *gid = current->fsgid;
965 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_unlinked *ul)
967 struct gfs2_sbd *sdp = dip->i_sbd;
972 dip->i_alloc.al_requested = RES_DINODE;
973 error = gfs2_inplace_reserve(dip);
977 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
982 ul->ul_ut.ut_inum.no_addr = gfs2_alloc_di(dip);
984 ul->ul_ut.ut_flags = GFS2_UTF_UNINIT;
985 error = gfs2_unlinked_ondisk_add(sdp, ul);
990 gfs2_inplace_release(dip);
999 * init_dinode - Fill in a new dinode structure
1000 * @dip: the directory this inode is being created in
1001 * @gl: The glock covering the new inode
1002 * @inum: the inode number
1003 * @mode: the file permissions
1009 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1010 struct gfs2_inum *inum, unsigned int mode,
1011 unsigned int uid, unsigned int gid)
1013 struct gfs2_sbd *sdp = dip->i_sbd;
1014 struct gfs2_dinode *di;
1015 struct buffer_head *dibh;
1017 dibh = gfs2_meta_new(gl, inum->no_addr);
1018 gfs2_trans_add_bh(gl, dibh, 1);
1019 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
1020 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1021 di = (struct gfs2_dinode *)dibh->b_data;
1023 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
1024 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
1025 di->di_mode = cpu_to_be32(mode);
1026 di->di_uid = cpu_to_be32(uid);
1027 di->di_gid = cpu_to_be32(gid);
1028 di->di_nlink = cpu_to_be32(0);
1029 di->di_size = cpu_to_be64(0);
1030 di->di_blocks = cpu_to_be64(1);
1031 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
1032 di->di_major = di->di_minor = cpu_to_be32(0);
1033 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
1034 di->__pad[0] = di->__pad[1] = 0;
1035 di->di_flags = cpu_to_be32(0);
1037 if (S_ISREG(mode)) {
1038 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
1039 gfs2_tune_get(sdp, gt_new_files_jdata))
1040 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
1041 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
1042 gfs2_tune_get(sdp, gt_new_files_directio))
1043 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
1044 } else if (S_ISDIR(mode)) {
1045 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1046 GFS2_DIF_INHERIT_DIRECTIO);
1047 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1048 GFS2_DIF_INHERIT_JDATA);
1052 di->di_height = cpu_to_be32(0);
1055 di->di_depth = cpu_to_be16(0);
1056 di->di_entries = cpu_to_be32(0);
1057 memset(&di->__pad4, 0, sizeof(di->__pad4));
1058 di->di_eattr = cpu_to_be64(0);
1059 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
1064 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1065 unsigned int mode, struct gfs2_unlinked *ul)
1067 struct gfs2_sbd *sdp = dip->i_sbd;
1068 unsigned int uid, gid;
1071 munge_mode_uid_gid(dip, &mode, &uid, &gid);
1073 gfs2_alloc_get(dip);
1075 error = gfs2_quota_lock(dip, uid, gid);
1079 error = gfs2_quota_check(dip, uid, gid);
1083 error = gfs2_trans_begin(sdp, RES_DINODE + RES_UNLINKED +
1088 ul->ul_ut.ut_flags = 0;
1089 error = gfs2_unlinked_ondisk_munge(sdp, ul);
1091 init_dinode(dip, gl, &ul->ul_ut.ut_inum,
1094 gfs2_quota_change(dip, +1, uid, gid);
1096 gfs2_trans_end(sdp);
1099 gfs2_quota_unlock(dip);
1102 gfs2_alloc_put(dip);
1107 static int link_dinode(struct gfs2_inode *dip, struct qstr *name,
1108 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1110 struct gfs2_sbd *sdp = dip->i_sbd;
1111 struct gfs2_alloc *al;
1113 struct buffer_head *dibh;
1116 al = gfs2_alloc_get(dip);
1118 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1122 error = alloc_required = gfs2_diradd_alloc_required(dip->i_vnode, name);
1123 if (alloc_required < 0)
1125 if (alloc_required) {
1126 error = gfs2_quota_check(dip, dip->i_di.di_uid,
1129 goto fail_quota_locks;
1131 al->al_requested = sdp->sd_max_dirres;
1133 error = gfs2_inplace_reserve(dip);
1135 goto fail_quota_locks;
1137 error = gfs2_trans_begin(sdp,
1138 sdp->sd_max_dirres +
1139 al->al_rgd->rd_ri.ri_length +
1140 2 * RES_DINODE + RES_UNLINKED +
1141 RES_STATFS + RES_QUOTA, 0);
1145 error = gfs2_trans_begin(sdp,
1150 goto fail_quota_locks;
1153 error = gfs2_dir_add(dip->i_vnode, name, &ip->i_num, IF2DT(ip->i_di.di_mode));
1155 goto fail_end_trans;
1157 error = gfs2_meta_inode_buffer(ip, &dibh);
1159 goto fail_end_trans;
1160 ip->i_di.di_nlink = 1;
1161 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1162 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1165 error = gfs2_unlinked_ondisk_rm(sdp, ul);
1167 goto fail_end_trans;
1172 gfs2_trans_end(sdp);
1175 if (dip->i_alloc.al_rgd)
1176 gfs2_inplace_release(dip);
1179 gfs2_quota_unlock(dip);
1182 gfs2_alloc_put(dip);
1188 * gfs2_createi - Create a new inode
1189 * @ghs: An array of two holders
1190 * @name: The name of the new file
1191 * @mode: the permissions on the new inode
1193 * @ghs[0] is an initialized holder for the directory
1194 * @ghs[1] is the holder for the inode lock
1196 * If the return value is not NULL, the glocks on both the directory and the new
1197 * file are held. A transaction has been started and an inplace reservation
1203 struct inode *gfs2_createi(struct gfs2_holder *ghs, struct qstr *name,
1206 struct inode *inode;
1207 struct gfs2_inode *dip = ghs->gh_gl->gl_object;
1208 struct gfs2_sbd *sdp = dip->i_sbd;
1209 struct gfs2_unlinked *ul;
1210 struct gfs2_inode *ip;
1213 if (!name->len || name->len > GFS2_FNAMESIZE)
1214 return ERR_PTR(-ENAMETOOLONG);
1216 error = gfs2_unlinked_get(sdp, &ul);
1218 return ERR_PTR(error);
1220 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1221 error = gfs2_glock_nq(ghs);
1225 error = create_ok(dip, name, mode);
1229 error = pick_formal_ino(sdp, &ul->ul_ut.ut_inum.no_formal_ino);
1233 error = alloc_dinode(dip, ul);
1237 if (ul->ul_ut.ut_inum.no_addr < dip->i_num.no_addr) {
1240 error = gfs2_glock_nq_num(sdp,
1241 ul->ul_ut.ut_inum.no_addr,
1243 LM_ST_EXCLUSIVE, GL_SKIP,
1246 gfs2_unlinked_put(sdp, ul);
1247 return ERR_PTR(error);
1250 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1251 error = gfs2_glock_nq(ghs);
1253 gfs2_glock_dq_uninit(ghs + 1);
1254 gfs2_unlinked_put(sdp, ul);
1255 return ERR_PTR(error);
1258 error = create_ok(dip, name, mode);
1262 error = gfs2_glock_nq_num(sdp,
1263 ul->ul_ut.ut_inum.no_addr,
1265 LM_ST_EXCLUSIVE, GL_SKIP,
1271 error = make_dinode(dip, ghs[1].gh_gl, mode, ul);
1275 error = gfs2_inode_get(ghs[1].gh_gl, &ul->ul_ut.ut_inum, CREATE, &ip);
1279 error = gfs2_inode_refresh(ip);
1283 error = gfs2_acl_create(dip, ip);
1287 error = link_dinode(dip, name, ip, ul);
1291 gfs2_unlinked_put(sdp, ul);
1293 inode = gfs2_ip2v(ip);
1296 return ERR_PTR(-ENOMEM);
1303 gfs2_glock_dq_uninit(ghs + 1);
1309 gfs2_unlinked_put(sdp, ul);
1311 return ERR_PTR(error);
1315 * gfs2_unlinki - Unlink a file
1316 * @dip: The inode of the directory
1317 * @name: The name of the file to be unlinked
1318 * @ip: The inode of the file to be removed
1320 * Assumes Glocks on both dip and ip are held.
1325 int gfs2_unlinki(struct gfs2_inode *dip, struct qstr *name,
1326 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1328 struct gfs2_sbd *sdp = dip->i_sbd;
1331 error = gfs2_dir_del(dip, name);
1335 error = gfs2_change_nlink(ip, -1);
1339 /* If this inode is being unlinked from the directory structure,
1340 we need to mark that in the log so that it isn't lost during
1343 if (!ip->i_di.di_nlink) {
1344 ul->ul_ut.ut_inum = ip->i_num;
1345 error = gfs2_unlinked_ondisk_add(sdp, ul);
1347 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1354 * gfs2_rmdiri - Remove a directory
1355 * @dip: The parent directory of the directory to be removed
1356 * @name: The name of the directory to be removed
1357 * @ip: The GFS2 inode of the directory to be removed
1359 * Assumes Glocks on dip and ip are held
1364 int gfs2_rmdiri(struct gfs2_inode *dip, struct qstr *name,
1365 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1367 struct gfs2_sbd *sdp = dip->i_sbd;
1368 struct qstr dotname;
1371 if (ip->i_di.di_entries != 2) {
1372 if (gfs2_consist_inode(ip))
1373 gfs2_dinode_print(&ip->i_di);
1377 error = gfs2_dir_del(dip, name);
1381 error = gfs2_change_nlink(dip, -1);
1385 gfs2_str2qstr(&dotname, ".");
1386 error = gfs2_dir_del(ip, &dotname);
1391 dotname.name = "..";
1392 dotname.hash = gfs2_disk_hash(dotname.name, dotname.len);
1393 error = gfs2_dir_del(ip, &dotname);
1397 error = gfs2_change_nlink(ip, -2);
1401 /* This inode is being unlinked from the directory structure and
1402 we need to mark that in the log so that it isn't lost during
1405 ul->ul_ut.ut_inum = ip->i_num;
1406 error = gfs2_unlinked_ondisk_add(sdp, ul);
1408 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1414 * gfs2_unlink_ok - check to see that a inode is still in a directory
1415 * @dip: the directory
1416 * @name: the name of the file
1419 * Assumes that the lock on (at least) @dip is held.
1421 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1424 int gfs2_unlink_ok(struct gfs2_inode *dip, struct qstr *name,
1425 struct gfs2_inode *ip)
1427 struct gfs2_inum inum;
1431 if (IS_IMMUTABLE(ip->i_vnode) || IS_APPEND(ip->i_vnode))
1434 if ((dip->i_di.di_mode & S_ISVTX) &&
1435 dip->i_di.di_uid != current->fsuid &&
1436 ip->i_di.di_uid != current->fsuid &&
1437 !capable(CAP_FOWNER))
1440 if (IS_APPEND(dip->i_vnode))
1443 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
1447 error = gfs2_dir_search(dip->i_vnode, name, &inum, &type);
1451 if (!gfs2_inum_equal(&inum, &ip->i_num))
1454 if (IF2DT(ip->i_di.di_mode) != type) {
1455 gfs2_consist_inode(dip);
1463 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1467 * Follow @to back to the root and make sure we don't encounter @this
1468 * Assumes we already hold the rename lock.
1473 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1475 struct inode *dir = to->i_vnode;
1476 struct super_block *sb = dir->i_sb;
1481 gfs2_str2qstr(&dotdot, "..");
1486 if (dir == this->i_vnode) {
1490 if (dir == sb->s_root->d_inode) {
1495 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL);
1497 error = PTR_ERR(tmp);
1511 * gfs2_readlinki - return the contents of a symlink
1512 * @ip: the symlink's inode
1513 * @buf: a pointer to the buffer to be filled
1514 * @len: a pointer to the length of @buf
1516 * If @buf is too small, a piece of memory is kmalloc()ed and needs
1517 * to be freed by the caller.
1522 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1524 struct gfs2_holder i_gh;
1525 struct buffer_head *dibh;
1529 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1530 error = gfs2_glock_nq_atime(&i_gh);
1532 gfs2_holder_uninit(&i_gh);
1536 if (!ip->i_di.di_size) {
1537 gfs2_consist_inode(ip);
1542 error = gfs2_meta_inode_buffer(ip, &dibh);
1546 x = ip->i_di.di_size + 1;
1548 *buf = kmalloc(x, GFP_KERNEL);
1555 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1562 gfs2_glock_dq_uninit(&i_gh);
1568 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1569 * conditionally update the inode's atime
1570 * @gh: the holder to acquire
1572 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1573 * Update if the difference between the current time and the inode's current
1574 * atime is greater than an interval specified at mount.
1579 int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1581 struct gfs2_glock *gl = gh->gh_gl;
1582 struct gfs2_sbd *sdp = gl->gl_sbd;
1583 struct gfs2_inode *ip = gl->gl_object;
1584 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1589 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1590 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1591 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1594 state = gh->gh_state;
1595 flags = gh->gh_flags;
1597 error = gfs2_glock_nq(gh);
1601 if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1602 (sdp->sd_vfs->s_flags & MS_RDONLY))
1605 curtime = get_seconds();
1606 if (curtime - ip->i_di.di_atime >= quantum) {
1608 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY,
1610 error = gfs2_glock_nq(gh);
1614 /* Verify that atime hasn't been updated while we were
1615 trying to get exclusive lock. */
1617 curtime = get_seconds();
1618 if (curtime - ip->i_di.di_atime >= quantum) {
1619 struct buffer_head *dibh;
1621 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1622 if (error == -EROFS)
1627 error = gfs2_meta_inode_buffer(ip, &dibh);
1629 goto fail_end_trans;
1631 ip->i_di.di_atime = curtime;
1633 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1634 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1637 gfs2_trans_end(sdp);
1640 /* If someone else has asked for the glock,
1641 unlock and let them have it. Then reacquire
1642 in the original state. */
1643 if (gfs2_glock_is_blocking(gl)) {
1645 gfs2_holder_reinit(state, flags, gh);
1646 return gfs2_glock_nq(gh);
1653 gfs2_trans_end(sdp);
1662 * glock_compare_atime - Compare two struct gfs2_glock structures for sort
1663 * @arg_a: the first structure
1664 * @arg_b: the second structure
1666 * Returns: 1 if A > B
1671 static int glock_compare_atime(const void *arg_a, const void *arg_b)
1673 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1674 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1675 struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1676 struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1679 if (a->ln_number > b->ln_number)
1681 else if (a->ln_number < b->ln_number)
1684 if (gh_a->gh_state == LM_ST_SHARED &&
1685 gh_b->gh_state == LM_ST_EXCLUSIVE)
1687 else if (gh_a->gh_state == LM_ST_SHARED &&
1688 (gh_b->gh_flags & GL_ATIME))
1696 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an
1698 * @num_gh: the number of structures
1699 * @ghs: an array of struct gfs2_holder structures
1701 * Returns: 0 on success (all glocks acquired),
1702 * errno on failure (no glocks acquired)
1705 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs)
1707 struct gfs2_holder **p;
1715 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1716 if (ghs->gh_flags & GL_ATIME)
1717 error = gfs2_glock_nq_atime(ghs);
1719 error = gfs2_glock_nq(ghs);
1723 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1727 for (x = 0; x < num_gh; x++)
1730 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL);
1732 for (x = 0; x < num_gh; x++) {
1733 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1735 if (p[x]->gh_flags & GL_ATIME)
1736 error = gfs2_glock_nq_atime(p[x]);
1738 error = gfs2_glock_nq(p[x]);
1742 gfs2_glock_dq(p[x]);
1753 * gfs2_try_toss_vnode - See if we can toss a vnode from memory
1756 * Returns: 1 if the vnode was tossed
1759 void gfs2_try_toss_vnode(struct gfs2_inode *ip)
1761 struct inode *inode;
1763 inode = gfs2_ip2v_lookup(ip);
1767 d_prune_aliases(inode);
1769 if (S_ISDIR(ip->i_di.di_mode)) {
1770 struct list_head *head = &inode->i_dentry;
1771 struct dentry *d = NULL;
1773 spin_lock(&dcache_lock);
1774 if (list_empty(head))
1775 spin_unlock(&dcache_lock);
1777 d = list_entry(head->next, struct dentry, d_alias);
1779 spin_unlock(&dcache_lock);
1781 if (have_submounts(d))
1784 shrink_dcache_parent(d);
1786 d_prune_aliases(inode);
1797 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1799 struct buffer_head *dibh;
1802 error = gfs2_meta_inode_buffer(ip, &dibh);
1804 error = inode_setattr(ip->i_vnode, attr);
1805 gfs2_assert_warn(ip->i_sbd, !error);
1806 gfs2_inode_attr_out(ip);
1808 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1809 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1816 * gfs2_setattr_simple -
1820 * Called with a reference on the vnode.
1825 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1829 if (current->journal_info)
1830 return __gfs2_setattr_simple(ip, attr);
1832 error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0);
1836 error = __gfs2_setattr_simple(ip, attr);
1838 gfs2_trans_end(ip->i_sbd);
1843 int gfs2_repermission(struct inode *inode, int mask, struct nameidata *nd)
1845 return permission(inode, mask, nd);