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)
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);
315 error = gfs2_glock_nq_init(io_gl,
316 io_state, GL_LOCAL_EXCL | GL_EXACT,
320 ip->i_iopen_gh.gh_owner = NULL;
322 spin_lock(&io_gl->gl_spin);
323 gfs2_glock_hold(i_gl);
324 io_gl->gl_object = i_gl;
325 spin_unlock(&io_gl->gl_spin);
327 gfs2_glock_hold(i_gl);
328 i_gl->gl_object = ip;
330 atomic_inc(&sdp->sd_inode_count);
337 gfs2_meta_cache_flush(ip);
338 kmem_cache_free(gfs2_inode_cachep, ip);
345 * gfs2_inode_get - Create or get a reference on an inode
346 * @i_gl: The glock covering the inode
347 * @inum: The inode number
349 * @ipp: pointer to put the returned inode in
354 int gfs2_inode_get(struct gfs2_glock *i_gl, const struct gfs2_inum *inum,
355 int create, struct gfs2_inode **ipp)
357 struct gfs2_sbd *sdp = i_gl->gl_sbd;
358 struct gfs2_glock *io_gl;
361 gfs2_glmutex_lock(i_gl);
363 *ipp = i_gl->gl_object;
366 if ((*ipp)->i_num.no_formal_ino != inum->no_formal_ino)
368 atomic_inc(&(*ipp)->i_count);
376 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops,
379 error = inode_create(i_gl, inum, io_gl, LM_ST_SHARED, ipp);
380 gfs2_glock_put(io_gl);
384 gfs2_glmutex_unlock(i_gl);
389 void gfs2_inode_hold(struct gfs2_inode *ip)
391 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
392 atomic_inc(&ip->i_count);
395 void gfs2_inode_put(struct gfs2_inode *ip)
397 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
398 atomic_dec(&ip->i_count);
401 void gfs2_inode_destroy(struct gfs2_inode *ip)
403 struct gfs2_sbd *sdp = ip->i_sbd;
404 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl;
405 struct gfs2_glock *i_gl = ip->i_gl;
407 gfs2_assert_warn(sdp, !atomic_read(&ip->i_count));
408 gfs2_assert(sdp, io_gl->gl_object == i_gl);
410 spin_lock(&io_gl->gl_spin);
411 io_gl->gl_object = NULL;
412 spin_unlock(&io_gl->gl_spin);
413 gfs2_glock_put(i_gl);
415 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
417 gfs2_meta_cache_flush(ip);
418 kmem_cache_free(gfs2_inode_cachep, ip);
420 i_gl->gl_object = NULL;
421 gfs2_glock_put(i_gl);
423 atomic_dec(&sdp->sd_inode_count);
426 static int dinode_dealloc(struct gfs2_inode *ip, struct gfs2_unlinked *ul)
428 struct gfs2_sbd *sdp = ip->i_sbd;
429 struct gfs2_alloc *al;
430 struct gfs2_rgrpd *rgd;
433 if (ip->i_di.di_blocks != 1) {
434 if (gfs2_consist_inode(ip))
435 gfs2_dinode_print(&ip->i_di);
439 al = gfs2_alloc_get(ip);
441 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
445 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
449 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
451 gfs2_consist_inode(ip);
453 goto out_rindex_relse;
456 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
459 goto out_rindex_relse;
461 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
462 RES_STATFS + RES_QUOTA, 1);
466 gfs2_trans_add_gl(ip->i_gl);
468 gfs2_free_di(rgd, ip);
470 error = gfs2_unlinked_ondisk_rm(sdp, ul);
473 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags);
476 gfs2_glock_dq_uninit(&al->al_rgd_gh);
479 gfs2_glock_dq_uninit(&al->al_ri_gh);
482 gfs2_quota_unhold(ip);
491 * inode_dealloc - Deallocate all on-disk blocks for an inode (dinode)
492 * @sdp: the filesystem
493 * @inum: the inode number to deallocate
494 * @io_gh: a holder for the iopen glock for this inode
499 static int inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul,
500 struct gfs2_holder *io_gh)
502 struct gfs2_inode *ip;
503 struct gfs2_holder i_gh;
506 error = gfs2_glock_nq_num(sdp,
507 ul->ul_ut.ut_inum.no_addr, &gfs2_inode_glops,
508 LM_ST_EXCLUSIVE, 0, &i_gh);
512 /* We reacquire the iopen lock here to avoid a race with the NFS server
513 calling gfs2_read_inode() with the inode number of a inode we're in
514 the process of deallocating. And we can't keep our hold on the lock
515 from inode_dealloc_init() for deadlock reasons. */
517 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY, io_gh);
518 error = gfs2_glock_nq(io_gh);
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);
532 gfs2_glock_dq(io_gh);
537 error = gfs2_inode_refresh(ip);
541 if (ip->i_di.di_nlink) {
542 if (gfs2_consist_inode(ip))
543 gfs2_dinode_print(&ip->i_di);
548 if (S_ISDIR(ip->i_di.di_mode) &&
549 (ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
550 error = gfs2_dir_exhash_dealloc(ip);
555 if (ip->i_di.di_eattr) {
556 error = gfs2_ea_dealloc(ip);
561 if (!gfs2_is_stuffed(ip)) {
562 error = gfs2_file_dealloc(ip);
567 error = dinode_dealloc(ip, ul);
572 gfs2_glmutex_lock(i_gh.gh_gl);
574 gfs2_inode_destroy(ip);
575 gfs2_glmutex_unlock(i_gh.gh_gl);
578 gfs2_glock_dq_uninit(&i_gh);
584 * try_inode_dealloc - Try to deallocate an inode and all its blocks
585 * @sdp: the filesystem
587 * Returns: 0 on success, -errno on error, 1 on busy (inode open)
590 static int try_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
592 struct gfs2_holder io_gh;
595 gfs2_try_toss_inode(sdp, &ul->ul_ut.ut_inum);
597 error = gfs2_glock_nq_num(sdp,
598 ul->ul_ut.ut_inum.no_addr, &gfs2_iopen_glops,
599 LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB, &io_gh);
609 gfs2_glock_dq(&io_gh);
610 error = inode_dealloc(sdp, ul, &io_gh);
611 gfs2_holder_uninit(&io_gh);
616 static int inode_dealloc_uninit(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
618 struct gfs2_rgrpd *rgd;
619 struct gfs2_holder ri_gh, rgd_gh;
622 error = gfs2_rindex_hold(sdp, &ri_gh);
626 rgd = gfs2_blk2rgrpd(sdp, ul->ul_ut.ut_inum.no_addr);
633 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rgd_gh);
637 error = gfs2_trans_begin(sdp,
638 RES_RG_BIT + RES_UNLINKED + RES_STATFS,
643 gfs2_free_uninit_di(rgd, ul->ul_ut.ut_inum.no_addr);
644 gfs2_unlinked_ondisk_rm(sdp, ul);
649 gfs2_glock_dq_uninit(&rgd_gh);
651 gfs2_glock_dq_uninit(&ri_gh);
656 int gfs2_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
658 if (ul->ul_ut.ut_flags & GFS2_UTF_UNINIT)
659 return inode_dealloc_uninit(sdp, ul);
661 return try_inode_dealloc(sdp, ul);
665 * gfs2_change_nlink - Change nlink count on inode
666 * @ip: The GFS2 inode
667 * @diff: The change in the nlink count required
672 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
674 struct buffer_head *dibh;
678 nlink = ip->i_di.di_nlink + diff;
680 /* If we are reducing the nlink count, but the new value ends up being
681 bigger than the old one, we must have underflowed. */
682 if (diff < 0 && nlink > ip->i_di.di_nlink) {
683 if (gfs2_consist_inode(ip))
684 gfs2_dinode_print(&ip->i_di);
688 error = gfs2_meta_inode_buffer(ip, &dibh);
692 ip->i_di.di_nlink = nlink;
693 ip->i_di.di_ctime = get_seconds();
695 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
696 gfs2_dinode_out(&ip->i_di, dibh->b_data);
702 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
705 gfs2_str2qstr(&qstr, name);
706 return gfs2_lookupi(dip, &qstr, 1, NULL);
711 * gfs2_lookupi - Look up a filename in a directory and return its inode
712 * @d_gh: An initialized holder for the directory glock
713 * @name: The name of the inode to look for
714 * @is_root: If 1, ignore the caller's permissions
715 * @i_gh: An uninitialized holder for the new inode glock
717 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless
723 struct inode *gfs2_lookupi(struct inode *dir, struct qstr *name, int is_root,
724 struct nameidata *nd)
727 struct super_block *sb = dir->i_sb;
728 struct gfs2_inode *ipp;
729 struct gfs2_inode *dip = dir->u.generic_ip;
730 struct gfs2_sbd *sdp = dip->i_sbd;
731 struct gfs2_holder d_gh;
732 struct gfs2_inum inum;
734 struct gfs2_glock *gl;
736 struct inode *inode = NULL;
738 if (!name->len || name->len > GFS2_FNAMESIZE)
739 return ERR_PTR(-ENAMETOOLONG);
741 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
742 (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
743 dir == sb->s_root->d_inode)) {
744 gfs2_inode_hold(dip);
749 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
751 return ERR_PTR(error);
754 error = gfs2_repermission(dip->i_vnode, MAY_EXEC, NULL);
759 error = gfs2_dir_search(dir, name, &inum, &type);
763 error = gfs2_glock_get(sdp, inum.no_addr, &gfs2_inode_glops,
768 error = gfs2_inode_get(gl, &inum, CREATE, &ipp);
770 gfs2_inode_min_init(ipp, type);
775 gfs2_glock_dq_uninit(&d_gh);
777 if (error == -ENOENT)
780 inode = gfs2_ip2v(ipp);
783 return ERR_PTR(-ENOMEM);
786 return ERR_PTR(error);
789 static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino)
791 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
792 struct buffer_head *bh;
793 struct gfs2_inum_range ir;
796 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
799 mutex_lock(&sdp->sd_inum_mutex);
801 error = gfs2_meta_inode_buffer(ip, &bh);
803 mutex_unlock(&sdp->sd_inum_mutex);
808 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
811 *formal_ino = ir.ir_start++;
813 gfs2_trans_add_bh(ip->i_gl, bh, 1);
814 gfs2_inum_range_out(&ir,
815 bh->b_data + sizeof(struct gfs2_dinode));
817 mutex_unlock(&sdp->sd_inum_mutex);
824 mutex_unlock(&sdp->sd_inum_mutex);
830 static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino)
832 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip;
833 struct gfs2_inode *m_ip = sdp->sd_inum_inode->u.generic_ip;
834 struct gfs2_holder gh;
835 struct buffer_head *bh;
836 struct gfs2_inum_range ir;
839 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
843 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
846 mutex_lock(&sdp->sd_inum_mutex);
848 error = gfs2_meta_inode_buffer(ip, &bh);
852 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
855 struct buffer_head *m_bh;
858 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
862 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode));
863 x = y = be64_to_cpu(x);
865 ir.ir_length = GFS2_INUM_QUANTUM;
866 x += GFS2_INUM_QUANTUM;
868 gfs2_consist_inode(m_ip);
870 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
871 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x;
876 *formal_ino = ir.ir_start++;
879 gfs2_trans_add_bh(ip->i_gl, bh, 1);
880 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
886 mutex_unlock(&sdp->sd_inum_mutex);
890 gfs2_glock_dq_uninit(&gh);
895 static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum)
899 error = pick_formal_ino_1(sdp, inum);
903 error = pick_formal_ino_2(sdp, inum);
909 * create_ok - OK to create a new on-disk inode here?
910 * @dip: Directory in which dinode is to be created
911 * @name: Name of new dinode
917 static int create_ok(struct gfs2_inode *dip, struct qstr *name,
922 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
926 /* Don't create entries in an unlinked directory */
927 if (!dip->i_di.di_nlink)
930 error = gfs2_dir_search(dip->i_vnode, name, NULL, NULL);
941 if (dip->i_di.di_entries == (uint32_t)-1)
943 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1)
949 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
950 unsigned int *uid, unsigned int *gid)
952 if (dip->i_sbd->sd_args.ar_suiddir &&
953 (dip->i_di.di_mode & S_ISUID) &&
957 else if (dip->i_di.di_uid != current->fsuid)
959 *uid = dip->i_di.di_uid;
961 *uid = current->fsuid;
963 if (dip->i_di.di_mode & S_ISGID) {
966 *gid = dip->i_di.di_gid;
968 *gid = current->fsgid;
971 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_unlinked *ul)
973 struct gfs2_sbd *sdp = dip->i_sbd;
978 dip->i_alloc.al_requested = RES_DINODE;
979 error = gfs2_inplace_reserve(dip);
983 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
988 ul->ul_ut.ut_inum.no_addr = gfs2_alloc_di(dip);
990 ul->ul_ut.ut_flags = GFS2_UTF_UNINIT;
991 error = gfs2_unlinked_ondisk_add(sdp, ul);
996 gfs2_inplace_release(dip);
1005 * init_dinode - Fill in a new dinode structure
1006 * @dip: the directory this inode is being created in
1007 * @gl: The glock covering the new inode
1008 * @inum: the inode number
1009 * @mode: the file permissions
1015 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1016 struct gfs2_inum *inum, unsigned int mode,
1017 unsigned int uid, unsigned int gid)
1019 struct gfs2_sbd *sdp = dip->i_sbd;
1020 struct gfs2_dinode *di;
1021 struct buffer_head *dibh;
1023 dibh = gfs2_meta_new(gl, inum->no_addr);
1024 gfs2_trans_add_bh(gl, dibh, 1);
1025 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
1026 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1027 di = (struct gfs2_dinode *)dibh->b_data;
1029 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
1030 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
1031 di->di_mode = cpu_to_be32(mode);
1032 di->di_uid = cpu_to_be32(uid);
1033 di->di_gid = cpu_to_be32(gid);
1034 di->di_nlink = cpu_to_be32(0);
1035 di->di_size = cpu_to_be64(0);
1036 di->di_blocks = cpu_to_be64(1);
1037 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
1038 di->di_major = di->di_minor = cpu_to_be32(0);
1039 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
1040 di->__pad[0] = di->__pad[1] = 0;
1041 di->di_flags = cpu_to_be32(0);
1043 if (S_ISREG(mode)) {
1044 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
1045 gfs2_tune_get(sdp, gt_new_files_jdata))
1046 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
1047 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
1048 gfs2_tune_get(sdp, gt_new_files_directio))
1049 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
1050 } else if (S_ISDIR(mode)) {
1051 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1052 GFS2_DIF_INHERIT_DIRECTIO);
1053 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
1054 GFS2_DIF_INHERIT_JDATA);
1058 di->di_height = cpu_to_be32(0);
1061 di->di_depth = cpu_to_be16(0);
1062 di->di_entries = cpu_to_be32(0);
1063 memset(&di->__pad4, 0, sizeof(di->__pad4));
1064 di->di_eattr = cpu_to_be64(0);
1065 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
1070 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1071 unsigned int mode, struct gfs2_unlinked *ul)
1073 struct gfs2_sbd *sdp = dip->i_sbd;
1074 unsigned int uid, gid;
1077 munge_mode_uid_gid(dip, &mode, &uid, &gid);
1079 gfs2_alloc_get(dip);
1081 error = gfs2_quota_lock(dip, uid, gid);
1085 error = gfs2_quota_check(dip, uid, gid);
1089 error = gfs2_trans_begin(sdp, RES_DINODE + RES_UNLINKED +
1094 ul->ul_ut.ut_flags = 0;
1095 error = gfs2_unlinked_ondisk_munge(sdp, ul);
1097 init_dinode(dip, gl, &ul->ul_ut.ut_inum,
1100 gfs2_quota_change(dip, +1, uid, gid);
1102 gfs2_trans_end(sdp);
1105 gfs2_quota_unlock(dip);
1108 gfs2_alloc_put(dip);
1113 static int link_dinode(struct gfs2_inode *dip, struct qstr *name,
1114 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1116 struct gfs2_sbd *sdp = dip->i_sbd;
1117 struct gfs2_alloc *al;
1119 struct buffer_head *dibh;
1122 al = gfs2_alloc_get(dip);
1124 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1128 error = alloc_required = gfs2_diradd_alloc_required(dip->i_vnode, name);
1129 if (alloc_required < 0)
1131 if (alloc_required) {
1132 error = gfs2_quota_check(dip, dip->i_di.di_uid,
1135 goto fail_quota_locks;
1137 al->al_requested = sdp->sd_max_dirres;
1139 error = gfs2_inplace_reserve(dip);
1141 goto fail_quota_locks;
1143 error = gfs2_trans_begin(sdp,
1144 sdp->sd_max_dirres +
1145 al->al_rgd->rd_ri.ri_length +
1146 2 * RES_DINODE + RES_UNLINKED +
1147 RES_STATFS + RES_QUOTA, 0);
1151 error = gfs2_trans_begin(sdp,
1156 goto fail_quota_locks;
1159 error = gfs2_dir_add(dip->i_vnode, name, &ip->i_num, IF2DT(ip->i_di.di_mode));
1161 goto fail_end_trans;
1163 error = gfs2_meta_inode_buffer(ip, &dibh);
1165 goto fail_end_trans;
1166 ip->i_di.di_nlink = 1;
1167 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1168 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1171 error = gfs2_unlinked_ondisk_rm(sdp, ul);
1173 goto fail_end_trans;
1178 gfs2_trans_end(sdp);
1181 if (dip->i_alloc.al_rgd)
1182 gfs2_inplace_release(dip);
1185 gfs2_quota_unlock(dip);
1188 gfs2_alloc_put(dip);
1194 * gfs2_createi - Create a new inode
1195 * @ghs: An array of two holders
1196 * @name: The name of the new file
1197 * @mode: the permissions on the new inode
1199 * @ghs[0] is an initialized holder for the directory
1200 * @ghs[1] is the holder for the inode lock
1202 * If the return value is not NULL, the glocks on both the directory and the new
1203 * file are held. A transaction has been started and an inplace reservation
1209 struct inode *gfs2_createi(struct gfs2_holder *ghs, struct qstr *name,
1212 struct inode *inode;
1213 struct gfs2_inode *dip = ghs->gh_gl->gl_object;
1214 struct gfs2_sbd *sdp = dip->i_sbd;
1215 struct gfs2_unlinked *ul;
1216 struct gfs2_inode *ip;
1219 if (!name->len || name->len > GFS2_FNAMESIZE)
1220 return ERR_PTR(-ENAMETOOLONG);
1222 error = gfs2_unlinked_get(sdp, &ul);
1224 return ERR_PTR(error);
1226 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1227 error = gfs2_glock_nq(ghs);
1231 error = create_ok(dip, name, mode);
1235 error = pick_formal_ino(sdp, &ul->ul_ut.ut_inum.no_formal_ino);
1239 error = alloc_dinode(dip, ul);
1243 if (ul->ul_ut.ut_inum.no_addr < dip->i_num.no_addr) {
1246 error = gfs2_glock_nq_num(sdp,
1247 ul->ul_ut.ut_inum.no_addr,
1249 LM_ST_EXCLUSIVE, GL_SKIP,
1252 gfs2_unlinked_put(sdp, ul);
1253 return ERR_PTR(error);
1256 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1257 error = gfs2_glock_nq(ghs);
1259 gfs2_glock_dq_uninit(ghs + 1);
1260 gfs2_unlinked_put(sdp, ul);
1261 return ERR_PTR(error);
1264 error = create_ok(dip, name, mode);
1268 error = gfs2_glock_nq_num(sdp,
1269 ul->ul_ut.ut_inum.no_addr,
1271 LM_ST_EXCLUSIVE, GL_SKIP,
1277 error = make_dinode(dip, ghs[1].gh_gl, mode, ul);
1281 error = gfs2_inode_get(ghs[1].gh_gl, &ul->ul_ut.ut_inum, CREATE, &ip);
1285 error = gfs2_inode_refresh(ip);
1289 error = gfs2_acl_create(dip, ip);
1293 error = link_dinode(dip, name, ip, ul);
1297 gfs2_unlinked_put(sdp, ul);
1299 inode = gfs2_ip2v(ip);
1302 return ERR_PTR(-ENOMEM);
1309 gfs2_glock_dq_uninit(ghs + 1);
1315 gfs2_unlinked_put(sdp, ul);
1317 return ERR_PTR(error);
1321 * gfs2_unlinki - Unlink a file
1322 * @dip: The inode of the directory
1323 * @name: The name of the file to be unlinked
1324 * @ip: The inode of the file to be removed
1326 * Assumes Glocks on both dip and ip are held.
1331 int gfs2_unlinki(struct gfs2_inode *dip, struct qstr *name,
1332 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1334 struct gfs2_sbd *sdp = dip->i_sbd;
1337 error = gfs2_dir_del(dip, name);
1341 error = gfs2_change_nlink(ip, -1);
1345 /* If this inode is being unlinked from the directory structure,
1346 we need to mark that in the log so that it isn't lost during
1349 if (!ip->i_di.di_nlink) {
1350 ul->ul_ut.ut_inum = ip->i_num;
1351 error = gfs2_unlinked_ondisk_add(sdp, ul);
1353 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1360 * gfs2_rmdiri - Remove a directory
1361 * @dip: The parent directory of the directory to be removed
1362 * @name: The name of the directory to be removed
1363 * @ip: The GFS2 inode of the directory to be removed
1365 * Assumes Glocks on dip and ip are held
1370 int gfs2_rmdiri(struct gfs2_inode *dip, struct qstr *name,
1371 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1373 struct gfs2_sbd *sdp = dip->i_sbd;
1374 struct qstr dotname;
1377 if (ip->i_di.di_entries != 2) {
1378 if (gfs2_consist_inode(ip))
1379 gfs2_dinode_print(&ip->i_di);
1383 error = gfs2_dir_del(dip, name);
1387 error = gfs2_change_nlink(dip, -1);
1391 gfs2_str2qstr(&dotname, ".");
1392 error = gfs2_dir_del(ip, &dotname);
1397 dotname.name = "..";
1398 dotname.hash = gfs2_disk_hash(dotname.name, dotname.len);
1399 error = gfs2_dir_del(ip, &dotname);
1403 error = gfs2_change_nlink(ip, -2);
1407 /* This inode is being unlinked from the directory structure and
1408 we need to mark that in the log so that it isn't lost during
1411 ul->ul_ut.ut_inum = ip->i_num;
1412 error = gfs2_unlinked_ondisk_add(sdp, ul);
1414 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1420 * gfs2_unlink_ok - check to see that a inode is still in a directory
1421 * @dip: the directory
1422 * @name: the name of the file
1425 * Assumes that the lock on (at least) @dip is held.
1427 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1430 int gfs2_unlink_ok(struct gfs2_inode *dip, struct qstr *name,
1431 struct gfs2_inode *ip)
1433 struct gfs2_inum inum;
1437 if (IS_IMMUTABLE(ip->i_vnode) || IS_APPEND(ip->i_vnode))
1440 if ((dip->i_di.di_mode & S_ISVTX) &&
1441 dip->i_di.di_uid != current->fsuid &&
1442 ip->i_di.di_uid != current->fsuid &&
1443 !capable(CAP_FOWNER))
1446 if (IS_APPEND(dip->i_vnode))
1449 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
1453 error = gfs2_dir_search(dip->i_vnode, name, &inum, &type);
1457 if (!gfs2_inum_equal(&inum, &ip->i_num))
1460 if (IF2DT(ip->i_di.di_mode) != type) {
1461 gfs2_consist_inode(dip);
1469 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1473 * Follow @to back to the root and make sure we don't encounter @this
1474 * Assumes we already hold the rename lock.
1479 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1481 struct inode *dir = to->i_vnode;
1482 struct super_block *sb = dir->i_sb;
1487 gfs2_str2qstr(&dotdot, "..");
1492 if (dir == this->i_vnode) {
1496 if (dir == sb->s_root->d_inode) {
1501 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL);
1503 error = PTR_ERR(tmp);
1517 * gfs2_readlinki - return the contents of a symlink
1518 * @ip: the symlink's inode
1519 * @buf: a pointer to the buffer to be filled
1520 * @len: a pointer to the length of @buf
1522 * If @buf is too small, a piece of memory is kmalloc()ed and needs
1523 * to be freed by the caller.
1528 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1530 struct gfs2_holder i_gh;
1531 struct buffer_head *dibh;
1535 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1536 error = gfs2_glock_nq_atime(&i_gh);
1538 gfs2_holder_uninit(&i_gh);
1542 if (!ip->i_di.di_size) {
1543 gfs2_consist_inode(ip);
1548 error = gfs2_meta_inode_buffer(ip, &dibh);
1552 x = ip->i_di.di_size + 1;
1554 *buf = kmalloc(x, GFP_KERNEL);
1561 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1568 gfs2_glock_dq_uninit(&i_gh);
1574 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1575 * conditionally update the inode's atime
1576 * @gh: the holder to acquire
1578 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1579 * Update if the difference between the current time and the inode's current
1580 * atime is greater than an interval specified at mount.
1585 int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1587 struct gfs2_glock *gl = gh->gh_gl;
1588 struct gfs2_sbd *sdp = gl->gl_sbd;
1589 struct gfs2_inode *ip = gl->gl_object;
1590 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1595 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1596 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1597 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1600 state = gh->gh_state;
1601 flags = gh->gh_flags;
1603 error = gfs2_glock_nq(gh);
1607 if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1608 (sdp->sd_vfs->s_flags & MS_RDONLY))
1611 curtime = get_seconds();
1612 if (curtime - ip->i_di.di_atime >= quantum) {
1614 gfs2_holder_reinit(LM_ST_EXCLUSIVE,
1615 gh->gh_flags & ~LM_FLAG_ANY,
1617 error = gfs2_glock_nq(gh);
1621 /* Verify that atime hasn't been updated while we were
1622 trying to get exclusive lock. */
1624 curtime = get_seconds();
1625 if (curtime - ip->i_di.di_atime >= quantum) {
1626 struct buffer_head *dibh;
1628 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1629 if (error == -EROFS)
1634 error = gfs2_meta_inode_buffer(ip, &dibh);
1636 goto fail_end_trans;
1638 ip->i_di.di_atime = curtime;
1640 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1641 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1644 gfs2_trans_end(sdp);
1647 /* If someone else has asked for the glock,
1648 unlock and let them have it. Then reacquire
1649 in the original state. */
1650 if (gfs2_glock_is_blocking(gl)) {
1652 gfs2_holder_reinit(state, flags, gh);
1653 return gfs2_glock_nq(gh);
1660 gfs2_trans_end(sdp);
1669 * glock_compare_atime - Compare two struct gfs2_glock structures for sort
1670 * @arg_a: the first structure
1671 * @arg_b: the second structure
1673 * Returns: 1 if A > B
1678 static int glock_compare_atime(const void *arg_a, const void *arg_b)
1680 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1681 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1682 struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1683 struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1686 if (a->ln_number > b->ln_number)
1688 else if (a->ln_number < b->ln_number)
1691 if (gh_a->gh_state == LM_ST_SHARED &&
1692 gh_b->gh_state == LM_ST_EXCLUSIVE)
1694 else if (gh_a->gh_state == LM_ST_SHARED &&
1695 (gh_b->gh_flags & GL_ATIME))
1703 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an
1705 * @num_gh: the number of structures
1706 * @ghs: an array of struct gfs2_holder structures
1708 * Returns: 0 on success (all glocks acquired),
1709 * errno on failure (no glocks acquired)
1712 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs)
1714 struct gfs2_holder **p;
1722 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1723 if (ghs->gh_flags & GL_ATIME)
1724 error = gfs2_glock_nq_atime(ghs);
1726 error = gfs2_glock_nq(ghs);
1730 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1734 for (x = 0; x < num_gh; x++)
1737 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL);
1739 for (x = 0; x < num_gh; x++) {
1740 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1742 if (p[x]->gh_flags & GL_ATIME)
1743 error = gfs2_glock_nq_atime(p[x]);
1745 error = gfs2_glock_nq(p[x]);
1749 gfs2_glock_dq(p[x]);
1760 * gfs2_try_toss_vnode - See if we can toss a vnode from memory
1763 * Returns: 1 if the vnode was tossed
1766 void gfs2_try_toss_vnode(struct gfs2_inode *ip)
1768 struct inode *inode;
1770 inode = gfs2_ip2v_lookup(ip);
1774 d_prune_aliases(inode);
1776 if (S_ISDIR(ip->i_di.di_mode)) {
1777 struct list_head *head = &inode->i_dentry;
1778 struct dentry *d = NULL;
1780 spin_lock(&dcache_lock);
1781 if (list_empty(head))
1782 spin_unlock(&dcache_lock);
1784 d = list_entry(head->next, struct dentry, d_alias);
1786 spin_unlock(&dcache_lock);
1788 if (have_submounts(d))
1791 shrink_dcache_parent(d);
1793 d_prune_aliases(inode);
1804 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1806 struct buffer_head *dibh;
1809 error = gfs2_meta_inode_buffer(ip, &dibh);
1811 error = inode_setattr(ip->i_vnode, attr);
1812 gfs2_assert_warn(ip->i_sbd, !error);
1813 gfs2_inode_attr_out(ip);
1815 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1816 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1823 * gfs2_setattr_simple -
1827 * Called with a reference on the vnode.
1832 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1836 if (current->journal_info)
1837 return __gfs2_setattr_simple(ip, attr);
1839 error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0);
1843 error = __gfs2_setattr_simple(ip, attr);
1845 gfs2_trans_end(ip->i_sbd);
1850 int gfs2_repermission(struct inode *inode, int mask, struct nameidata *nd)
1852 return permission(inode, mask, nd);