xref: /openbmc/linux/fs/gfs2/inode.c (revision 4161b450)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2011 Red Hat, Inc.  All rights reserved.
4  *
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 version 2.
8  */
9 
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/namei.h>
15 #include <linux/mm.h>
16 #include <linux/xattr.h>
17 #include <linux/posix_acl.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/crc32.h>
20 #include <linux/fiemap.h>
21 #include <linux/security.h>
22 #include <asm/uaccess.h>
23 
24 #include "gfs2.h"
25 #include "incore.h"
26 #include "acl.h"
27 #include "bmap.h"
28 #include "dir.h"
29 #include "xattr.h"
30 #include "glock.h"
31 #include "inode.h"
32 #include "meta_io.h"
33 #include "quota.h"
34 #include "rgrp.h"
35 #include "trans.h"
36 #include "util.h"
37 #include "super.h"
38 #include "glops.h"
39 
40 struct gfs2_skip_data {
41 	u64 no_addr;
42 	int skipped;
43 	int non_block;
44 };
45 
46 static int iget_test(struct inode *inode, void *opaque)
47 {
48 	struct gfs2_inode *ip = GFS2_I(inode);
49 	struct gfs2_skip_data *data = opaque;
50 
51 	if (ip->i_no_addr == data->no_addr) {
52 		if (data->non_block &&
53 		    inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
54 			data->skipped = 1;
55 			return 0;
56 		}
57 		return 1;
58 	}
59 	return 0;
60 }
61 
62 static int iget_set(struct inode *inode, void *opaque)
63 {
64 	struct gfs2_inode *ip = GFS2_I(inode);
65 	struct gfs2_skip_data *data = opaque;
66 
67 	if (data->skipped)
68 		return -ENOENT;
69 	inode->i_ino = (unsigned long)(data->no_addr);
70 	ip->i_no_addr = data->no_addr;
71 	return 0;
72 }
73 
74 struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr, int non_block)
75 {
76 	unsigned long hash = (unsigned long)no_addr;
77 	struct gfs2_skip_data data;
78 
79 	data.no_addr = no_addr;
80 	data.skipped = 0;
81 	data.non_block = non_block;
82 	return ilookup5(sb, hash, iget_test, &data);
83 }
84 
85 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr,
86 			       int non_block)
87 {
88 	struct gfs2_skip_data data;
89 	unsigned long hash = (unsigned long)no_addr;
90 
91 	data.no_addr = no_addr;
92 	data.skipped = 0;
93 	data.non_block = non_block;
94 	return iget5_locked(sb, hash, iget_test, iget_set, &data);
95 }
96 
97 /**
98  * gfs2_set_iop - Sets inode operations
99  * @inode: The inode with correct i_mode filled in
100  *
101  * GFS2 lookup code fills in vfs inode contents based on info obtained
102  * from directory entry inside gfs2_inode_lookup().
103  */
104 
105 static void gfs2_set_iop(struct inode *inode)
106 {
107 	struct gfs2_sbd *sdp = GFS2_SB(inode);
108 	umode_t mode = inode->i_mode;
109 
110 	if (S_ISREG(mode)) {
111 		inode->i_op = &gfs2_file_iops;
112 		if (gfs2_localflocks(sdp))
113 			inode->i_fop = &gfs2_file_fops_nolock;
114 		else
115 			inode->i_fop = &gfs2_file_fops;
116 	} else if (S_ISDIR(mode)) {
117 		inode->i_op = &gfs2_dir_iops;
118 		if (gfs2_localflocks(sdp))
119 			inode->i_fop = &gfs2_dir_fops_nolock;
120 		else
121 			inode->i_fop = &gfs2_dir_fops;
122 	} else if (S_ISLNK(mode)) {
123 		inode->i_op = &gfs2_symlink_iops;
124 	} else {
125 		inode->i_op = &gfs2_file_iops;
126 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
127 	}
128 }
129 
130 /**
131  * gfs2_inode_lookup - Lookup an inode
132  * @sb: The super block
133  * @no_addr: The inode number
134  * @type: The type of the inode
135  * non_block: Can we block on inodes that are being freed?
136  *
137  * Returns: A VFS inode, or an error
138  */
139 
140 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
141 				u64 no_addr, u64 no_formal_ino, int non_block)
142 {
143 	struct inode *inode;
144 	struct gfs2_inode *ip;
145 	struct gfs2_glock *io_gl = NULL;
146 	int error;
147 
148 	inode = gfs2_iget(sb, no_addr, non_block);
149 	ip = GFS2_I(inode);
150 
151 	if (!inode)
152 		return ERR_PTR(-ENOMEM);
153 
154 	if (inode->i_state & I_NEW) {
155 		struct gfs2_sbd *sdp = GFS2_SB(inode);
156 		ip->i_no_formal_ino = no_formal_ino;
157 
158 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
159 		if (unlikely(error))
160 			goto fail;
161 		ip->i_gl->gl_object = ip;
162 
163 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
164 		if (unlikely(error))
165 			goto fail_put;
166 
167 		set_bit(GIF_INVALID, &ip->i_flags);
168 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
169 		if (unlikely(error))
170 			goto fail_iopen;
171 
172 		ip->i_iopen_gh.gh_gl->gl_object = ip;
173 		gfs2_glock_put(io_gl);
174 		io_gl = NULL;
175 
176 		if (type == DT_UNKNOWN) {
177 			/* Inode glock must be locked already */
178 			error = gfs2_inode_refresh(GFS2_I(inode));
179 			if (error)
180 				goto fail_refresh;
181 		} else {
182 			inode->i_mode = DT2IF(type);
183 		}
184 
185 		gfs2_set_iop(inode);
186 		unlock_new_inode(inode);
187 	}
188 
189 	return inode;
190 
191 fail_refresh:
192 	ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
193 	ip->i_iopen_gh.gh_gl->gl_object = NULL;
194 	gfs2_glock_dq_uninit(&ip->i_iopen_gh);
195 fail_iopen:
196 	if (io_gl)
197 		gfs2_glock_put(io_gl);
198 fail_put:
199 	ip->i_gl->gl_object = NULL;
200 	gfs2_glock_put(ip->i_gl);
201 fail:
202 	iget_failed(inode);
203 	return ERR_PTR(error);
204 }
205 
206 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
207 				  u64 *no_formal_ino, unsigned int blktype)
208 {
209 	struct super_block *sb = sdp->sd_vfs;
210 	struct gfs2_holder i_gh;
211 	struct inode *inode = NULL;
212 	int error;
213 
214 	/* Must not read in block until block type is verified */
215 	error = gfs2_glock_nq_num(sdp, no_addr, &gfs2_inode_glops,
216 				  LM_ST_EXCLUSIVE, GL_SKIP, &i_gh);
217 	if (error)
218 		return ERR_PTR(error);
219 
220 	error = gfs2_check_blk_type(sdp, no_addr, blktype);
221 	if (error)
222 		goto fail;
223 
224 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0, 1);
225 	if (IS_ERR(inode))
226 		goto fail;
227 
228 	/* Two extra checks for NFS only */
229 	if (no_formal_ino) {
230 		error = -ESTALE;
231 		if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
232 			goto fail_iput;
233 
234 		error = -EIO;
235 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
236 			goto fail_iput;
237 
238 		error = 0;
239 	}
240 
241 fail:
242 	gfs2_glock_dq_uninit(&i_gh);
243 	return error ? ERR_PTR(error) : inode;
244 fail_iput:
245 	iput(inode);
246 	goto fail;
247 }
248 
249 
250 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
251 {
252 	struct qstr qstr;
253 	struct inode *inode;
254 	gfs2_str2qstr(&qstr, name);
255 	inode = gfs2_lookupi(dip, &qstr, 1);
256 	/* gfs2_lookupi has inconsistent callers: vfs
257 	 * related routines expect NULL for no entry found,
258 	 * gfs2_lookup_simple callers expect ENOENT
259 	 * and do not check for NULL.
260 	 */
261 	if (inode == NULL)
262 		return ERR_PTR(-ENOENT);
263 	else
264 		return inode;
265 }
266 
267 
268 /**
269  * gfs2_lookupi - Look up a filename in a directory and return its inode
270  * @d_gh: An initialized holder for the directory glock
271  * @name: The name of the inode to look for
272  * @is_root: If 1, ignore the caller's permissions
273  * @i_gh: An uninitialized holder for the new inode glock
274  *
275  * This can be called via the VFS filldir function when NFS is doing
276  * a readdirplus and the inode which its intending to stat isn't
277  * already in cache. In this case we must not take the directory glock
278  * again, since the readdir call will have already taken that lock.
279  *
280  * Returns: errno
281  */
282 
283 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
284 			   int is_root)
285 {
286 	struct super_block *sb = dir->i_sb;
287 	struct gfs2_inode *dip = GFS2_I(dir);
288 	struct gfs2_holder d_gh;
289 	int error = 0;
290 	struct inode *inode = NULL;
291 	int unlock = 0;
292 
293 	if (!name->len || name->len > GFS2_FNAMESIZE)
294 		return ERR_PTR(-ENAMETOOLONG);
295 
296 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
297 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
298 	     dir == sb->s_root->d_inode)) {
299 		igrab(dir);
300 		return dir;
301 	}
302 
303 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
304 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
305 		if (error)
306 			return ERR_PTR(error);
307 		unlock = 1;
308 	}
309 
310 	if (!is_root) {
311 		error = gfs2_permission(dir, MAY_EXEC);
312 		if (error)
313 			goto out;
314 	}
315 
316 	inode = gfs2_dir_search(dir, name, false);
317 	if (IS_ERR(inode))
318 		error = PTR_ERR(inode);
319 out:
320 	if (unlock)
321 		gfs2_glock_dq_uninit(&d_gh);
322 	if (error == -ENOENT)
323 		return NULL;
324 	return inode ? inode : ERR_PTR(error);
325 }
326 
327 /**
328  * create_ok - OK to create a new on-disk inode here?
329  * @dip:  Directory in which dinode is to be created
330  * @name:  Name of new dinode
331  * @mode:
332  *
333  * Returns: errno
334  */
335 
336 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
337 		     umode_t mode)
338 {
339 	int error;
340 
341 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
342 	if (error)
343 		return error;
344 
345 	/*  Don't create entries in an unlinked directory  */
346 	if (!dip->i_inode.i_nlink)
347 		return -ENOENT;
348 
349 	if (dip->i_entries == (u32)-1)
350 		return -EFBIG;
351 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
352 		return -EMLINK;
353 
354 	return 0;
355 }
356 
357 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
358 			       struct inode *inode)
359 {
360 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
361 	    (dip->i_inode.i_mode & S_ISUID) &&
362 	    !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
363 		if (S_ISDIR(inode->i_mode))
364 			inode->i_mode |= S_ISUID;
365 		else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
366 			inode->i_mode &= ~07111;
367 		inode->i_uid = dip->i_inode.i_uid;
368 	} else
369 		inode->i_uid = current_fsuid();
370 
371 	if (dip->i_inode.i_mode & S_ISGID) {
372 		if (S_ISDIR(inode->i_mode))
373 			inode->i_mode |= S_ISGID;
374 		inode->i_gid = dip->i_inode.i_gid;
375 	} else
376 		inode->i_gid = current_fsgid();
377 }
378 
379 static int alloc_dinode(struct gfs2_inode *ip, u32 flags, unsigned *dblocks)
380 {
381 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
382 	struct gfs2_alloc_parms ap = { .target = *dblocks, .aflags = flags, };
383 	int error;
384 
385 	error = gfs2_quota_lock_check(ip);
386 	if (error)
387 		goto out;
388 
389 	error = gfs2_inplace_reserve(ip, &ap);
390 	if (error)
391 		goto out_quota;
392 
393 	error = gfs2_trans_begin(sdp, (*dblocks * RES_RG_BIT) + RES_STATFS + RES_QUOTA, 0);
394 	if (error)
395 		goto out_ipreserv;
396 
397 	error = gfs2_alloc_blocks(ip, &ip->i_no_addr, dblocks, 1, &ip->i_generation);
398 	ip->i_no_formal_ino = ip->i_generation;
399 	ip->i_inode.i_ino = ip->i_no_addr;
400 	ip->i_goal = ip->i_no_addr;
401 
402 	gfs2_trans_end(sdp);
403 
404 out_ipreserv:
405 	gfs2_inplace_release(ip);
406 out_quota:
407 	gfs2_quota_unlock(ip);
408 out:
409 	return error;
410 }
411 
412 static void gfs2_init_dir(struct buffer_head *dibh,
413 			  const struct gfs2_inode *parent)
414 {
415 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
416 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
417 
418 	gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
419 	dent->de_inum = di->di_num; /* already GFS2 endian */
420 	dent->de_type = cpu_to_be16(DT_DIR);
421 
422 	dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
423 	gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
424 	gfs2_inum_out(parent, dent);
425 	dent->de_type = cpu_to_be16(DT_DIR);
426 
427 }
428 
429 /**
430  * gfs2_init_xattr - Initialise an xattr block for a new inode
431  * @ip: The inode in question
432  *
433  * This sets up an empty xattr block for a new inode, ready to
434  * take any ACLs, LSM xattrs, etc.
435  */
436 
437 static void gfs2_init_xattr(struct gfs2_inode *ip)
438 {
439 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
440 	struct buffer_head *bh;
441 	struct gfs2_ea_header *ea;
442 
443 	bh = gfs2_meta_new(ip->i_gl, ip->i_eattr);
444 	gfs2_trans_add_meta(ip->i_gl, bh);
445 	gfs2_metatype_set(bh, GFS2_METATYPE_EA, GFS2_FORMAT_EA);
446 	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
447 
448 	ea = GFS2_EA_BH2FIRST(bh);
449 	ea->ea_rec_len = cpu_to_be32(sdp->sd_jbsize);
450 	ea->ea_type = GFS2_EATYPE_UNUSED;
451 	ea->ea_flags = GFS2_EAFLAG_LAST;
452 
453 	brelse(bh);
454 }
455 
456 /**
457  * init_dinode - Fill in a new dinode structure
458  * @dip: The directory this inode is being created in
459  * @ip: The inode
460  * @symname: The symlink destination (if a symlink)
461  * @bhp: The buffer head (returned to caller)
462  *
463  */
464 
465 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
466 			const char *symname)
467 {
468 	struct gfs2_dinode *di;
469 	struct buffer_head *dibh;
470 
471 	dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
472 	gfs2_trans_add_meta(ip->i_gl, dibh);
473 	di = (struct gfs2_dinode *)dibh->b_data;
474 	gfs2_dinode_out(ip, di);
475 
476 	di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
477 	di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
478 	di->__pad1 = 0;
479 	di->__pad2 = 0;
480 	di->__pad3 = 0;
481 	memset(&di->__pad4, 0, sizeof(di->__pad4));
482 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
483 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
484 
485 	switch(ip->i_inode.i_mode & S_IFMT) {
486 	case S_IFDIR:
487 		gfs2_init_dir(dibh, dip);
488 		break;
489 	case S_IFLNK:
490 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
491 		break;
492 	}
493 
494 	set_buffer_uptodate(dibh);
495 	brelse(dibh);
496 }
497 
498 /**
499  * gfs2_trans_da_blocks - Calculate number of blocks to link inode
500  * @dip: The directory we are linking into
501  * @da: The dir add information
502  * @nr_inodes: The number of inodes involved
503  *
504  * This calculate the number of blocks we need to reserve in a
505  * transaction to link @nr_inodes into a directory. In most cases
506  * @nr_inodes will be 2 (the directory plus the inode being linked in)
507  * but in case of rename, 4 may be required.
508  *
509  * Returns: Number of blocks
510  */
511 
512 static unsigned gfs2_trans_da_blks(const struct gfs2_inode *dip,
513 				   const struct gfs2_diradd *da,
514 				   unsigned nr_inodes)
515 {
516 	return da->nr_blocks + gfs2_rg_blocks(dip, da->nr_blocks) +
517 	       (nr_inodes * RES_DINODE) + RES_QUOTA + RES_STATFS;
518 }
519 
520 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
521 		       struct gfs2_inode *ip, struct gfs2_diradd *da)
522 {
523 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
524 	struct gfs2_alloc_parms ap = { .target = da->nr_blocks, };
525 	int error;
526 
527 	if (da->nr_blocks) {
528 		error = gfs2_quota_lock_check(dip);
529 		if (error)
530 			goto fail_quota_locks;
531 
532 		error = gfs2_inplace_reserve(dip, &ap);
533 		if (error)
534 			goto fail_quota_locks;
535 
536 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, da, 2), 0);
537 		if (error)
538 			goto fail_ipreserv;
539 	} else {
540 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
541 		if (error)
542 			goto fail_quota_locks;
543 	}
544 
545 	error = gfs2_dir_add(&dip->i_inode, name, ip, da);
546 	if (error)
547 		goto fail_end_trans;
548 
549 fail_end_trans:
550 	gfs2_trans_end(sdp);
551 fail_ipreserv:
552 	gfs2_inplace_release(dip);
553 fail_quota_locks:
554 	gfs2_quota_unlock(dip);
555 	return error;
556 }
557 
558 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
559 		    void *fs_info)
560 {
561 	const struct xattr *xattr;
562 	int err = 0;
563 
564 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
565 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
566 				       xattr->value_len, 0,
567 				       GFS2_EATYPE_SECURITY);
568 		if (err < 0)
569 			break;
570 	}
571 	return err;
572 }
573 
574 /**
575  * gfs2_create_inode - Create a new inode
576  * @dir: The parent directory
577  * @dentry: The new dentry
578  * @file: If non-NULL, the file which is being opened
579  * @mode: The permissions on the new inode
580  * @dev: For device nodes, this is the device number
581  * @symname: For symlinks, this is the link destination
582  * @size: The initial size of the inode (ignored for directories)
583  *
584  * Returns: 0 on success, or error code
585  */
586 
587 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
588 			     struct file *file,
589 			     umode_t mode, dev_t dev, const char *symname,
590 			     unsigned int size, int excl, int *opened)
591 {
592 	const struct qstr *name = &dentry->d_name;
593 	struct posix_acl *default_acl, *acl;
594 	struct gfs2_holder ghs[2];
595 	struct inode *inode = NULL;
596 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
597 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
598 	struct gfs2_glock *io_gl;
599 	int error, free_vfs_inode = 0;
600 	u32 aflags = 0;
601 	unsigned blocks = 1;
602 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
603 
604 	if (!name->len || name->len > GFS2_FNAMESIZE)
605 		return -ENAMETOOLONG;
606 
607 	error = gfs2_rs_alloc(dip);
608 	if (error)
609 		return error;
610 
611 	error = gfs2_rindex_update(sdp);
612 	if (error)
613 		return error;
614 
615 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
616 	if (error)
617 		goto fail;
618 
619 	error = create_ok(dip, name, mode);
620 	if (error)
621 		goto fail_gunlock;
622 
623 	inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
624 	error = PTR_ERR(inode);
625 	if (!IS_ERR(inode)) {
626 		if (S_ISDIR(inode->i_mode)) {
627 			iput(inode);
628 			inode = ERR_PTR(-EISDIR);
629 			goto fail_gunlock;
630 		}
631 		d_instantiate(dentry, inode);
632 		error = 0;
633 		if (file) {
634 			if (S_ISREG(inode->i_mode))
635 				error = finish_open(file, dentry, gfs2_open_common, opened);
636 			else
637 				error = finish_no_open(file, NULL);
638 		}
639 		gfs2_glock_dq_uninit(ghs);
640 		return error;
641 	} else if (error != -ENOENT) {
642 		goto fail_gunlock;
643 	}
644 
645 	error = gfs2_diradd_alloc_required(dir, name, &da);
646 	if (error < 0)
647 		goto fail_gunlock;
648 
649 	inode = new_inode(sdp->sd_vfs);
650 	error = -ENOMEM;
651 	if (!inode)
652 		goto fail_gunlock;
653 
654 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
655 	if (error)
656 		goto fail_free_vfs_inode;
657 
658 	ip = GFS2_I(inode);
659 	error = gfs2_rs_alloc(ip);
660 	if (error)
661 		goto fail_free_acls;
662 
663 	inode->i_mode = mode;
664 	set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
665 	inode->i_rdev = dev;
666 	inode->i_size = size;
667 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
668 	gfs2_set_inode_blocks(inode, 1);
669 	munge_mode_uid_gid(dip, inode);
670 	check_and_update_goal(dip);
671 	ip->i_goal = dip->i_goal;
672 	ip->i_diskflags = 0;
673 	ip->i_eattr = 0;
674 	ip->i_height = 0;
675 	ip->i_depth = 0;
676 	ip->i_entries = 0;
677 
678 	switch(mode & S_IFMT) {
679 	case S_IFREG:
680 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
681 		    gfs2_tune_get(sdp, gt_new_files_jdata))
682 			ip->i_diskflags |= GFS2_DIF_JDATA;
683 		gfs2_set_aops(inode);
684 		break;
685 	case S_IFDIR:
686 		ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
687 		ip->i_diskflags |= GFS2_DIF_JDATA;
688 		ip->i_entries = 2;
689 		break;
690 	}
691 	gfs2_set_inode_flags(inode);
692 
693 	if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
694 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
695 		aflags |= GFS2_AF_ORLOV;
696 
697 	if (default_acl || acl)
698 		blocks++;
699 
700 	error = alloc_dinode(ip, aflags, &blocks);
701 	if (error)
702 		goto fail_free_inode;
703 
704 	gfs2_set_inode_blocks(inode, blocks);
705 
706 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
707 	if (error)
708 		goto fail_free_inode;
709 
710 	ip->i_gl->gl_object = ip;
711 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
712 	if (error)
713 		goto fail_free_inode;
714 
715 	error = gfs2_trans_begin(sdp, blocks, 0);
716 	if (error)
717 		goto fail_gunlock2;
718 
719 	if (blocks > 1) {
720 		ip->i_eattr = ip->i_no_addr + 1;
721 		gfs2_init_xattr(ip);
722 	}
723 	init_dinode(dip, ip, symname);
724 	gfs2_trans_end(sdp);
725 
726 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
727 	if (error)
728 		goto fail_gunlock2;
729 
730 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
731 	if (error)
732 		goto fail_gunlock2;
733 
734 	ip->i_iopen_gh.gh_gl->gl_object = ip;
735 	gfs2_glock_put(io_gl);
736 	gfs2_set_iop(inode);
737 	insert_inode_hash(inode);
738 
739 	if (default_acl) {
740 		error = gfs2_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
741 		posix_acl_release(default_acl);
742 	}
743 	if (acl) {
744 		if (!error)
745 			error = gfs2_set_acl(inode, acl, ACL_TYPE_ACCESS);
746 		posix_acl_release(acl);
747 	}
748 
749 	if (error)
750 		goto fail_gunlock3;
751 
752 	error = security_inode_init_security(&ip->i_inode, &dip->i_inode, name,
753 					     &gfs2_initxattrs, NULL);
754 	if (error)
755 		goto fail_gunlock3;
756 
757 	error = link_dinode(dip, name, ip, &da);
758 	if (error)
759 		goto fail_gunlock3;
760 
761 	mark_inode_dirty(inode);
762 	d_instantiate(dentry, inode);
763 	if (file) {
764 		*opened |= FILE_CREATED;
765 		error = finish_open(file, dentry, gfs2_open_common, opened);
766 	}
767 	gfs2_glock_dq_uninit(ghs);
768 	gfs2_glock_dq_uninit(ghs + 1);
769 	return error;
770 
771 fail_gunlock3:
772 	gfs2_glock_dq_uninit(ghs + 1);
773 	if (ip->i_gl)
774 		gfs2_glock_put(ip->i_gl);
775 	goto fail_gunlock;
776 
777 fail_gunlock2:
778 	gfs2_glock_dq_uninit(ghs + 1);
779 fail_free_inode:
780 	if (ip->i_gl)
781 		gfs2_glock_put(ip->i_gl);
782 	gfs2_rs_delete(ip, NULL);
783 fail_free_acls:
784 	if (default_acl)
785 		posix_acl_release(default_acl);
786 	if (acl)
787 		posix_acl_release(acl);
788 fail_free_vfs_inode:
789 	free_vfs_inode = 1;
790 fail_gunlock:
791 	gfs2_dir_no_add(&da);
792 	gfs2_glock_dq_uninit(ghs);
793 	if (inode && !IS_ERR(inode)) {
794 		clear_nlink(inode);
795 		if (!free_vfs_inode)
796 			mark_inode_dirty(inode);
797 		set_bit(free_vfs_inode ? GIF_FREE_VFS_INODE : GIF_ALLOC_FAILED,
798 			&GFS2_I(inode)->i_flags);
799 		iput(inode);
800 	}
801 fail:
802 	return error;
803 }
804 
805 /**
806  * gfs2_create - Create a file
807  * @dir: The directory in which to create the file
808  * @dentry: The dentry of the new file
809  * @mode: The mode of the new file
810  *
811  * Returns: errno
812  */
813 
814 static int gfs2_create(struct inode *dir, struct dentry *dentry,
815 		       umode_t mode, bool excl)
816 {
817 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl, NULL);
818 }
819 
820 /**
821  * __gfs2_lookup - Look up a filename in a directory and return its inode
822  * @dir: The directory inode
823  * @dentry: The dentry of the new inode
824  * @file: File to be opened
825  * @opened: atomic_open flags
826  *
827  *
828  * Returns: errno
829  */
830 
831 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
832 				    struct file *file, int *opened)
833 {
834 	struct inode *inode;
835 	struct dentry *d;
836 	struct gfs2_holder gh;
837 	struct gfs2_glock *gl;
838 	int error;
839 
840 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
841 	if (inode == NULL) {
842 		d_add(dentry, NULL);
843 		return NULL;
844 	}
845 	if (IS_ERR(inode))
846 		return ERR_CAST(inode);
847 
848 	gl = GFS2_I(inode)->i_gl;
849 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
850 	if (error) {
851 		iput(inode);
852 		return ERR_PTR(error);
853 	}
854 
855 	d = d_splice_alias(inode, dentry);
856 	if (IS_ERR(d)) {
857 		gfs2_glock_dq_uninit(&gh);
858 		return d;
859 	}
860 	if (file && S_ISREG(inode->i_mode))
861 		error = finish_open(file, dentry, gfs2_open_common, opened);
862 
863 	gfs2_glock_dq_uninit(&gh);
864 	if (error) {
865 		dput(d);
866 		return ERR_PTR(error);
867 	}
868 	return d;
869 }
870 
871 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
872 				  unsigned flags)
873 {
874 	return __gfs2_lookup(dir, dentry, NULL, NULL);
875 }
876 
877 /**
878  * gfs2_link - Link to a file
879  * @old_dentry: The inode to link
880  * @dir: Add link to this directory
881  * @dentry: The name of the link
882  *
883  * Link the inode in "old_dentry" into the directory "dir" with the
884  * name in "dentry".
885  *
886  * Returns: errno
887  */
888 
889 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
890 		     struct dentry *dentry)
891 {
892 	struct gfs2_inode *dip = GFS2_I(dir);
893 	struct gfs2_sbd *sdp = GFS2_SB(dir);
894 	struct inode *inode = old_dentry->d_inode;
895 	struct gfs2_inode *ip = GFS2_I(inode);
896 	struct gfs2_holder ghs[2];
897 	struct buffer_head *dibh;
898 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
899 	int error;
900 
901 	if (S_ISDIR(inode->i_mode))
902 		return -EPERM;
903 
904 	error = gfs2_rs_alloc(dip);
905 	if (error)
906 		return error;
907 
908 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
909 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
910 
911 	error = gfs2_glock_nq(ghs); /* parent */
912 	if (error)
913 		goto out_parent;
914 
915 	error = gfs2_glock_nq(ghs + 1); /* child */
916 	if (error)
917 		goto out_child;
918 
919 	error = -ENOENT;
920 	if (inode->i_nlink == 0)
921 		goto out_gunlock;
922 
923 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
924 	if (error)
925 		goto out_gunlock;
926 
927 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
928 	switch (error) {
929 	case -ENOENT:
930 		break;
931 	case 0:
932 		error = -EEXIST;
933 	default:
934 		goto out_gunlock;
935 	}
936 
937 	error = -EINVAL;
938 	if (!dip->i_inode.i_nlink)
939 		goto out_gunlock;
940 	error = -EFBIG;
941 	if (dip->i_entries == (u32)-1)
942 		goto out_gunlock;
943 	error = -EPERM;
944 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
945 		goto out_gunlock;
946 	error = -EINVAL;
947 	if (!ip->i_inode.i_nlink)
948 		goto out_gunlock;
949 	error = -EMLINK;
950 	if (ip->i_inode.i_nlink == (u32)-1)
951 		goto out_gunlock;
952 
953 	error = gfs2_diradd_alloc_required(dir, &dentry->d_name, &da);
954 	if (error < 0)
955 		goto out_gunlock;
956 
957 	if (da.nr_blocks) {
958 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
959 		error = gfs2_quota_lock_check(dip);
960 		if (error)
961 			goto out_gunlock;
962 
963 		error = gfs2_inplace_reserve(dip, &ap);
964 		if (error)
965 			goto out_gunlock_q;
966 
967 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, &da, 2), 0);
968 		if (error)
969 			goto out_ipres;
970 	} else {
971 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
972 		if (error)
973 			goto out_ipres;
974 	}
975 
976 	error = gfs2_meta_inode_buffer(ip, &dibh);
977 	if (error)
978 		goto out_end_trans;
979 
980 	error = gfs2_dir_add(dir, &dentry->d_name, ip, &da);
981 	if (error)
982 		goto out_brelse;
983 
984 	gfs2_trans_add_meta(ip->i_gl, dibh);
985 	inc_nlink(&ip->i_inode);
986 	ip->i_inode.i_ctime = CURRENT_TIME;
987 	ihold(inode);
988 	d_instantiate(dentry, inode);
989 	mark_inode_dirty(inode);
990 
991 out_brelse:
992 	brelse(dibh);
993 out_end_trans:
994 	gfs2_trans_end(sdp);
995 out_ipres:
996 	if (da.nr_blocks)
997 		gfs2_inplace_release(dip);
998 out_gunlock_q:
999 	if (da.nr_blocks)
1000 		gfs2_quota_unlock(dip);
1001 out_gunlock:
1002 	gfs2_dir_no_add(&da);
1003 	gfs2_glock_dq(ghs + 1);
1004 out_child:
1005 	gfs2_glock_dq(ghs);
1006 out_parent:
1007 	gfs2_holder_uninit(ghs);
1008 	gfs2_holder_uninit(ghs + 1);
1009 	return error;
1010 }
1011 
1012 /*
1013  * gfs2_unlink_ok - check to see that a inode is still in a directory
1014  * @dip: the directory
1015  * @name: the name of the file
1016  * @ip: the inode
1017  *
1018  * Assumes that the lock on (at least) @dip is held.
1019  *
1020  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1021  */
1022 
1023 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
1024 			  const struct gfs2_inode *ip)
1025 {
1026 	int error;
1027 
1028 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
1029 		return -EPERM;
1030 
1031 	if ((dip->i_inode.i_mode & S_ISVTX) &&
1032 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
1033 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
1034 		return -EPERM;
1035 
1036 	if (IS_APPEND(&dip->i_inode))
1037 		return -EPERM;
1038 
1039 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
1040 	if (error)
1041 		return error;
1042 
1043 	return gfs2_dir_check(&dip->i_inode, name, ip);
1044 }
1045 
1046 /**
1047  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1048  * @dip: The parent directory
1049  * @name: The name of the entry in the parent directory
1050  * @inode: The inode to be removed
1051  *
1052  * Called with all the locks and in a transaction. This will only be
1053  * called for a directory after it has been checked to ensure it is empty.
1054  *
1055  * Returns: 0 on success, or an error
1056  */
1057 
1058 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1059 			     const struct dentry *dentry)
1060 {
1061 	struct inode *inode = dentry->d_inode;
1062 	struct gfs2_inode *ip = GFS2_I(inode);
1063 	int error;
1064 
1065 	error = gfs2_dir_del(dip, dentry);
1066 	if (error)
1067 		return error;
1068 
1069 	ip->i_entries = 0;
1070 	inode->i_ctime = CURRENT_TIME;
1071 	if (S_ISDIR(inode->i_mode))
1072 		clear_nlink(inode);
1073 	else
1074 		drop_nlink(inode);
1075 	mark_inode_dirty(inode);
1076 	if (inode->i_nlink == 0)
1077 		gfs2_unlink_di(inode);
1078 	return 0;
1079 }
1080 
1081 
1082 /**
1083  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1084  * @dir: The inode of the directory containing the inode to unlink
1085  * @dentry: The file itself
1086  *
1087  * This routine uses the type of the inode as a flag to figure out
1088  * whether this is an unlink or an rmdir.
1089  *
1090  * Returns: errno
1091  */
1092 
1093 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1094 {
1095 	struct gfs2_inode *dip = GFS2_I(dir);
1096 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1097 	struct inode *inode = dentry->d_inode;
1098 	struct gfs2_inode *ip = GFS2_I(inode);
1099 	struct gfs2_holder ghs[3];
1100 	struct gfs2_rgrpd *rgd;
1101 	int error;
1102 
1103 	error = gfs2_rindex_update(sdp);
1104 	if (error)
1105 		return error;
1106 
1107 	error = -EROFS;
1108 
1109 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1110 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1111 
1112 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1113 	if (!rgd)
1114 		goto out_inodes;
1115 
1116 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1117 
1118 
1119 	error = gfs2_glock_nq(ghs); /* parent */
1120 	if (error)
1121 		goto out_parent;
1122 
1123 	error = gfs2_glock_nq(ghs + 1); /* child */
1124 	if (error)
1125 		goto out_child;
1126 
1127 	error = -ENOENT;
1128 	if (inode->i_nlink == 0)
1129 		goto out_rgrp;
1130 
1131 	if (S_ISDIR(inode->i_mode)) {
1132 		error = -ENOTEMPTY;
1133 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1134 			goto out_rgrp;
1135 	}
1136 
1137 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1138 	if (error)
1139 		goto out_rgrp;
1140 
1141 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1142 	if (error)
1143 		goto out_gunlock;
1144 
1145 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1146 	if (error)
1147 		goto out_end_trans;
1148 
1149 	error = gfs2_unlink_inode(dip, dentry);
1150 
1151 out_end_trans:
1152 	gfs2_trans_end(sdp);
1153 out_gunlock:
1154 	gfs2_glock_dq(ghs + 2);
1155 out_rgrp:
1156 	gfs2_glock_dq(ghs + 1);
1157 out_child:
1158 	gfs2_glock_dq(ghs);
1159 out_parent:
1160 	gfs2_holder_uninit(ghs + 2);
1161 out_inodes:
1162 	gfs2_holder_uninit(ghs + 1);
1163 	gfs2_holder_uninit(ghs);
1164 	return error;
1165 }
1166 
1167 /**
1168  * gfs2_symlink - Create a symlink
1169  * @dir: The directory to create the symlink in
1170  * @dentry: The dentry to put the symlink in
1171  * @symname: The thing which the link points to
1172  *
1173  * Returns: errno
1174  */
1175 
1176 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1177 			const char *symname)
1178 {
1179 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1180 	unsigned int size;
1181 
1182 	size = strlen(symname);
1183 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1184 		return -ENAMETOOLONG;
1185 
1186 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0, NULL);
1187 }
1188 
1189 /**
1190  * gfs2_mkdir - Make a directory
1191  * @dir: The parent directory of the new one
1192  * @dentry: The dentry of the new directory
1193  * @mode: The mode of the new directory
1194  *
1195  * Returns: errno
1196  */
1197 
1198 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1199 {
1200 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1201 	unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1202 	return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0, NULL);
1203 }
1204 
1205 /**
1206  * gfs2_mknod - Make a special file
1207  * @dir: The directory in which the special file will reside
1208  * @dentry: The dentry of the special file
1209  * @mode: The mode of the special file
1210  * @dev: The device specification of the special file
1211  *
1212  */
1213 
1214 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1215 		      dev_t dev)
1216 {
1217 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0, NULL);
1218 }
1219 
1220 /**
1221  * gfs2_atomic_open - Atomically open a file
1222  * @dir: The directory
1223  * @dentry: The proposed new entry
1224  * @file: The proposed new struct file
1225  * @flags: open flags
1226  * @mode: File mode
1227  * @opened: Flag to say whether the file has been opened or not
1228  *
1229  * Returns: error code or 0 for success
1230  */
1231 
1232 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1233                             struct file *file, unsigned flags,
1234                             umode_t mode, int *opened)
1235 {
1236 	struct dentry *d;
1237 	bool excl = !!(flags & O_EXCL);
1238 
1239 	if (!d_unhashed(dentry))
1240 		goto skip_lookup;
1241 
1242 	d = __gfs2_lookup(dir, dentry, file, opened);
1243 	if (IS_ERR(d))
1244 		return PTR_ERR(d);
1245 	if (d != NULL)
1246 		dentry = d;
1247 	if (dentry->d_inode) {
1248 		if (!(*opened & FILE_OPENED))
1249 			return finish_no_open(file, d);
1250 		dput(d);
1251 		return 0;
1252 	}
1253 
1254 	BUG_ON(d != NULL);
1255 
1256 skip_lookup:
1257 	if (!(flags & O_CREAT))
1258 		return -ENOENT;
1259 
1260 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl, opened);
1261 }
1262 
1263 /*
1264  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1265  * @this: move this
1266  * @to: to here
1267  *
1268  * Follow @to back to the root and make sure we don't encounter @this
1269  * Assumes we already hold the rename lock.
1270  *
1271  * Returns: errno
1272  */
1273 
1274 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1275 {
1276 	struct inode *dir = &to->i_inode;
1277 	struct super_block *sb = dir->i_sb;
1278 	struct inode *tmp;
1279 	int error = 0;
1280 
1281 	igrab(dir);
1282 
1283 	for (;;) {
1284 		if (dir == &this->i_inode) {
1285 			error = -EINVAL;
1286 			break;
1287 		}
1288 		if (dir == sb->s_root->d_inode) {
1289 			error = 0;
1290 			break;
1291 		}
1292 
1293 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1294 		if (!tmp) {
1295 			error = -ENOENT;
1296 			break;
1297 		}
1298 		if (IS_ERR(tmp)) {
1299 			error = PTR_ERR(tmp);
1300 			break;
1301 		}
1302 
1303 		iput(dir);
1304 		dir = tmp;
1305 	}
1306 
1307 	iput(dir);
1308 
1309 	return error;
1310 }
1311 
1312 /**
1313  * gfs2_rename - Rename a file
1314  * @odir: Parent directory of old file name
1315  * @odentry: The old dentry of the file
1316  * @ndir: Parent directory of new file name
1317  * @ndentry: The new dentry of the file
1318  *
1319  * Returns: errno
1320  */
1321 
1322 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1323 		       struct inode *ndir, struct dentry *ndentry)
1324 {
1325 	struct gfs2_inode *odip = GFS2_I(odir);
1326 	struct gfs2_inode *ndip = GFS2_I(ndir);
1327 	struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1328 	struct gfs2_inode *nip = NULL;
1329 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1330 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1331 	struct gfs2_rgrpd *nrgd;
1332 	unsigned int num_gh;
1333 	int dir_rename = 0;
1334 	struct gfs2_diradd da = { .nr_blocks = 0, .save_loc = 0, };
1335 	unsigned int x;
1336 	int error;
1337 
1338 	if (ndentry->d_inode) {
1339 		nip = GFS2_I(ndentry->d_inode);
1340 		if (ip == nip)
1341 			return 0;
1342 	}
1343 
1344 	error = gfs2_rindex_update(sdp);
1345 	if (error)
1346 		return error;
1347 
1348 	error = gfs2_rs_alloc(ndip);
1349 	if (error)
1350 		return error;
1351 
1352 	if (odip != ndip) {
1353 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1354 					   0, &r_gh);
1355 		if (error)
1356 			goto out;
1357 
1358 		if (S_ISDIR(ip->i_inode.i_mode)) {
1359 			dir_rename = 1;
1360 			/* don't move a dirctory into it's subdir */
1361 			error = gfs2_ok_to_move(ip, ndip);
1362 			if (error)
1363 				goto out_gunlock_r;
1364 		}
1365 	}
1366 
1367 	num_gh = 1;
1368 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1369 	if (odip != ndip) {
1370 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1371 		num_gh++;
1372 	}
1373 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1374 	num_gh++;
1375 
1376 	if (nip) {
1377 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1378 		num_gh++;
1379 		/* grab the resource lock for unlink flag twiddling
1380 		 * this is the case of the target file already existing
1381 		 * so we unlink before doing the rename
1382 		 */
1383 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1384 		if (nrgd)
1385 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1386 	}
1387 
1388 	for (x = 0; x < num_gh; x++) {
1389 		error = gfs2_glock_nq(ghs + x);
1390 		if (error)
1391 			goto out_gunlock;
1392 	}
1393 
1394 	error = -ENOENT;
1395 	if (ip->i_inode.i_nlink == 0)
1396 		goto out_gunlock;
1397 
1398 	/* Check out the old directory */
1399 
1400 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1401 	if (error)
1402 		goto out_gunlock;
1403 
1404 	/* Check out the new directory */
1405 
1406 	if (nip) {
1407 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1408 		if (error)
1409 			goto out_gunlock;
1410 
1411 		if (nip->i_inode.i_nlink == 0) {
1412 			error = -EAGAIN;
1413 			goto out_gunlock;
1414 		}
1415 
1416 		if (S_ISDIR(nip->i_inode.i_mode)) {
1417 			if (nip->i_entries < 2) {
1418 				gfs2_consist_inode(nip);
1419 				error = -EIO;
1420 				goto out_gunlock;
1421 			}
1422 			if (nip->i_entries > 2) {
1423 				error = -ENOTEMPTY;
1424 				goto out_gunlock;
1425 			}
1426 		}
1427 	} else {
1428 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1429 		if (error)
1430 			goto out_gunlock;
1431 
1432 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1433 		switch (error) {
1434 		case -ENOENT:
1435 			error = 0;
1436 			break;
1437 		case 0:
1438 			error = -EEXIST;
1439 		default:
1440 			goto out_gunlock;
1441 		};
1442 
1443 		if (odip != ndip) {
1444 			if (!ndip->i_inode.i_nlink) {
1445 				error = -ENOENT;
1446 				goto out_gunlock;
1447 			}
1448 			if (ndip->i_entries == (u32)-1) {
1449 				error = -EFBIG;
1450 				goto out_gunlock;
1451 			}
1452 			if (S_ISDIR(ip->i_inode.i_mode) &&
1453 			    ndip->i_inode.i_nlink == (u32)-1) {
1454 				error = -EMLINK;
1455 				goto out_gunlock;
1456 			}
1457 		}
1458 	}
1459 
1460 	/* Check out the dir to be renamed */
1461 
1462 	if (dir_rename) {
1463 		error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1464 		if (error)
1465 			goto out_gunlock;
1466 	}
1467 
1468 	if (nip == NULL) {
1469 		error = gfs2_diradd_alloc_required(ndir, &ndentry->d_name, &da);
1470 		if (error)
1471 			goto out_gunlock;
1472 	}
1473 
1474 	if (da.nr_blocks) {
1475 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1476 		error = gfs2_quota_lock_check(ndip);
1477 		if (error)
1478 			goto out_gunlock;
1479 
1480 		error = gfs2_inplace_reserve(ndip, &ap);
1481 		if (error)
1482 			goto out_gunlock_q;
1483 
1484 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(ndip, &da, 4) +
1485 					 4 * RES_LEAF + 4, 0);
1486 		if (error)
1487 			goto out_ipreserv;
1488 	} else {
1489 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1490 					 5 * RES_LEAF + 4, 0);
1491 		if (error)
1492 			goto out_gunlock;
1493 	}
1494 
1495 	/* Remove the target file, if it exists */
1496 
1497 	if (nip)
1498 		error = gfs2_unlink_inode(ndip, ndentry);
1499 
1500 	if (dir_rename) {
1501 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1502 		if (error)
1503 			goto out_end_trans;
1504 	} else {
1505 		struct buffer_head *dibh;
1506 		error = gfs2_meta_inode_buffer(ip, &dibh);
1507 		if (error)
1508 			goto out_end_trans;
1509 		ip->i_inode.i_ctime = CURRENT_TIME;
1510 		gfs2_trans_add_meta(ip->i_gl, dibh);
1511 		gfs2_dinode_out(ip, dibh->b_data);
1512 		brelse(dibh);
1513 	}
1514 
1515 	error = gfs2_dir_del(odip, odentry);
1516 	if (error)
1517 		goto out_end_trans;
1518 
1519 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip, &da);
1520 	if (error)
1521 		goto out_end_trans;
1522 
1523 out_end_trans:
1524 	gfs2_trans_end(sdp);
1525 out_ipreserv:
1526 	if (da.nr_blocks)
1527 		gfs2_inplace_release(ndip);
1528 out_gunlock_q:
1529 	if (da.nr_blocks)
1530 		gfs2_quota_unlock(ndip);
1531 out_gunlock:
1532 	gfs2_dir_no_add(&da);
1533 	while (x--) {
1534 		gfs2_glock_dq(ghs + x);
1535 		gfs2_holder_uninit(ghs + x);
1536 	}
1537 out_gunlock_r:
1538 	if (r_gh.gh_gl)
1539 		gfs2_glock_dq_uninit(&r_gh);
1540 out:
1541 	return error;
1542 }
1543 
1544 /**
1545  * gfs2_follow_link - Follow a symbolic link
1546  * @dentry: The dentry of the link
1547  * @nd: Data that we pass to vfs_follow_link()
1548  *
1549  * This can handle symlinks of any size.
1550  *
1551  * Returns: 0 on success or error code
1552  */
1553 
1554 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1555 {
1556 	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1557 	struct gfs2_holder i_gh;
1558 	struct buffer_head *dibh;
1559 	unsigned int size;
1560 	char *buf;
1561 	int error;
1562 
1563 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1564 	error = gfs2_glock_nq(&i_gh);
1565 	if (error) {
1566 		gfs2_holder_uninit(&i_gh);
1567 		nd_set_link(nd, ERR_PTR(error));
1568 		return NULL;
1569 	}
1570 
1571 	size = (unsigned int)i_size_read(&ip->i_inode);
1572 	if (size == 0) {
1573 		gfs2_consist_inode(ip);
1574 		buf = ERR_PTR(-EIO);
1575 		goto out;
1576 	}
1577 
1578 	error = gfs2_meta_inode_buffer(ip, &dibh);
1579 	if (error) {
1580 		buf = ERR_PTR(error);
1581 		goto out;
1582 	}
1583 
1584 	buf = kzalloc(size + 1, GFP_NOFS);
1585 	if (!buf)
1586 		buf = ERR_PTR(-ENOMEM);
1587 	else
1588 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1589 	brelse(dibh);
1590 out:
1591 	gfs2_glock_dq_uninit(&i_gh);
1592 	nd_set_link(nd, buf);
1593 	return NULL;
1594 }
1595 
1596 /**
1597  * gfs2_permission -
1598  * @inode: The inode
1599  * @mask: The mask to be tested
1600  * @flags: Indicates whether this is an RCU path walk or not
1601  *
1602  * This may be called from the VFS directly, or from within GFS2 with the
1603  * inode locked, so we look to see if the glock is already locked and only
1604  * lock the glock if its not already been done.
1605  *
1606  * Returns: errno
1607  */
1608 
1609 int gfs2_permission(struct inode *inode, int mask)
1610 {
1611 	struct gfs2_inode *ip;
1612 	struct gfs2_holder i_gh;
1613 	int error;
1614 	int unlock = 0;
1615 
1616 
1617 	ip = GFS2_I(inode);
1618 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1619 		if (mask & MAY_NOT_BLOCK)
1620 			return -ECHILD;
1621 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1622 		if (error)
1623 			return error;
1624 		unlock = 1;
1625 	}
1626 
1627 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1628 		error = -EACCES;
1629 	else
1630 		error = generic_permission(inode, mask);
1631 	if (unlock)
1632 		gfs2_glock_dq_uninit(&i_gh);
1633 
1634 	return error;
1635 }
1636 
1637 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1638 {
1639 	setattr_copy(inode, attr);
1640 	mark_inode_dirty(inode);
1641 	return 0;
1642 }
1643 
1644 /**
1645  * gfs2_setattr_simple -
1646  * @ip:
1647  * @attr:
1648  *
1649  * Returns: errno
1650  */
1651 
1652 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1653 {
1654 	int error;
1655 
1656 	if (current->journal_info)
1657 		return __gfs2_setattr_simple(inode, attr);
1658 
1659 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1660 	if (error)
1661 		return error;
1662 
1663 	error = __gfs2_setattr_simple(inode, attr);
1664 	gfs2_trans_end(GFS2_SB(inode));
1665 	return error;
1666 }
1667 
1668 static int setattr_chown(struct inode *inode, struct iattr *attr)
1669 {
1670 	struct gfs2_inode *ip = GFS2_I(inode);
1671 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1672 	kuid_t ouid, nuid;
1673 	kgid_t ogid, ngid;
1674 	int error;
1675 
1676 	ouid = inode->i_uid;
1677 	ogid = inode->i_gid;
1678 	nuid = attr->ia_uid;
1679 	ngid = attr->ia_gid;
1680 
1681 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1682 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1683 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1684 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1685 
1686 	error = get_write_access(inode);
1687 	if (error)
1688 		return error;
1689 
1690 	error = gfs2_rs_alloc(ip);
1691 	if (error)
1692 		goto out;
1693 
1694 	error = gfs2_rindex_update(sdp);
1695 	if (error)
1696 		goto out;
1697 
1698 	error = gfs2_quota_lock(ip, nuid, ngid);
1699 	if (error)
1700 		goto out;
1701 
1702 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1703 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1704 		error = gfs2_quota_check(ip, nuid, ngid);
1705 		if (error)
1706 			goto out_gunlock_q;
1707 	}
1708 
1709 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1710 	if (error)
1711 		goto out_gunlock_q;
1712 
1713 	error = gfs2_setattr_simple(inode, attr);
1714 	if (error)
1715 		goto out_end_trans;
1716 
1717 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1718 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1719 		u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1720 		gfs2_quota_change(ip, -blocks, ouid, ogid);
1721 		gfs2_quota_change(ip, blocks, nuid, ngid);
1722 	}
1723 
1724 out_end_trans:
1725 	gfs2_trans_end(sdp);
1726 out_gunlock_q:
1727 	gfs2_quota_unlock(ip);
1728 out:
1729 	put_write_access(inode);
1730 	return error;
1731 }
1732 
1733 /**
1734  * gfs2_setattr - Change attributes on an inode
1735  * @dentry: The dentry which is changing
1736  * @attr: The structure describing the change
1737  *
1738  * The VFS layer wants to change one or more of an inodes attributes.  Write
1739  * that change out to disk.
1740  *
1741  * Returns: errno
1742  */
1743 
1744 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1745 {
1746 	struct inode *inode = dentry->d_inode;
1747 	struct gfs2_inode *ip = GFS2_I(inode);
1748 	struct gfs2_holder i_gh;
1749 	int error;
1750 
1751 	error = gfs2_rs_alloc(ip);
1752 	if (error)
1753 		return error;
1754 
1755 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1756 	if (error)
1757 		return error;
1758 
1759 	error = -EPERM;
1760 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1761 		goto out;
1762 
1763 	error = inode_change_ok(inode, attr);
1764 	if (error)
1765 		goto out;
1766 
1767 	if (attr->ia_valid & ATTR_SIZE)
1768 		error = gfs2_setattr_size(inode, attr->ia_size);
1769 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1770 		error = setattr_chown(inode, attr);
1771 	else {
1772 		error = gfs2_setattr_simple(inode, attr);
1773 		if (!error && attr->ia_valid & ATTR_MODE)
1774 			error = posix_acl_chmod(inode, inode->i_mode);
1775 	}
1776 
1777 out:
1778 	if (!error)
1779 		mark_inode_dirty(inode);
1780 	gfs2_glock_dq_uninit(&i_gh);
1781 	return error;
1782 }
1783 
1784 /**
1785  * gfs2_getattr - Read out an inode's attributes
1786  * @mnt: The vfsmount the inode is being accessed from
1787  * @dentry: The dentry to stat
1788  * @stat: The inode's stats
1789  *
1790  * This may be called from the VFS directly, or from within GFS2 with the
1791  * inode locked, so we look to see if the glock is already locked and only
1792  * lock the glock if its not already been done. Note that its the NFS
1793  * readdirplus operation which causes this to be called (from filldir)
1794  * with the glock already held.
1795  *
1796  * Returns: errno
1797  */
1798 
1799 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1800 			struct kstat *stat)
1801 {
1802 	struct inode *inode = dentry->d_inode;
1803 	struct gfs2_inode *ip = GFS2_I(inode);
1804 	struct gfs2_holder gh;
1805 	int error;
1806 	int unlock = 0;
1807 
1808 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1809 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1810 		if (error)
1811 			return error;
1812 		unlock = 1;
1813 	}
1814 
1815 	generic_fillattr(inode, stat);
1816 	if (unlock)
1817 		gfs2_glock_dq_uninit(&gh);
1818 
1819 	return 0;
1820 }
1821 
1822 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1823 			 const void *data, size_t size, int flags)
1824 {
1825 	struct inode *inode = dentry->d_inode;
1826 	struct gfs2_inode *ip = GFS2_I(inode);
1827 	struct gfs2_holder gh;
1828 	int ret;
1829 
1830 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1831 	ret = gfs2_glock_nq(&gh);
1832 	if (ret == 0) {
1833 		ret = gfs2_rs_alloc(ip);
1834 		if (ret == 0)
1835 			ret = generic_setxattr(dentry, name, data, size, flags);
1836 		gfs2_glock_dq(&gh);
1837 	}
1838 	gfs2_holder_uninit(&gh);
1839 	return ret;
1840 }
1841 
1842 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1843 			     void *data, size_t size)
1844 {
1845 	struct inode *inode = dentry->d_inode;
1846 	struct gfs2_inode *ip = GFS2_I(inode);
1847 	struct gfs2_holder gh;
1848 	int ret;
1849 
1850 	/* For selinux during lookup */
1851 	if (gfs2_glock_is_locked_by_me(ip->i_gl))
1852 		return generic_getxattr(dentry, name, data, size);
1853 
1854 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1855 	ret = gfs2_glock_nq(&gh);
1856 	if (ret == 0) {
1857 		ret = generic_getxattr(dentry, name, data, size);
1858 		gfs2_glock_dq(&gh);
1859 	}
1860 	gfs2_holder_uninit(&gh);
1861 	return ret;
1862 }
1863 
1864 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1865 {
1866 	struct inode *inode = dentry->d_inode;
1867 	struct gfs2_inode *ip = GFS2_I(inode);
1868 	struct gfs2_holder gh;
1869 	int ret;
1870 
1871 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1872 	ret = gfs2_glock_nq(&gh);
1873 	if (ret == 0) {
1874 		ret = gfs2_rs_alloc(ip);
1875 		if (ret == 0)
1876 			ret = generic_removexattr(dentry, name);
1877 		gfs2_glock_dq(&gh);
1878 	}
1879 	gfs2_holder_uninit(&gh);
1880 	return ret;
1881 }
1882 
1883 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1884 		       u64 start, u64 len)
1885 {
1886 	struct gfs2_inode *ip = GFS2_I(inode);
1887 	struct gfs2_holder gh;
1888 	int ret;
1889 
1890 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1891 	if (ret)
1892 		return ret;
1893 
1894 	mutex_lock(&inode->i_mutex);
1895 
1896 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1897 	if (ret)
1898 		goto out;
1899 
1900 	if (gfs2_is_stuffed(ip)) {
1901 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1902 		u64 size = i_size_read(inode);
1903 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1904 			    FIEMAP_EXTENT_DATA_INLINE;
1905 		phys += sizeof(struct gfs2_dinode);
1906 		phys += start;
1907 		if (start + len > size)
1908 			len = size - start;
1909 		if (start < size)
1910 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1911 						      len, flags);
1912 		if (ret == 1)
1913 			ret = 0;
1914 	} else {
1915 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1916 					     gfs2_block_map);
1917 	}
1918 
1919 	gfs2_glock_dq_uninit(&gh);
1920 out:
1921 	mutex_unlock(&inode->i_mutex);
1922 	return ret;
1923 }
1924 
1925 const struct inode_operations gfs2_file_iops = {
1926 	.permission = gfs2_permission,
1927 	.setattr = gfs2_setattr,
1928 	.getattr = gfs2_getattr,
1929 	.setxattr = gfs2_setxattr,
1930 	.getxattr = gfs2_getxattr,
1931 	.listxattr = gfs2_listxattr,
1932 	.removexattr = gfs2_removexattr,
1933 	.fiemap = gfs2_fiemap,
1934 	.get_acl = gfs2_get_acl,
1935 	.set_acl = gfs2_set_acl,
1936 };
1937 
1938 const struct inode_operations gfs2_dir_iops = {
1939 	.create = gfs2_create,
1940 	.lookup = gfs2_lookup,
1941 	.link = gfs2_link,
1942 	.unlink = gfs2_unlink,
1943 	.symlink = gfs2_symlink,
1944 	.mkdir = gfs2_mkdir,
1945 	.rmdir = gfs2_unlink,
1946 	.mknod = gfs2_mknod,
1947 	.rename = gfs2_rename,
1948 	.permission = gfs2_permission,
1949 	.setattr = gfs2_setattr,
1950 	.getattr = gfs2_getattr,
1951 	.setxattr = gfs2_setxattr,
1952 	.getxattr = gfs2_getxattr,
1953 	.listxattr = gfs2_listxattr,
1954 	.removexattr = gfs2_removexattr,
1955 	.fiemap = gfs2_fiemap,
1956 	.get_acl = gfs2_get_acl,
1957 	.set_acl = gfs2_set_acl,
1958 	.atomic_open = gfs2_atomic_open,
1959 };
1960 
1961 const struct inode_operations gfs2_symlink_iops = {
1962 	.readlink = generic_readlink,
1963 	.follow_link = gfs2_follow_link,
1964 	.put_link = kfree_put_link,
1965 	.permission = gfs2_permission,
1966 	.setattr = gfs2_setattr,
1967 	.getattr = gfs2_getattr,
1968 	.setxattr = gfs2_setxattr,
1969 	.getxattr = gfs2_getxattr,
1970 	.listxattr = gfs2_listxattr,
1971 	.removexattr = gfs2_removexattr,
1972 	.fiemap = gfs2_fiemap,
1973 };
1974 
1975