xref: /openbmc/linux/fs/gfs2/inode.c (revision 0d456bad)
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(-ENOBUFS);
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_gl->gl_object = NULL;
193 	gfs2_glock_dq_uninit(&ip->i_iopen_gh);
194 fail_iopen:
195 	if (io_gl)
196 		gfs2_glock_put(io_gl);
197 fail_put:
198 	ip->i_gl->gl_object = NULL;
199 	gfs2_glock_put(ip->i_gl);
200 fail:
201 	iget_failed(inode);
202 	return ERR_PTR(error);
203 }
204 
205 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
206 				  u64 *no_formal_ino, unsigned int blktype)
207 {
208 	struct super_block *sb = sdp->sd_vfs;
209 	struct gfs2_holder i_gh;
210 	struct inode *inode = NULL;
211 	int error;
212 
213 	/* Must not read in block until block type is verified */
214 	error = gfs2_glock_nq_num(sdp, no_addr, &gfs2_inode_glops,
215 				  LM_ST_EXCLUSIVE, GL_SKIP, &i_gh);
216 	if (error)
217 		return ERR_PTR(error);
218 
219 	error = gfs2_check_blk_type(sdp, no_addr, blktype);
220 	if (error)
221 		goto fail;
222 
223 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0, 1);
224 	if (IS_ERR(inode))
225 		goto fail;
226 
227 	/* Two extra checks for NFS only */
228 	if (no_formal_ino) {
229 		error = -ESTALE;
230 		if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
231 			goto fail_iput;
232 
233 		error = -EIO;
234 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
235 			goto fail_iput;
236 
237 		error = 0;
238 	}
239 
240 fail:
241 	gfs2_glock_dq_uninit(&i_gh);
242 	return error ? ERR_PTR(error) : inode;
243 fail_iput:
244 	iput(inode);
245 	goto fail;
246 }
247 
248 
249 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
250 {
251 	struct qstr qstr;
252 	struct inode *inode;
253 	gfs2_str2qstr(&qstr, name);
254 	inode = gfs2_lookupi(dip, &qstr, 1);
255 	/* gfs2_lookupi has inconsistent callers: vfs
256 	 * related routines expect NULL for no entry found,
257 	 * gfs2_lookup_simple callers expect ENOENT
258 	 * and do not check for NULL.
259 	 */
260 	if (inode == NULL)
261 		return ERR_PTR(-ENOENT);
262 	else
263 		return inode;
264 }
265 
266 
267 /**
268  * gfs2_lookupi - Look up a filename in a directory and return its inode
269  * @d_gh: An initialized holder for the directory glock
270  * @name: The name of the inode to look for
271  * @is_root: If 1, ignore the caller's permissions
272  * @i_gh: An uninitialized holder for the new inode glock
273  *
274  * This can be called via the VFS filldir function when NFS is doing
275  * a readdirplus and the inode which its intending to stat isn't
276  * already in cache. In this case we must not take the directory glock
277  * again, since the readdir call will have already taken that lock.
278  *
279  * Returns: errno
280  */
281 
282 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
283 			   int is_root)
284 {
285 	struct super_block *sb = dir->i_sb;
286 	struct gfs2_inode *dip = GFS2_I(dir);
287 	struct gfs2_holder d_gh;
288 	int error = 0;
289 	struct inode *inode = NULL;
290 	int unlock = 0;
291 
292 	if (!name->len || name->len > GFS2_FNAMESIZE)
293 		return ERR_PTR(-ENAMETOOLONG);
294 
295 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
296 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
297 	     dir == sb->s_root->d_inode)) {
298 		igrab(dir);
299 		return dir;
300 	}
301 
302 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
303 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
304 		if (error)
305 			return ERR_PTR(error);
306 		unlock = 1;
307 	}
308 
309 	if (!is_root) {
310 		error = gfs2_permission(dir, MAY_EXEC);
311 		if (error)
312 			goto out;
313 	}
314 
315 	inode = gfs2_dir_search(dir, name);
316 	if (IS_ERR(inode))
317 		error = PTR_ERR(inode);
318 out:
319 	if (unlock)
320 		gfs2_glock_dq_uninit(&d_gh);
321 	if (error == -ENOENT)
322 		return NULL;
323 	return inode ? inode : ERR_PTR(error);
324 }
325 
326 /**
327  * create_ok - OK to create a new on-disk inode here?
328  * @dip:  Directory in which dinode is to be created
329  * @name:  Name of new dinode
330  * @mode:
331  *
332  * Returns: errno
333  */
334 
335 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
336 		     umode_t mode)
337 {
338 	int error;
339 
340 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
341 	if (error)
342 		return error;
343 
344 	/*  Don't create entries in an unlinked directory  */
345 	if (!dip->i_inode.i_nlink)
346 		return -ENOENT;
347 
348 	error = gfs2_dir_check(&dip->i_inode, name, NULL);
349 	switch (error) {
350 	case -ENOENT:
351 		error = 0;
352 		break;
353 	case 0:
354 		return -EEXIST;
355 	default:
356 		return error;
357 	}
358 
359 	if (dip->i_entries == (u32)-1)
360 		return -EFBIG;
361 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
362 		return -EMLINK;
363 
364 	return 0;
365 }
366 
367 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
368 			       struct inode *inode)
369 {
370 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
371 	    (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
372 		if (S_ISDIR(inode->i_mode))
373 			inode->i_mode |= S_ISUID;
374 		else if (dip->i_inode.i_uid != current_fsuid())
375 			inode->i_mode &= ~07111;
376 		inode->i_uid = dip->i_inode.i_uid;
377 	} else
378 		inode->i_uid = current_fsuid();
379 
380 	if (dip->i_inode.i_mode & S_ISGID) {
381 		if (S_ISDIR(inode->i_mode))
382 			inode->i_mode |= S_ISGID;
383 		inode->i_gid = dip->i_inode.i_gid;
384 	} else
385 		inode->i_gid = current_fsgid();
386 }
387 
388 static int alloc_dinode(struct gfs2_inode *ip, u32 flags)
389 {
390 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
391 	int error;
392 	int dblocks = 1;
393 
394 	error = gfs2_inplace_reserve(ip, RES_DINODE, flags);
395 	if (error)
396 		goto out;
397 
398 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
399 	if (error)
400 		goto out_ipreserv;
401 
402 	error = gfs2_alloc_blocks(ip, &ip->i_no_addr, &dblocks, 1, &ip->i_generation);
403 	ip->i_no_formal_ino = ip->i_generation;
404 	ip->i_inode.i_ino = ip->i_no_addr;
405 	ip->i_goal = ip->i_no_addr;
406 
407 	gfs2_trans_end(sdp);
408 
409 out_ipreserv:
410 	gfs2_inplace_release(ip);
411 out:
412 	return error;
413 }
414 
415 static void gfs2_init_dir(struct buffer_head *dibh,
416 			  const struct gfs2_inode *parent)
417 {
418 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
419 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
420 
421 	gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
422 	dent->de_inum = di->di_num; /* already GFS2 endian */
423 	dent->de_type = cpu_to_be16(DT_DIR);
424 
425 	dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
426 	gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
427 	gfs2_inum_out(parent, dent);
428 	dent->de_type = cpu_to_be16(DT_DIR);
429 
430 }
431 
432 /**
433  * init_dinode - Fill in a new dinode structure
434  * @dip: The directory this inode is being created in
435  * @ip: The inode
436  * @symname: The symlink destination (if a symlink)
437  * @bhp: The buffer head (returned to caller)
438  *
439  */
440 
441 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
442 			const char *symname, struct buffer_head **bhp)
443 {
444 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
445 	struct gfs2_dinode *di;
446 	struct buffer_head *dibh;
447 	struct timespec tv = CURRENT_TIME;
448 
449 	dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
450 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
451 	gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
452 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
453 	di = (struct gfs2_dinode *)dibh->b_data;
454 
455 	di->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
456 	di->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
457 	di->di_mode = cpu_to_be32(ip->i_inode.i_mode);
458 	di->di_uid = cpu_to_be32(ip->i_inode.i_uid);
459 	di->di_gid = cpu_to_be32(ip->i_inode.i_gid);
460 	di->di_nlink = 0;
461 	di->di_size = cpu_to_be64(ip->i_inode.i_size);
462 	di->di_blocks = cpu_to_be64(1);
463 	di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(tv.tv_sec);
464 	di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
465 	di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
466 	di->di_goal_meta = di->di_goal_data = cpu_to_be64(ip->i_no_addr);
467 	di->di_generation = cpu_to_be64(ip->i_generation);
468 	di->di_flags = 0;
469 	di->__pad1 = 0;
470 	di->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) ? GFS2_FORMAT_DE : 0);
471 	di->di_height = 0;
472 	di->__pad2 = 0;
473 	di->__pad3 = 0;
474 	di->di_depth = 0;
475 	di->di_entries = 0;
476 	memset(&di->__pad4, 0, sizeof(di->__pad4));
477 	di->di_eattr = 0;
478 	di->di_atime_nsec = cpu_to_be32(tv.tv_nsec);
479 	di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec);
480 	di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec);
481 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
482 
483 	switch(ip->i_inode.i_mode & S_IFMT) {
484 	case S_IFREG:
485 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
486 		    gfs2_tune_get(sdp, gt_new_files_jdata))
487 			di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
488 		break;
489 	case S_IFDIR:
490 		di->di_flags |= cpu_to_be32(dip->i_diskflags &
491 					    GFS2_DIF_INHERIT_JDATA);
492 		di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
493 		di->di_size = cpu_to_be64(sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode));
494 		di->di_entries = cpu_to_be32(2);
495 		gfs2_init_dir(dibh, dip);
496 		break;
497 	case S_IFLNK:
498 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
499 		break;
500 	}
501 
502 	set_buffer_uptodate(dibh);
503 
504 	*bhp = dibh;
505 }
506 
507 static int make_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
508 		       const char *symname, struct buffer_head **bhp)
509 {
510 	struct inode *inode = &ip->i_inode;
511 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
512 	int error;
513 
514 	error = gfs2_rindex_update(sdp);
515 	if (error)
516 		return error;
517 
518 	error = gfs2_quota_lock(dip, inode->i_uid, inode->i_gid);
519 	if (error)
520 		return error;
521 
522 	error = gfs2_quota_check(dip, inode->i_uid, inode->i_gid);
523 	if (error)
524 		goto out_quota;
525 
526 	error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
527 	if (error)
528 		goto out_quota;
529 
530 	init_dinode(dip, ip, symname, bhp);
531 	gfs2_quota_change(dip, +1, inode->i_uid, inode->i_gid);
532 	gfs2_trans_end(sdp);
533 
534 out_quota:
535 	gfs2_quota_unlock(dip);
536 	return error;
537 }
538 
539 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
540 		       struct gfs2_inode *ip)
541 {
542 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
543 	int alloc_required;
544 	struct buffer_head *dibh;
545 	int error;
546 
547 	error = gfs2_rindex_update(sdp);
548 	if (error)
549 		return error;
550 
551 	error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
552 	if (error)
553 		goto fail;
554 
555 	error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
556 	if (alloc_required < 0)
557 		goto fail_quota_locks;
558 	if (alloc_required) {
559 		error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
560 		if (error)
561 			goto fail_quota_locks;
562 
563 		error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
564 		if (error)
565 			goto fail_quota_locks;
566 
567 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
568 					 dip->i_rgd->rd_length +
569 					 2 * RES_DINODE +
570 					 RES_STATFS + RES_QUOTA, 0);
571 		if (error)
572 			goto fail_ipreserv;
573 	} else {
574 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
575 		if (error)
576 			goto fail_quota_locks;
577 	}
578 
579 	error = gfs2_dir_add(&dip->i_inode, name, ip);
580 	if (error)
581 		goto fail_end_trans;
582 
583 	error = gfs2_meta_inode_buffer(ip, &dibh);
584 	if (error)
585 		goto fail_end_trans;
586 	set_nlink(&ip->i_inode, S_ISDIR(ip->i_inode.i_mode) ? 2 : 1);
587 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
588 	gfs2_dinode_out(ip, dibh->b_data);
589 	brelse(dibh);
590 	return 0;
591 
592 fail_end_trans:
593 	gfs2_trans_end(sdp);
594 
595 fail_ipreserv:
596 	if (alloc_required)
597 		gfs2_inplace_release(dip);
598 
599 fail_quota_locks:
600 	gfs2_quota_unlock(dip);
601 
602 fail:
603 	return error;
604 }
605 
606 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
607 		    void *fs_info)
608 {
609 	const struct xattr *xattr;
610 	int err = 0;
611 
612 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
613 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
614 				       xattr->value_len, 0,
615 				       GFS2_EATYPE_SECURITY);
616 		if (err < 0)
617 			break;
618 	}
619 	return err;
620 }
621 
622 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
623 			      const struct qstr *qstr)
624 {
625 	return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
626 					    &gfs2_initxattrs, NULL);
627 }
628 
629 /**
630  * gfs2_create_inode - Create a new inode
631  * @dir: The parent directory
632  * @dentry: The new dentry
633  * @mode: The permissions on the new inode
634  * @dev: For device nodes, this is the device number
635  * @symname: For symlinks, this is the link destination
636  * @size: The initial size of the inode (ignored for directories)
637  *
638  * Returns: 0 on success, or error code
639  */
640 
641 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
642 			     umode_t mode, dev_t dev, const char *symname,
643 			     unsigned int size, int excl)
644 {
645 	const struct qstr *name = &dentry->d_name;
646 	struct gfs2_holder ghs[2];
647 	struct inode *inode = NULL;
648 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
649 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
650 	struct gfs2_glock *io_gl;
651 	int error;
652 	struct buffer_head *bh = NULL;
653 	u32 aflags = 0;
654 
655 	if (!name->len || name->len > GFS2_FNAMESIZE)
656 		return -ENAMETOOLONG;
657 
658 	error = gfs2_rs_alloc(dip);
659 	if (error)
660 		return error;
661 
662 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
663 	if (error)
664 		goto fail;
665 
666 	error = create_ok(dip, name, mode);
667 	if ((error == -EEXIST) && S_ISREG(mode) && !excl) {
668 		inode = gfs2_lookupi(dir, &dentry->d_name, 0);
669 		gfs2_glock_dq_uninit(ghs);
670 		d_instantiate(dentry, inode);
671 		return IS_ERR(inode) ? PTR_ERR(inode) : 0;
672 	}
673 	if (error)
674 		goto fail_gunlock;
675 
676 	inode = new_inode(sdp->sd_vfs);
677 	if (!inode) {
678 		gfs2_glock_dq_uninit(ghs);
679 		return -ENOMEM;
680 	}
681 	ip = GFS2_I(inode);
682 	error = gfs2_rs_alloc(ip);
683 	if (error)
684 		goto fail_free_inode;
685 
686 	set_bit(GIF_INVALID, &ip->i_flags);
687 	inode->i_mode = mode;
688 	inode->i_rdev = dev;
689 	inode->i_size = size;
690 	munge_mode_uid_gid(dip, inode);
691 	ip->i_goal = dip->i_goal;
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 	error = alloc_dinode(ip, aflags);
698 	if (error)
699 		goto fail_free_inode;
700 
701 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
702 	if (error)
703 		goto fail_free_inode;
704 
705 	ip->i_gl->gl_object = ip;
706 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
707 	if (error)
708 		goto fail_free_inode;
709 
710 	error = make_dinode(dip, ip, symname, &bh);
711 	if (error)
712 		goto fail_gunlock2;
713 
714 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
715 	if (error)
716 		goto fail_gunlock2;
717 
718 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
719 	if (error)
720 		goto fail_gunlock2;
721 
722 	ip->i_iopen_gh.gh_gl->gl_object = ip;
723 	gfs2_glock_put(io_gl);
724 	gfs2_set_iop(inode);
725 	insert_inode_hash(inode);
726 
727 	error = gfs2_inode_refresh(ip);
728 	if (error)
729 		goto fail_gunlock3;
730 
731 	error = gfs2_acl_create(dip, inode);
732 	if (error)
733 		goto fail_gunlock3;
734 
735 	error = gfs2_security_init(dip, ip, name);
736 	if (error)
737 		goto fail_gunlock3;
738 
739 	error = link_dinode(dip, name, ip);
740 	if (error)
741 		goto fail_gunlock3;
742 
743 	if (bh)
744 		brelse(bh);
745 
746 	gfs2_trans_end(sdp);
747 	gfs2_inplace_release(dip);
748 	gfs2_quota_unlock(dip);
749 	mark_inode_dirty(inode);
750 	gfs2_glock_dq_uninit_m(2, ghs);
751 	d_instantiate(dentry, inode);
752 	return 0;
753 
754 fail_gunlock3:
755 	gfs2_glock_dq_uninit(ghs + 1);
756 	if (ip->i_gl)
757 		gfs2_glock_put(ip->i_gl);
758 	goto fail_gunlock;
759 
760 fail_gunlock2:
761 	gfs2_glock_dq_uninit(ghs + 1);
762 fail_free_inode:
763 	if (ip->i_gl)
764 		gfs2_glock_put(ip->i_gl);
765 	gfs2_rs_delete(ip);
766 	free_inode_nonrcu(inode);
767 	inode = NULL;
768 fail_gunlock:
769 	gfs2_glock_dq_uninit(ghs);
770 	if (inode && !IS_ERR(inode)) {
771 		set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
772 		iput(inode);
773 	}
774 fail:
775 	if (bh)
776 		brelse(bh);
777 	return error;
778 }
779 
780 /**
781  * gfs2_create - Create a file
782  * @dir: The directory in which to create the file
783  * @dentry: The dentry of the new file
784  * @mode: The mode of the new file
785  *
786  * Returns: errno
787  */
788 
789 static int gfs2_create(struct inode *dir, struct dentry *dentry,
790 		       umode_t mode, bool excl)
791 {
792 	return gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0, excl);
793 }
794 
795 /**
796  * gfs2_lookup - Look up a filename in a directory and return its inode
797  * @dir: The directory inode
798  * @dentry: The dentry of the new inode
799  * @nd: passed from Linux VFS, ignored by us
800  *
801  * Called by the VFS layer. Lock dir and call gfs2_lookupi()
802  *
803  * Returns: errno
804  */
805 
806 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
807 				  unsigned int flags)
808 {
809 	struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
810 	if (inode && !IS_ERR(inode)) {
811 		struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
812 		struct gfs2_holder gh;
813 		int error;
814 		error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
815 		if (error) {
816 			iput(inode);
817 			return ERR_PTR(error);
818 		}
819 		gfs2_glock_dq_uninit(&gh);
820 	}
821 	return d_splice_alias(inode, dentry);
822 }
823 
824 /**
825  * gfs2_link - Link to a file
826  * @old_dentry: The inode to link
827  * @dir: Add link to this directory
828  * @dentry: The name of the link
829  *
830  * Link the inode in "old_dentry" into the directory "dir" with the
831  * name in "dentry".
832  *
833  * Returns: errno
834  */
835 
836 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
837 		     struct dentry *dentry)
838 {
839 	struct gfs2_inode *dip = GFS2_I(dir);
840 	struct gfs2_sbd *sdp = GFS2_SB(dir);
841 	struct inode *inode = old_dentry->d_inode;
842 	struct gfs2_inode *ip = GFS2_I(inode);
843 	struct gfs2_holder ghs[2];
844 	struct buffer_head *dibh;
845 	int alloc_required;
846 	int error;
847 
848 	if (S_ISDIR(inode->i_mode))
849 		return -EPERM;
850 
851 	error = gfs2_rs_alloc(dip);
852 	if (error)
853 		return error;
854 
855 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
856 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
857 
858 	error = gfs2_glock_nq(ghs); /* parent */
859 	if (error)
860 		goto out_parent;
861 
862 	error = gfs2_glock_nq(ghs + 1); /* child */
863 	if (error)
864 		goto out_child;
865 
866 	error = -ENOENT;
867 	if (inode->i_nlink == 0)
868 		goto out_gunlock;
869 
870 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
871 	if (error)
872 		goto out_gunlock;
873 
874 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
875 	switch (error) {
876 	case -ENOENT:
877 		break;
878 	case 0:
879 		error = -EEXIST;
880 	default:
881 		goto out_gunlock;
882 	}
883 
884 	error = -EINVAL;
885 	if (!dip->i_inode.i_nlink)
886 		goto out_gunlock;
887 	error = -EFBIG;
888 	if (dip->i_entries == (u32)-1)
889 		goto out_gunlock;
890 	error = -EPERM;
891 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
892 		goto out_gunlock;
893 	error = -EINVAL;
894 	if (!ip->i_inode.i_nlink)
895 		goto out_gunlock;
896 	error = -EMLINK;
897 	if (ip->i_inode.i_nlink == (u32)-1)
898 		goto out_gunlock;
899 
900 	alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
901 	if (error < 0)
902 		goto out_gunlock;
903 	error = 0;
904 
905 	if (alloc_required) {
906 		error = gfs2_quota_lock_check(dip);
907 		if (error)
908 			goto out_gunlock;
909 
910 		error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
911 		if (error)
912 			goto out_gunlock_q;
913 
914 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
915 					 gfs2_rg_blocks(dip, sdp->sd_max_dirres) +
916 					 2 * RES_DINODE + RES_STATFS +
917 					 RES_QUOTA, 0);
918 		if (error)
919 			goto out_ipres;
920 	} else {
921 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
922 		if (error)
923 			goto out_ipres;
924 	}
925 
926 	error = gfs2_meta_inode_buffer(ip, &dibh);
927 	if (error)
928 		goto out_end_trans;
929 
930 	error = gfs2_dir_add(dir, &dentry->d_name, ip);
931 	if (error)
932 		goto out_brelse;
933 
934 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
935 	inc_nlink(&ip->i_inode);
936 	ip->i_inode.i_ctime = CURRENT_TIME;
937 	ihold(inode);
938 	d_instantiate(dentry, inode);
939 	mark_inode_dirty(inode);
940 
941 out_brelse:
942 	brelse(dibh);
943 out_end_trans:
944 	gfs2_trans_end(sdp);
945 out_ipres:
946 	if (alloc_required)
947 		gfs2_inplace_release(dip);
948 out_gunlock_q:
949 	if (alloc_required)
950 		gfs2_quota_unlock(dip);
951 out_gunlock:
952 	gfs2_glock_dq(ghs + 1);
953 out_child:
954 	gfs2_glock_dq(ghs);
955 out_parent:
956 	gfs2_holder_uninit(ghs);
957 	gfs2_holder_uninit(ghs + 1);
958 	return error;
959 }
960 
961 /*
962  * gfs2_unlink_ok - check to see that a inode is still in a directory
963  * @dip: the directory
964  * @name: the name of the file
965  * @ip: the inode
966  *
967  * Assumes that the lock on (at least) @dip is held.
968  *
969  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
970  */
971 
972 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
973 			  const struct gfs2_inode *ip)
974 {
975 	int error;
976 
977 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
978 		return -EPERM;
979 
980 	if ((dip->i_inode.i_mode & S_ISVTX) &&
981 	    dip->i_inode.i_uid != current_fsuid() &&
982 	    ip->i_inode.i_uid != current_fsuid() && !capable(CAP_FOWNER))
983 		return -EPERM;
984 
985 	if (IS_APPEND(&dip->i_inode))
986 		return -EPERM;
987 
988 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
989 	if (error)
990 		return error;
991 
992 	error = gfs2_dir_check(&dip->i_inode, name, ip);
993 	if (error)
994 		return error;
995 
996 	return 0;
997 }
998 
999 /**
1000  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1001  * @dip: The parent directory
1002  * @name: The name of the entry in the parent directory
1003  * @inode: The inode to be removed
1004  *
1005  * Called with all the locks and in a transaction. This will only be
1006  * called for a directory after it has been checked to ensure it is empty.
1007  *
1008  * Returns: 0 on success, or an error
1009  */
1010 
1011 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1012 			     const struct dentry *dentry)
1013 {
1014 	struct inode *inode = dentry->d_inode;
1015 	struct gfs2_inode *ip = GFS2_I(inode);
1016 	int error;
1017 
1018 	error = gfs2_dir_del(dip, dentry);
1019 	if (error)
1020 		return error;
1021 
1022 	ip->i_entries = 0;
1023 	inode->i_ctime = CURRENT_TIME;
1024 	if (S_ISDIR(inode->i_mode))
1025 		clear_nlink(inode);
1026 	else
1027 		drop_nlink(inode);
1028 	mark_inode_dirty(inode);
1029 	if (inode->i_nlink == 0)
1030 		gfs2_unlink_di(inode);
1031 	return 0;
1032 }
1033 
1034 
1035 /**
1036  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1037  * @dir: The inode of the directory containing the inode to unlink
1038  * @dentry: The file itself
1039  *
1040  * This routine uses the type of the inode as a flag to figure out
1041  * whether this is an unlink or an rmdir.
1042  *
1043  * Returns: errno
1044  */
1045 
1046 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1047 {
1048 	struct gfs2_inode *dip = GFS2_I(dir);
1049 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1050 	struct inode *inode = dentry->d_inode;
1051 	struct gfs2_inode *ip = GFS2_I(inode);
1052 	struct gfs2_holder ghs[3];
1053 	struct gfs2_rgrpd *rgd;
1054 	int error;
1055 
1056 	error = gfs2_rindex_update(sdp);
1057 	if (error)
1058 		return error;
1059 
1060 	error = -EROFS;
1061 
1062 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1063 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1064 
1065 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1066 	if (!rgd)
1067 		goto out_inodes;
1068 
1069 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1070 
1071 
1072 	error = gfs2_glock_nq(ghs); /* parent */
1073 	if (error)
1074 		goto out_parent;
1075 
1076 	error = gfs2_glock_nq(ghs + 1); /* child */
1077 	if (error)
1078 		goto out_child;
1079 
1080 	error = -ENOENT;
1081 	if (inode->i_nlink == 0)
1082 		goto out_rgrp;
1083 
1084 	if (S_ISDIR(inode->i_mode)) {
1085 		error = -ENOTEMPTY;
1086 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1087 			goto out_rgrp;
1088 	}
1089 
1090 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1091 	if (error)
1092 		goto out_rgrp;
1093 
1094 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1095 	if (error)
1096 		goto out_gunlock;
1097 
1098 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1099 	if (error)
1100 		goto out_end_trans;
1101 
1102 	error = gfs2_unlink_inode(dip, dentry);
1103 
1104 out_end_trans:
1105 	gfs2_trans_end(sdp);
1106 out_gunlock:
1107 	gfs2_glock_dq(ghs + 2);
1108 out_rgrp:
1109 	gfs2_glock_dq(ghs + 1);
1110 out_child:
1111 	gfs2_glock_dq(ghs);
1112 out_parent:
1113 	gfs2_holder_uninit(ghs + 2);
1114 out_inodes:
1115 	gfs2_holder_uninit(ghs + 1);
1116 	gfs2_holder_uninit(ghs);
1117 	return error;
1118 }
1119 
1120 /**
1121  * gfs2_symlink - Create a symlink
1122  * @dir: The directory to create the symlink in
1123  * @dentry: The dentry to put the symlink in
1124  * @symname: The thing which the link points to
1125  *
1126  * Returns: errno
1127  */
1128 
1129 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1130 			const char *symname)
1131 {
1132 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1133 	unsigned int size;
1134 
1135 	size = strlen(symname);
1136 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1137 		return -ENAMETOOLONG;
1138 
1139 	return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1140 }
1141 
1142 /**
1143  * gfs2_mkdir - Make a directory
1144  * @dir: The parent directory of the new one
1145  * @dentry: The dentry of the new directory
1146  * @mode: The mode of the new directory
1147  *
1148  * Returns: errno
1149  */
1150 
1151 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1152 {
1153 	return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, 0, 0);
1154 }
1155 
1156 /**
1157  * gfs2_mknod - Make a special file
1158  * @dir: The directory in which the special file will reside
1159  * @dentry: The dentry of the special file
1160  * @mode: The mode of the special file
1161  * @dev: The device specification of the special file
1162  *
1163  */
1164 
1165 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1166 		      dev_t dev)
1167 {
1168 	return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0, 0);
1169 }
1170 
1171 /*
1172  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1173  * @this: move this
1174  * @to: to here
1175  *
1176  * Follow @to back to the root and make sure we don't encounter @this
1177  * Assumes we already hold the rename lock.
1178  *
1179  * Returns: errno
1180  */
1181 
1182 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1183 {
1184 	struct inode *dir = &to->i_inode;
1185 	struct super_block *sb = dir->i_sb;
1186 	struct inode *tmp;
1187 	int error = 0;
1188 
1189 	igrab(dir);
1190 
1191 	for (;;) {
1192 		if (dir == &this->i_inode) {
1193 			error = -EINVAL;
1194 			break;
1195 		}
1196 		if (dir == sb->s_root->d_inode) {
1197 			error = 0;
1198 			break;
1199 		}
1200 
1201 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1202 		if (IS_ERR(tmp)) {
1203 			error = PTR_ERR(tmp);
1204 			break;
1205 		}
1206 
1207 		iput(dir);
1208 		dir = tmp;
1209 	}
1210 
1211 	iput(dir);
1212 
1213 	return error;
1214 }
1215 
1216 /**
1217  * gfs2_rename - Rename a file
1218  * @odir: Parent directory of old file name
1219  * @odentry: The old dentry of the file
1220  * @ndir: Parent directory of new file name
1221  * @ndentry: The new dentry of the file
1222  *
1223  * Returns: errno
1224  */
1225 
1226 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1227 		       struct inode *ndir, struct dentry *ndentry)
1228 {
1229 	struct gfs2_inode *odip = GFS2_I(odir);
1230 	struct gfs2_inode *ndip = GFS2_I(ndir);
1231 	struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1232 	struct gfs2_inode *nip = NULL;
1233 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1234 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1235 	struct gfs2_rgrpd *nrgd;
1236 	unsigned int num_gh;
1237 	int dir_rename = 0;
1238 	int alloc_required = 0;
1239 	unsigned int x;
1240 	int error;
1241 
1242 	if (ndentry->d_inode) {
1243 		nip = GFS2_I(ndentry->d_inode);
1244 		if (ip == nip)
1245 			return 0;
1246 	}
1247 
1248 	error = gfs2_rindex_update(sdp);
1249 	if (error)
1250 		return error;
1251 
1252 	error = gfs2_rs_alloc(ndip);
1253 	if (error)
1254 		return error;
1255 
1256 	if (odip != ndip) {
1257 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1258 					   0, &r_gh);
1259 		if (error)
1260 			goto out;
1261 
1262 		if (S_ISDIR(ip->i_inode.i_mode)) {
1263 			dir_rename = 1;
1264 			/* don't move a dirctory into it's subdir */
1265 			error = gfs2_ok_to_move(ip, ndip);
1266 			if (error)
1267 				goto out_gunlock_r;
1268 		}
1269 	}
1270 
1271 	num_gh = 1;
1272 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1273 	if (odip != ndip) {
1274 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1275 		num_gh++;
1276 	}
1277 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1278 	num_gh++;
1279 
1280 	if (nip) {
1281 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1282 		num_gh++;
1283 		/* grab the resource lock for unlink flag twiddling
1284 		 * this is the case of the target file already existing
1285 		 * so we unlink before doing the rename
1286 		 */
1287 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1288 		if (nrgd)
1289 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1290 	}
1291 
1292 	for (x = 0; x < num_gh; x++) {
1293 		error = gfs2_glock_nq(ghs + x);
1294 		if (error)
1295 			goto out_gunlock;
1296 	}
1297 
1298 	error = -ENOENT;
1299 	if (ip->i_inode.i_nlink == 0)
1300 		goto out_gunlock;
1301 
1302 	/* Check out the old directory */
1303 
1304 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1305 	if (error)
1306 		goto out_gunlock;
1307 
1308 	/* Check out the new directory */
1309 
1310 	if (nip) {
1311 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1312 		if (error)
1313 			goto out_gunlock;
1314 
1315 		if (nip->i_inode.i_nlink == 0) {
1316 			error = -EAGAIN;
1317 			goto out_gunlock;
1318 		}
1319 
1320 		if (S_ISDIR(nip->i_inode.i_mode)) {
1321 			if (nip->i_entries < 2) {
1322 				gfs2_consist_inode(nip);
1323 				error = -EIO;
1324 				goto out_gunlock;
1325 			}
1326 			if (nip->i_entries > 2) {
1327 				error = -ENOTEMPTY;
1328 				goto out_gunlock;
1329 			}
1330 		}
1331 	} else {
1332 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1333 		if (error)
1334 			goto out_gunlock;
1335 
1336 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1337 		switch (error) {
1338 		case -ENOENT:
1339 			error = 0;
1340 			break;
1341 		case 0:
1342 			error = -EEXIST;
1343 		default:
1344 			goto out_gunlock;
1345 		};
1346 
1347 		if (odip != ndip) {
1348 			if (!ndip->i_inode.i_nlink) {
1349 				error = -ENOENT;
1350 				goto out_gunlock;
1351 			}
1352 			if (ndip->i_entries == (u32)-1) {
1353 				error = -EFBIG;
1354 				goto out_gunlock;
1355 			}
1356 			if (S_ISDIR(ip->i_inode.i_mode) &&
1357 			    ndip->i_inode.i_nlink == (u32)-1) {
1358 				error = -EMLINK;
1359 				goto out_gunlock;
1360 			}
1361 		}
1362 	}
1363 
1364 	/* Check out the dir to be renamed */
1365 
1366 	if (dir_rename) {
1367 		error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1368 		if (error)
1369 			goto out_gunlock;
1370 	}
1371 
1372 	if (nip == NULL)
1373 		alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1374 	error = alloc_required;
1375 	if (error < 0)
1376 		goto out_gunlock;
1377 
1378 	if (alloc_required) {
1379 		error = gfs2_quota_lock_check(ndip);
1380 		if (error)
1381 			goto out_gunlock;
1382 
1383 		error = gfs2_inplace_reserve(ndip, sdp->sd_max_dirres, 0);
1384 		if (error)
1385 			goto out_gunlock_q;
1386 
1387 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1388 					 gfs2_rg_blocks(ndip, sdp->sd_max_dirres) +
1389 					 4 * RES_DINODE + 4 * RES_LEAF +
1390 					 RES_STATFS + RES_QUOTA + 4, 0);
1391 		if (error)
1392 			goto out_ipreserv;
1393 	} else {
1394 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1395 					 5 * RES_LEAF + 4, 0);
1396 		if (error)
1397 			goto out_gunlock;
1398 	}
1399 
1400 	/* Remove the target file, if it exists */
1401 
1402 	if (nip)
1403 		error = gfs2_unlink_inode(ndip, ndentry);
1404 
1405 	if (dir_rename) {
1406 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1407 		if (error)
1408 			goto out_end_trans;
1409 	} else {
1410 		struct buffer_head *dibh;
1411 		error = gfs2_meta_inode_buffer(ip, &dibh);
1412 		if (error)
1413 			goto out_end_trans;
1414 		ip->i_inode.i_ctime = CURRENT_TIME;
1415 		gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1416 		gfs2_dinode_out(ip, dibh->b_data);
1417 		brelse(dibh);
1418 	}
1419 
1420 	error = gfs2_dir_del(odip, odentry);
1421 	if (error)
1422 		goto out_end_trans;
1423 
1424 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1425 	if (error)
1426 		goto out_end_trans;
1427 
1428 out_end_trans:
1429 	gfs2_trans_end(sdp);
1430 out_ipreserv:
1431 	if (alloc_required)
1432 		gfs2_inplace_release(ndip);
1433 out_gunlock_q:
1434 	if (alloc_required)
1435 		gfs2_quota_unlock(ndip);
1436 out_gunlock:
1437 	while (x--) {
1438 		gfs2_glock_dq(ghs + x);
1439 		gfs2_holder_uninit(ghs + x);
1440 	}
1441 out_gunlock_r:
1442 	if (r_gh.gh_gl)
1443 		gfs2_glock_dq_uninit(&r_gh);
1444 out:
1445 	return error;
1446 }
1447 
1448 /**
1449  * gfs2_follow_link - Follow a symbolic link
1450  * @dentry: The dentry of the link
1451  * @nd: Data that we pass to vfs_follow_link()
1452  *
1453  * This can handle symlinks of any size.
1454  *
1455  * Returns: 0 on success or error code
1456  */
1457 
1458 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1459 {
1460 	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1461 	struct gfs2_holder i_gh;
1462 	struct buffer_head *dibh;
1463 	unsigned int size;
1464 	char *buf;
1465 	int error;
1466 
1467 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1468 	error = gfs2_glock_nq(&i_gh);
1469 	if (error) {
1470 		gfs2_holder_uninit(&i_gh);
1471 		nd_set_link(nd, ERR_PTR(error));
1472 		return NULL;
1473 	}
1474 
1475 	size = (unsigned int)i_size_read(&ip->i_inode);
1476 	if (size == 0) {
1477 		gfs2_consist_inode(ip);
1478 		buf = ERR_PTR(-EIO);
1479 		goto out;
1480 	}
1481 
1482 	error = gfs2_meta_inode_buffer(ip, &dibh);
1483 	if (error) {
1484 		buf = ERR_PTR(error);
1485 		goto out;
1486 	}
1487 
1488 	buf = kzalloc(size + 1, GFP_NOFS);
1489 	if (!buf)
1490 		buf = ERR_PTR(-ENOMEM);
1491 	else
1492 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1493 	brelse(dibh);
1494 out:
1495 	gfs2_glock_dq_uninit(&i_gh);
1496 	nd_set_link(nd, buf);
1497 	return NULL;
1498 }
1499 
1500 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1501 {
1502 	char *s = nd_get_link(nd);
1503 	if (!IS_ERR(s))
1504 		kfree(s);
1505 }
1506 
1507 /**
1508  * gfs2_permission -
1509  * @inode: The inode
1510  * @mask: The mask to be tested
1511  * @flags: Indicates whether this is an RCU path walk or not
1512  *
1513  * This may be called from the VFS directly, or from within GFS2 with the
1514  * inode locked, so we look to see if the glock is already locked and only
1515  * lock the glock if its not already been done.
1516  *
1517  * Returns: errno
1518  */
1519 
1520 int gfs2_permission(struct inode *inode, int mask)
1521 {
1522 	struct gfs2_inode *ip;
1523 	struct gfs2_holder i_gh;
1524 	int error;
1525 	int unlock = 0;
1526 
1527 
1528 	ip = GFS2_I(inode);
1529 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1530 		if (mask & MAY_NOT_BLOCK)
1531 			return -ECHILD;
1532 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1533 		if (error)
1534 			return error;
1535 		unlock = 1;
1536 	}
1537 
1538 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1539 		error = -EACCES;
1540 	else
1541 		error = generic_permission(inode, mask);
1542 	if (unlock)
1543 		gfs2_glock_dq_uninit(&i_gh);
1544 
1545 	return error;
1546 }
1547 
1548 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1549 {
1550 	setattr_copy(inode, attr);
1551 	mark_inode_dirty(inode);
1552 	return 0;
1553 }
1554 
1555 /**
1556  * gfs2_setattr_simple -
1557  * @ip:
1558  * @attr:
1559  *
1560  * Returns: errno
1561  */
1562 
1563 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1564 {
1565 	int error;
1566 
1567 	if (current->journal_info)
1568 		return __gfs2_setattr_simple(inode, attr);
1569 
1570 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1571 	if (error)
1572 		return error;
1573 
1574 	error = __gfs2_setattr_simple(inode, attr);
1575 	gfs2_trans_end(GFS2_SB(inode));
1576 	return error;
1577 }
1578 
1579 static int setattr_chown(struct inode *inode, struct iattr *attr)
1580 {
1581 	struct gfs2_inode *ip = GFS2_I(inode);
1582 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1583 	u32 ouid, ogid, nuid, ngid;
1584 	int error;
1585 
1586 	ouid = inode->i_uid;
1587 	ogid = inode->i_gid;
1588 	nuid = attr->ia_uid;
1589 	ngid = attr->ia_gid;
1590 
1591 	if (!(attr->ia_valid & ATTR_UID) || ouid == nuid)
1592 		ouid = nuid = NO_QUOTA_CHANGE;
1593 	if (!(attr->ia_valid & ATTR_GID) || ogid == ngid)
1594 		ogid = ngid = NO_QUOTA_CHANGE;
1595 
1596 	error = gfs2_quota_lock(ip, nuid, ngid);
1597 	if (error)
1598 		return error;
1599 
1600 	if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1601 		error = gfs2_quota_check(ip, nuid, ngid);
1602 		if (error)
1603 			goto out_gunlock_q;
1604 	}
1605 
1606 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1607 	if (error)
1608 		goto out_gunlock_q;
1609 
1610 	error = gfs2_setattr_simple(inode, attr);
1611 	if (error)
1612 		goto out_end_trans;
1613 
1614 	if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1615 		u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1616 		gfs2_quota_change(ip, -blocks, ouid, ogid);
1617 		gfs2_quota_change(ip, blocks, nuid, ngid);
1618 	}
1619 
1620 out_end_trans:
1621 	gfs2_trans_end(sdp);
1622 out_gunlock_q:
1623 	gfs2_quota_unlock(ip);
1624 	return error;
1625 }
1626 
1627 /**
1628  * gfs2_setattr - Change attributes on an inode
1629  * @dentry: The dentry which is changing
1630  * @attr: The structure describing the change
1631  *
1632  * The VFS layer wants to change one or more of an inodes attributes.  Write
1633  * that change out to disk.
1634  *
1635  * Returns: errno
1636  */
1637 
1638 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1639 {
1640 	struct inode *inode = dentry->d_inode;
1641 	struct gfs2_inode *ip = GFS2_I(inode);
1642 	struct gfs2_holder i_gh;
1643 	int error;
1644 
1645 	error = gfs2_rs_alloc(ip);
1646 	if (error)
1647 		return error;
1648 
1649 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1650 	if (error)
1651 		return error;
1652 
1653 	error = -EPERM;
1654 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1655 		goto out;
1656 
1657 	error = inode_change_ok(inode, attr);
1658 	if (error)
1659 		goto out;
1660 
1661 	if (attr->ia_valid & ATTR_SIZE)
1662 		error = gfs2_setattr_size(inode, attr->ia_size);
1663 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1664 		error = setattr_chown(inode, attr);
1665 	else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1666 		error = gfs2_acl_chmod(ip, attr);
1667 	else
1668 		error = gfs2_setattr_simple(inode, attr);
1669 
1670 out:
1671 	if (!error)
1672 		mark_inode_dirty(inode);
1673 	gfs2_glock_dq_uninit(&i_gh);
1674 	return error;
1675 }
1676 
1677 /**
1678  * gfs2_getattr - Read out an inode's attributes
1679  * @mnt: The vfsmount the inode is being accessed from
1680  * @dentry: The dentry to stat
1681  * @stat: The inode's stats
1682  *
1683  * This may be called from the VFS directly, or from within GFS2 with the
1684  * inode locked, so we look to see if the glock is already locked and only
1685  * lock the glock if its not already been done. Note that its the NFS
1686  * readdirplus operation which causes this to be called (from filldir)
1687  * with the glock already held.
1688  *
1689  * Returns: errno
1690  */
1691 
1692 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1693 			struct kstat *stat)
1694 {
1695 	struct inode *inode = dentry->d_inode;
1696 	struct gfs2_inode *ip = GFS2_I(inode);
1697 	struct gfs2_holder gh;
1698 	int error;
1699 	int unlock = 0;
1700 
1701 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1702 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1703 		if (error)
1704 			return error;
1705 		unlock = 1;
1706 	}
1707 
1708 	generic_fillattr(inode, stat);
1709 	if (unlock)
1710 		gfs2_glock_dq_uninit(&gh);
1711 
1712 	return 0;
1713 }
1714 
1715 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1716 			 const void *data, size_t size, int flags)
1717 {
1718 	struct inode *inode = dentry->d_inode;
1719 	struct gfs2_inode *ip = GFS2_I(inode);
1720 	struct gfs2_holder gh;
1721 	int ret;
1722 
1723 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1724 	ret = gfs2_glock_nq(&gh);
1725 	if (ret == 0) {
1726 		ret = gfs2_rs_alloc(ip);
1727 		if (ret == 0)
1728 			ret = generic_setxattr(dentry, name, data, size, flags);
1729 		gfs2_glock_dq(&gh);
1730 	}
1731 	gfs2_holder_uninit(&gh);
1732 	return ret;
1733 }
1734 
1735 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1736 			     void *data, size_t size)
1737 {
1738 	struct inode *inode = dentry->d_inode;
1739 	struct gfs2_inode *ip = GFS2_I(inode);
1740 	struct gfs2_holder gh;
1741 	int ret;
1742 
1743 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1744 	ret = gfs2_glock_nq(&gh);
1745 	if (ret == 0) {
1746 		ret = generic_getxattr(dentry, name, data, size);
1747 		gfs2_glock_dq(&gh);
1748 	}
1749 	gfs2_holder_uninit(&gh);
1750 	return ret;
1751 }
1752 
1753 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1754 {
1755 	struct inode *inode = dentry->d_inode;
1756 	struct gfs2_inode *ip = GFS2_I(inode);
1757 	struct gfs2_holder gh;
1758 	int ret;
1759 
1760 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1761 	ret = gfs2_glock_nq(&gh);
1762 	if (ret == 0) {
1763 		ret = gfs2_rs_alloc(ip);
1764 		if (ret == 0)
1765 			ret = generic_removexattr(dentry, name);
1766 		gfs2_glock_dq(&gh);
1767 	}
1768 	gfs2_holder_uninit(&gh);
1769 	return ret;
1770 }
1771 
1772 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1773 		       u64 start, u64 len)
1774 {
1775 	struct gfs2_inode *ip = GFS2_I(inode);
1776 	struct gfs2_holder gh;
1777 	int ret;
1778 
1779 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1780 	if (ret)
1781 		return ret;
1782 
1783 	mutex_lock(&inode->i_mutex);
1784 
1785 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1786 	if (ret)
1787 		goto out;
1788 
1789 	if (gfs2_is_stuffed(ip)) {
1790 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1791 		u64 size = i_size_read(inode);
1792 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1793 			    FIEMAP_EXTENT_DATA_INLINE;
1794 		phys += sizeof(struct gfs2_dinode);
1795 		phys += start;
1796 		if (start + len > size)
1797 			len = size - start;
1798 		if (start < size)
1799 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1800 						      len, flags);
1801 		if (ret == 1)
1802 			ret = 0;
1803 	} else {
1804 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1805 					     gfs2_block_map);
1806 	}
1807 
1808 	gfs2_glock_dq_uninit(&gh);
1809 out:
1810 	mutex_unlock(&inode->i_mutex);
1811 	return ret;
1812 }
1813 
1814 const struct inode_operations gfs2_file_iops = {
1815 	.permission = gfs2_permission,
1816 	.setattr = gfs2_setattr,
1817 	.getattr = gfs2_getattr,
1818 	.setxattr = gfs2_setxattr,
1819 	.getxattr = gfs2_getxattr,
1820 	.listxattr = gfs2_listxattr,
1821 	.removexattr = gfs2_removexattr,
1822 	.fiemap = gfs2_fiemap,
1823 	.get_acl = gfs2_get_acl,
1824 };
1825 
1826 const struct inode_operations gfs2_dir_iops = {
1827 	.create = gfs2_create,
1828 	.lookup = gfs2_lookup,
1829 	.link = gfs2_link,
1830 	.unlink = gfs2_unlink,
1831 	.symlink = gfs2_symlink,
1832 	.mkdir = gfs2_mkdir,
1833 	.rmdir = gfs2_unlink,
1834 	.mknod = gfs2_mknod,
1835 	.rename = gfs2_rename,
1836 	.permission = gfs2_permission,
1837 	.setattr = gfs2_setattr,
1838 	.getattr = gfs2_getattr,
1839 	.setxattr = gfs2_setxattr,
1840 	.getxattr = gfs2_getxattr,
1841 	.listxattr = gfs2_listxattr,
1842 	.removexattr = gfs2_removexattr,
1843 	.fiemap = gfs2_fiemap,
1844 	.get_acl = gfs2_get_acl,
1845 };
1846 
1847 const struct inode_operations gfs2_symlink_iops = {
1848 	.readlink = generic_readlink,
1849 	.follow_link = gfs2_follow_link,
1850 	.put_link = gfs2_put_link,
1851 	.permission = gfs2_permission,
1852 	.setattr = gfs2_setattr,
1853 	.getattr = gfs2_getattr,
1854 	.setxattr = gfs2_setxattr,
1855 	.getxattr = gfs2_getxattr,
1856 	.listxattr = gfs2_listxattr,
1857 	.removexattr = gfs2_removexattr,
1858 	.fiemap = gfs2_fiemap,
1859 	.get_acl = gfs2_get_acl,
1860 };
1861 
1862