xref: /linux/fs/gfs2/inode.c (revision 068df0f34e81bc06c5eb5012ec2eda25624e87aa)
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 		     unsigned int 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(struct gfs2_inode *dip, unsigned int *mode,
368 			       unsigned int *uid, unsigned int *gid)
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(*mode))
373 			*mode |= S_ISUID;
374 		else if (dip->i_inode.i_uid != current_fsuid())
375 			*mode &= ~07111;
376 		*uid = dip->i_inode.i_uid;
377 	} else
378 		*uid = current_fsuid();
379 
380 	if (dip->i_inode.i_mode & S_ISGID) {
381 		if (S_ISDIR(*mode))
382 			*mode |= S_ISGID;
383 		*gid = dip->i_inode.i_gid;
384 	} else
385 		*gid = current_fsgid();
386 }
387 
388 static int alloc_dinode(struct gfs2_inode *dip, u64 *no_addr, u64 *generation)
389 {
390 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
391 	int error;
392 
393 	if (gfs2_alloc_get(dip) == NULL)
394 		return -ENOMEM;
395 
396 	dip->i_alloc->al_requested = RES_DINODE;
397 	error = gfs2_inplace_reserve(dip);
398 	if (error)
399 		goto out;
400 
401 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
402 	if (error)
403 		goto out_ipreserv;
404 
405 	error = gfs2_alloc_di(dip, no_addr, generation);
406 
407 	gfs2_trans_end(sdp);
408 
409 out_ipreserv:
410 	gfs2_inplace_release(dip);
411 out:
412 	gfs2_alloc_put(dip);
413 	return error;
414 }
415 
416 static void gfs2_init_dir(struct buffer_head *dibh,
417 			  const struct gfs2_inode *parent)
418 {
419 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
420 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
421 
422 	gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
423 	dent->de_inum = di->di_num; /* already GFS2 endian */
424 	dent->de_type = cpu_to_be16(DT_DIR);
425 
426 	dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
427 	gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
428 	gfs2_inum_out(parent, dent);
429 	dent->de_type = cpu_to_be16(DT_DIR);
430 
431 }
432 
433 /**
434  * init_dinode - Fill in a new dinode structure
435  * @dip: The directory this inode is being created in
436  * @gl: The glock covering the new inode
437  * @inum: The inode number
438  * @mode: The file permissions
439  * @uid: The uid of the new inode
440  * @gid: The gid of the new inode
441  * @generation: The generation number of the new inode
442  * @dev: The device number (if a device node)
443  * @symname: The symlink destination (if a symlink)
444  * @size: The inode size (ignored for directories)
445  * @bhp: The buffer head (returned to caller)
446  *
447  */
448 
449 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
450 			const struct gfs2_inum_host *inum, unsigned int mode,
451 			unsigned int uid, unsigned int gid,
452 			const u64 *generation, dev_t dev, const char *symname,
453 			unsigned size, struct buffer_head **bhp)
454 {
455 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
456 	struct gfs2_dinode *di;
457 	struct buffer_head *dibh;
458 	struct timespec tv = CURRENT_TIME;
459 
460 	dibh = gfs2_meta_new(gl, inum->no_addr);
461 	gfs2_trans_add_bh(gl, dibh, 1);
462 	gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
463 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
464 	di = (struct gfs2_dinode *)dibh->b_data;
465 
466 	di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
467 	di->di_num.no_addr = cpu_to_be64(inum->no_addr);
468 	di->di_mode = cpu_to_be32(mode);
469 	di->di_uid = cpu_to_be32(uid);
470 	di->di_gid = cpu_to_be32(gid);
471 	di->di_nlink = 0;
472 	di->di_size = cpu_to_be64(size);
473 	di->di_blocks = cpu_to_be64(1);
474 	di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(tv.tv_sec);
475 	di->di_major = cpu_to_be32(MAJOR(dev));
476 	di->di_minor = cpu_to_be32(MINOR(dev));
477 	di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
478 	di->di_generation = cpu_to_be64(*generation);
479 	di->di_flags = 0;
480 	di->__pad1 = 0;
481 	di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
482 	di->di_height = 0;
483 	di->__pad2 = 0;
484 	di->__pad3 = 0;
485 	di->di_depth = 0;
486 	di->di_entries = 0;
487 	memset(&di->__pad4, 0, sizeof(di->__pad4));
488 	di->di_eattr = 0;
489 	di->di_atime_nsec = cpu_to_be32(tv.tv_nsec);
490 	di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec);
491 	di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec);
492 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
493 
494 	switch(mode & S_IFMT) {
495 	case S_IFREG:
496 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
497 		    gfs2_tune_get(sdp, gt_new_files_jdata))
498 			di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
499 		break;
500 	case S_IFDIR:
501 		di->di_flags |= cpu_to_be32(dip->i_diskflags &
502 					    GFS2_DIF_INHERIT_JDATA);
503 		di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
504 		di->di_size = cpu_to_be64(sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode));
505 		di->di_entries = cpu_to_be32(2);
506 		gfs2_init_dir(dibh, dip);
507 		break;
508 	case S_IFLNK:
509 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, size);
510 		break;
511 	}
512 
513 	set_buffer_uptodate(dibh);
514 
515 	*bhp = dibh;
516 }
517 
518 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
519 		       unsigned int mode, const struct gfs2_inum_host *inum,
520 		       const u64 *generation, dev_t dev, const char *symname,
521 		       unsigned int size, struct buffer_head **bhp)
522 {
523 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
524 	unsigned int uid, gid;
525 	int error;
526 
527 	munge_mode_uid_gid(dip, &mode, &uid, &gid);
528 	if (!gfs2_alloc_get(dip))
529 		return -ENOMEM;
530 
531 	error = gfs2_quota_lock(dip, uid, gid);
532 	if (error)
533 		goto out;
534 
535 	error = gfs2_quota_check(dip, uid, gid);
536 	if (error)
537 		goto out_quota;
538 
539 	error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
540 	if (error)
541 		goto out_quota;
542 
543 	init_dinode(dip, gl, inum, mode, uid, gid, generation, dev, symname, size, bhp);
544 	gfs2_quota_change(dip, +1, uid, gid);
545 	gfs2_trans_end(sdp);
546 
547 out_quota:
548 	gfs2_quota_unlock(dip);
549 out:
550 	gfs2_alloc_put(dip);
551 	return error;
552 }
553 
554 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
555 		       struct gfs2_inode *ip)
556 {
557 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
558 	struct gfs2_alloc *al;
559 	int alloc_required;
560 	struct buffer_head *dibh;
561 	int error;
562 
563 	al = gfs2_alloc_get(dip);
564 	if (!al)
565 		return -ENOMEM;
566 
567 	error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
568 	if (error)
569 		goto fail;
570 
571 	error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
572 	if (alloc_required < 0)
573 		goto fail_quota_locks;
574 	if (alloc_required) {
575 		error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
576 		if (error)
577 			goto fail_quota_locks;
578 
579 		al->al_requested = sdp->sd_max_dirres;
580 
581 		error = gfs2_inplace_reserve(dip);
582 		if (error)
583 			goto fail_quota_locks;
584 
585 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
586 					 dip->i_rgd->rd_length +
587 					 2 * RES_DINODE +
588 					 RES_STATFS + RES_QUOTA, 0);
589 		if (error)
590 			goto fail_ipreserv;
591 	} else {
592 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
593 		if (error)
594 			goto fail_quota_locks;
595 	}
596 
597 	error = gfs2_dir_add(&dip->i_inode, name, ip);
598 	if (error)
599 		goto fail_end_trans;
600 
601 	error = gfs2_meta_inode_buffer(ip, &dibh);
602 	if (error)
603 		goto fail_end_trans;
604 	inc_nlink(&ip->i_inode);
605 	if (S_ISDIR(ip->i_inode.i_mode))
606 		inc_nlink(&ip->i_inode);
607 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
608 	gfs2_dinode_out(ip, dibh->b_data);
609 	brelse(dibh);
610 	return 0;
611 
612 fail_end_trans:
613 	gfs2_trans_end(sdp);
614 
615 fail_ipreserv:
616 	gfs2_inplace_release(dip);
617 
618 fail_quota_locks:
619 	gfs2_quota_unlock(dip);
620 
621 fail:
622 	gfs2_alloc_put(dip);
623 	return error;
624 }
625 
626 int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
627 		    void *fs_info)
628 {
629 	const struct xattr *xattr;
630 	int err = 0;
631 
632 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
633 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
634 				       xattr->value_len, 0,
635 				       GFS2_EATYPE_SECURITY);
636 		if (err < 0)
637 			break;
638 	}
639 	return err;
640 }
641 
642 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
643 			      const struct qstr *qstr)
644 {
645 	return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
646 					    &gfs2_initxattrs, NULL);
647 }
648 
649 /**
650  * gfs2_create_inode - Create a new inode
651  * @dir: The parent directory
652  * @dentry: The new dentry
653  * @mode: The permissions on the new inode
654  * @dev: For device nodes, this is the device number
655  * @symname: For symlinks, this is the link destination
656  * @size: The initial size of the inode (ignored for directories)
657  *
658  * Returns: 0 on success, or error code
659  */
660 
661 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
662 			     unsigned int mode, dev_t dev, const char *symname,
663 			     unsigned int size, int excl)
664 {
665 	const struct qstr *name = &dentry->d_name;
666 	struct gfs2_holder ghs[2];
667 	struct inode *inode = NULL;
668 	struct gfs2_inode *dip = GFS2_I(dir);
669 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
670 	struct gfs2_inum_host inum = { .no_addr = 0, .no_formal_ino = 0 };
671 	int error;
672 	u64 generation;
673 	struct buffer_head *bh = NULL;
674 
675 	if (!name->len || name->len > GFS2_FNAMESIZE)
676 		return -ENAMETOOLONG;
677 
678 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
679 	if (error)
680 		goto fail;
681 
682 	error = create_ok(dip, name, mode);
683 	if ((error == -EEXIST) && S_ISREG(mode) && !excl) {
684 		inode = gfs2_lookupi(dir, &dentry->d_name, 0);
685 		gfs2_glock_dq_uninit(ghs);
686 		d_instantiate(dentry, inode);
687 		return IS_ERR(inode) ? PTR_ERR(inode) : 0;
688 	}
689 	if (error)
690 		goto fail_gunlock;
691 
692 	error = alloc_dinode(dip, &inum.no_addr, &generation);
693 	if (error)
694 		goto fail_gunlock;
695 	inum.no_formal_ino = generation;
696 
697 	error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops,
698 				  LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
699 	if (error)
700 		goto fail_gunlock;
701 
702 	error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev, symname, size, &bh);
703 	if (error)
704 		goto fail_gunlock2;
705 
706 	inode = gfs2_inode_lookup(dir->i_sb, IF2DT(mode), inum.no_addr,
707 				  inum.no_formal_ino, 0);
708 	if (IS_ERR(inode))
709 		goto fail_gunlock2;
710 
711 	error = gfs2_inode_refresh(GFS2_I(inode));
712 	if (error)
713 		goto fail_gunlock2;
714 
715 	error = gfs2_acl_create(dip, inode);
716 	if (error)
717 		goto fail_gunlock2;
718 
719 	error = gfs2_security_init(dip, GFS2_I(inode), name);
720 	if (error)
721 		goto fail_gunlock2;
722 
723 	error = link_dinode(dip, name, GFS2_I(inode));
724 	if (error)
725 		goto fail_gunlock2;
726 
727 	if (bh)
728 		brelse(bh);
729 
730 	gfs2_trans_end(sdp);
731 	gfs2_inplace_release(dip);
732 	gfs2_quota_unlock(dip);
733 	gfs2_alloc_put(dip);
734 	mark_inode_dirty(inode);
735 	gfs2_glock_dq_uninit_m(2, ghs);
736 	d_instantiate(dentry, inode);
737 	return 0;
738 
739 fail_gunlock2:
740 	gfs2_glock_dq_uninit(ghs + 1);
741 fail_gunlock:
742 	gfs2_glock_dq_uninit(ghs);
743 	if (inode && !IS_ERR(inode)) {
744 		set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
745 		iput(inode);
746 	}
747 fail:
748 	if (bh)
749 		brelse(bh);
750 	return error;
751 }
752 
753 /**
754  * gfs2_create - Create a file
755  * @dir: The directory in which to create the file
756  * @dentry: The dentry of the new file
757  * @mode: The mode of the new file
758  *
759  * Returns: errno
760  */
761 
762 static int gfs2_create(struct inode *dir, struct dentry *dentry,
763 		       int mode, struct nameidata *nd)
764 {
765 	int excl = 0;
766 	if (nd && (nd->flags & LOOKUP_EXCL))
767 		excl = 1;
768 	return gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0, excl);
769 }
770 
771 /**
772  * gfs2_lookup - Look up a filename in a directory and return its inode
773  * @dir: The directory inode
774  * @dentry: The dentry of the new inode
775  * @nd: passed from Linux VFS, ignored by us
776  *
777  * Called by the VFS layer. Lock dir and call gfs2_lookupi()
778  *
779  * Returns: errno
780  */
781 
782 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
783 				  struct nameidata *nd)
784 {
785 	struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
786 	if (inode && !IS_ERR(inode)) {
787 		struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
788 		struct gfs2_holder gh;
789 		int error;
790 		error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
791 		if (error) {
792 			iput(inode);
793 			return ERR_PTR(error);
794 		}
795 		gfs2_glock_dq_uninit(&gh);
796 	}
797 	return d_splice_alias(inode, dentry);
798 }
799 
800 /**
801  * gfs2_link - Link to a file
802  * @old_dentry: The inode to link
803  * @dir: Add link to this directory
804  * @dentry: The name of the link
805  *
806  * Link the inode in "old_dentry" into the directory "dir" with the
807  * name in "dentry".
808  *
809  * Returns: errno
810  */
811 
812 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
813 		     struct dentry *dentry)
814 {
815 	struct gfs2_inode *dip = GFS2_I(dir);
816 	struct gfs2_sbd *sdp = GFS2_SB(dir);
817 	struct inode *inode = old_dentry->d_inode;
818 	struct gfs2_inode *ip = GFS2_I(inode);
819 	struct gfs2_holder ghs[2];
820 	struct buffer_head *dibh;
821 	int alloc_required;
822 	int error;
823 
824 	if (S_ISDIR(inode->i_mode))
825 		return -EPERM;
826 
827 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
828 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
829 
830 	error = gfs2_glock_nq(ghs); /* parent */
831 	if (error)
832 		goto out_parent;
833 
834 	error = gfs2_glock_nq(ghs + 1); /* child */
835 	if (error)
836 		goto out_child;
837 
838 	error = -ENOENT;
839 	if (inode->i_nlink == 0)
840 		goto out_gunlock;
841 
842 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
843 	if (error)
844 		goto out_gunlock;
845 
846 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
847 	switch (error) {
848 	case -ENOENT:
849 		break;
850 	case 0:
851 		error = -EEXIST;
852 	default:
853 		goto out_gunlock;
854 	}
855 
856 	error = -EINVAL;
857 	if (!dip->i_inode.i_nlink)
858 		goto out_gunlock;
859 	error = -EFBIG;
860 	if (dip->i_entries == (u32)-1)
861 		goto out_gunlock;
862 	error = -EPERM;
863 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
864 		goto out_gunlock;
865 	error = -EINVAL;
866 	if (!ip->i_inode.i_nlink)
867 		goto out_gunlock;
868 	error = -EMLINK;
869 	if (ip->i_inode.i_nlink == (u32)-1)
870 		goto out_gunlock;
871 
872 	alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
873 	if (error < 0)
874 		goto out_gunlock;
875 	error = 0;
876 
877 	if (alloc_required) {
878 		struct gfs2_alloc *al = gfs2_alloc_get(dip);
879 		if (!al) {
880 			error = -ENOMEM;
881 			goto out_gunlock;
882 		}
883 
884 		error = gfs2_quota_lock_check(dip);
885 		if (error)
886 			goto out_alloc;
887 
888 		al->al_requested = sdp->sd_max_dirres;
889 
890 		error = gfs2_inplace_reserve(dip);
891 		if (error)
892 			goto out_gunlock_q;
893 
894 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
895 					 gfs2_rg_blocks(dip) +
896 					 2 * RES_DINODE + RES_STATFS +
897 					 RES_QUOTA, 0);
898 		if (error)
899 			goto out_ipres;
900 	} else {
901 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
902 		if (error)
903 			goto out_ipres;
904 	}
905 
906 	error = gfs2_meta_inode_buffer(ip, &dibh);
907 	if (error)
908 		goto out_end_trans;
909 
910 	error = gfs2_dir_add(dir, &dentry->d_name, ip);
911 	if (error)
912 		goto out_brelse;
913 
914 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
915 	inc_nlink(&ip->i_inode);
916 	ip->i_inode.i_ctime = CURRENT_TIME;
917 	ihold(inode);
918 	d_instantiate(dentry, inode);
919 	mark_inode_dirty(inode);
920 
921 out_brelse:
922 	brelse(dibh);
923 out_end_trans:
924 	gfs2_trans_end(sdp);
925 out_ipres:
926 	if (alloc_required)
927 		gfs2_inplace_release(dip);
928 out_gunlock_q:
929 	if (alloc_required)
930 		gfs2_quota_unlock(dip);
931 out_alloc:
932 	if (alloc_required)
933 		gfs2_alloc_put(dip);
934 out_gunlock:
935 	gfs2_glock_dq(ghs + 1);
936 out_child:
937 	gfs2_glock_dq(ghs);
938 out_parent:
939 	gfs2_holder_uninit(ghs);
940 	gfs2_holder_uninit(ghs + 1);
941 	return error;
942 }
943 
944 /*
945  * gfs2_unlink_ok - check to see that a inode is still in a directory
946  * @dip: the directory
947  * @name: the name of the file
948  * @ip: the inode
949  *
950  * Assumes that the lock on (at least) @dip is held.
951  *
952  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
953  */
954 
955 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
956 			  const struct gfs2_inode *ip)
957 {
958 	int error;
959 
960 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
961 		return -EPERM;
962 
963 	if ((dip->i_inode.i_mode & S_ISVTX) &&
964 	    dip->i_inode.i_uid != current_fsuid() &&
965 	    ip->i_inode.i_uid != current_fsuid() && !capable(CAP_FOWNER))
966 		return -EPERM;
967 
968 	if (IS_APPEND(&dip->i_inode))
969 		return -EPERM;
970 
971 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
972 	if (error)
973 		return error;
974 
975 	error = gfs2_dir_check(&dip->i_inode, name, ip);
976 	if (error)
977 		return error;
978 
979 	return 0;
980 }
981 
982 /**
983  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
984  * @dip: The parent directory
985  * @name: The name of the entry in the parent directory
986  * @bh: The inode buffer for the inode to be removed
987  * @inode: The inode to be removed
988  *
989  * Called with all the locks and in a transaction. This will only be
990  * called for a directory after it has been checked to ensure it is empty.
991  *
992  * Returns: 0 on success, or an error
993  */
994 
995 static int gfs2_unlink_inode(struct gfs2_inode *dip,
996 			     const struct dentry *dentry,
997 			     struct buffer_head *bh)
998 {
999 	struct inode *inode = dentry->d_inode;
1000 	struct gfs2_inode *ip = GFS2_I(inode);
1001 	int error;
1002 
1003 	error = gfs2_dir_del(dip, dentry);
1004 	if (error)
1005 		return error;
1006 
1007 	ip->i_entries = 0;
1008 	inode->i_ctime = CURRENT_TIME;
1009 	if (S_ISDIR(inode->i_mode))
1010 		clear_nlink(inode);
1011 	else
1012 		drop_nlink(inode);
1013 	mark_inode_dirty(inode);
1014 	if (inode->i_nlink == 0)
1015 		gfs2_unlink_di(inode);
1016 	return 0;
1017 }
1018 
1019 
1020 /**
1021  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1022  * @dir: The inode of the directory containing the inode to unlink
1023  * @dentry: The file itself
1024  *
1025  * This routine uses the type of the inode as a flag to figure out
1026  * whether this is an unlink or an rmdir.
1027  *
1028  * Returns: errno
1029  */
1030 
1031 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1032 {
1033 	struct gfs2_inode *dip = GFS2_I(dir);
1034 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1035 	struct inode *inode = dentry->d_inode;
1036 	struct gfs2_inode *ip = GFS2_I(inode);
1037 	struct buffer_head *bh;
1038 	struct gfs2_holder ghs[3];
1039 	struct gfs2_rgrpd *rgd;
1040 	int error;
1041 
1042 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1043 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1044 
1045 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
1046 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1047 
1048 
1049 	error = gfs2_glock_nq(ghs); /* parent */
1050 	if (error)
1051 		goto out_parent;
1052 
1053 	error = gfs2_glock_nq(ghs + 1); /* child */
1054 	if (error)
1055 		goto out_child;
1056 
1057 	error = -ENOENT;
1058 	if (inode->i_nlink == 0)
1059 		goto out_rgrp;
1060 
1061 	if (S_ISDIR(inode->i_mode)) {
1062 		error = -ENOTEMPTY;
1063 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1064 			goto out_rgrp;
1065 	}
1066 
1067 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1068 	if (error)
1069 		goto out_rgrp;
1070 
1071 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1072 	if (error)
1073 		goto out_gunlock;
1074 
1075 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1076 	if (error)
1077 		goto out_gunlock;
1078 
1079 	error = gfs2_meta_inode_buffer(ip, &bh);
1080 	if (error)
1081 		goto out_end_trans;
1082 
1083 	error = gfs2_unlink_inode(dip, dentry, bh);
1084 	brelse(bh);
1085 
1086 out_end_trans:
1087 	gfs2_trans_end(sdp);
1088 out_gunlock:
1089 	gfs2_glock_dq(ghs + 2);
1090 out_rgrp:
1091 	gfs2_holder_uninit(ghs + 2);
1092 	gfs2_glock_dq(ghs + 1);
1093 out_child:
1094 	gfs2_holder_uninit(ghs + 1);
1095 	gfs2_glock_dq(ghs);
1096 out_parent:
1097 	gfs2_holder_uninit(ghs);
1098 	return error;
1099 }
1100 
1101 /**
1102  * gfs2_symlink - Create a symlink
1103  * @dir: The directory to create the symlink in
1104  * @dentry: The dentry to put the symlink in
1105  * @symname: The thing which the link points to
1106  *
1107  * Returns: errno
1108  */
1109 
1110 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1111 			const char *symname)
1112 {
1113 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1114 	unsigned int size;
1115 
1116 	size = strlen(symname);
1117 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1118 		return -ENAMETOOLONG;
1119 
1120 	return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1121 }
1122 
1123 /**
1124  * gfs2_mkdir - Make a directory
1125  * @dir: The parent directory of the new one
1126  * @dentry: The dentry of the new directory
1127  * @mode: The mode of the new directory
1128  *
1129  * Returns: errno
1130  */
1131 
1132 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1133 {
1134 	return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, 0, 0);
1135 }
1136 
1137 /**
1138  * gfs2_mknod - Make a special file
1139  * @dir: The directory in which the special file will reside
1140  * @dentry: The dentry of the special file
1141  * @mode: The mode of the special file
1142  * @dev: The device specification of the special file
1143  *
1144  */
1145 
1146 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, int mode,
1147 		      dev_t dev)
1148 {
1149 	return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0, 0);
1150 }
1151 
1152 /*
1153  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1154  * @this: move this
1155  * @to: to here
1156  *
1157  * Follow @to back to the root and make sure we don't encounter @this
1158  * Assumes we already hold the rename lock.
1159  *
1160  * Returns: errno
1161  */
1162 
1163 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1164 {
1165 	struct inode *dir = &to->i_inode;
1166 	struct super_block *sb = dir->i_sb;
1167 	struct inode *tmp;
1168 	int error = 0;
1169 
1170 	igrab(dir);
1171 
1172 	for (;;) {
1173 		if (dir == &this->i_inode) {
1174 			error = -EINVAL;
1175 			break;
1176 		}
1177 		if (dir == sb->s_root->d_inode) {
1178 			error = 0;
1179 			break;
1180 		}
1181 
1182 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1183 		if (IS_ERR(tmp)) {
1184 			error = PTR_ERR(tmp);
1185 			break;
1186 		}
1187 
1188 		iput(dir);
1189 		dir = tmp;
1190 	}
1191 
1192 	iput(dir);
1193 
1194 	return error;
1195 }
1196 
1197 /**
1198  * gfs2_rename - Rename a file
1199  * @odir: Parent directory of old file name
1200  * @odentry: The old dentry of the file
1201  * @ndir: Parent directory of new file name
1202  * @ndentry: The new dentry of the file
1203  *
1204  * Returns: errno
1205  */
1206 
1207 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1208 		       struct inode *ndir, struct dentry *ndentry)
1209 {
1210 	struct gfs2_inode *odip = GFS2_I(odir);
1211 	struct gfs2_inode *ndip = GFS2_I(ndir);
1212 	struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1213 	struct gfs2_inode *nip = NULL;
1214 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1215 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1216 	struct gfs2_rgrpd *nrgd;
1217 	unsigned int num_gh;
1218 	int dir_rename = 0;
1219 	int alloc_required = 0;
1220 	unsigned int x;
1221 	int error;
1222 
1223 	if (ndentry->d_inode) {
1224 		nip = GFS2_I(ndentry->d_inode);
1225 		if (ip == nip)
1226 			return 0;
1227 	}
1228 
1229 	if (odip != ndip) {
1230 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1231 					   0, &r_gh);
1232 		if (error)
1233 			goto out;
1234 
1235 		if (S_ISDIR(ip->i_inode.i_mode)) {
1236 			dir_rename = 1;
1237 			/* don't move a dirctory into it's subdir */
1238 			error = gfs2_ok_to_move(ip, ndip);
1239 			if (error)
1240 				goto out_gunlock_r;
1241 		}
1242 	}
1243 
1244 	num_gh = 1;
1245 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1246 	if (odip != ndip) {
1247 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1248 		num_gh++;
1249 	}
1250 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1251 	num_gh++;
1252 
1253 	if (nip) {
1254 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1255 		num_gh++;
1256 		/* grab the resource lock for unlink flag twiddling
1257 		 * this is the case of the target file already existing
1258 		 * so we unlink before doing the rename
1259 		 */
1260 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr);
1261 		if (nrgd)
1262 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1263 	}
1264 
1265 	for (x = 0; x < num_gh; x++) {
1266 		error = gfs2_glock_nq(ghs + x);
1267 		if (error)
1268 			goto out_gunlock;
1269 	}
1270 
1271 	error = -ENOENT;
1272 	if (ip->i_inode.i_nlink == 0)
1273 		goto out_gunlock;
1274 
1275 	/* Check out the old directory */
1276 
1277 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1278 	if (error)
1279 		goto out_gunlock;
1280 
1281 	/* Check out the new directory */
1282 
1283 	if (nip) {
1284 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1285 		if (error)
1286 			goto out_gunlock;
1287 
1288 		if (nip->i_inode.i_nlink == 0) {
1289 			error = -EAGAIN;
1290 			goto out_gunlock;
1291 		}
1292 
1293 		if (S_ISDIR(nip->i_inode.i_mode)) {
1294 			if (nip->i_entries < 2) {
1295 				gfs2_consist_inode(nip);
1296 				error = -EIO;
1297 				goto out_gunlock;
1298 			}
1299 			if (nip->i_entries > 2) {
1300 				error = -ENOTEMPTY;
1301 				goto out_gunlock;
1302 			}
1303 		}
1304 	} else {
1305 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1306 		if (error)
1307 			goto out_gunlock;
1308 
1309 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1310 		switch (error) {
1311 		case -ENOENT:
1312 			error = 0;
1313 			break;
1314 		case 0:
1315 			error = -EEXIST;
1316 		default:
1317 			goto out_gunlock;
1318 		};
1319 
1320 		if (odip != ndip) {
1321 			if (!ndip->i_inode.i_nlink) {
1322 				error = -ENOENT;
1323 				goto out_gunlock;
1324 			}
1325 			if (ndip->i_entries == (u32)-1) {
1326 				error = -EFBIG;
1327 				goto out_gunlock;
1328 			}
1329 			if (S_ISDIR(ip->i_inode.i_mode) &&
1330 			    ndip->i_inode.i_nlink == (u32)-1) {
1331 				error = -EMLINK;
1332 				goto out_gunlock;
1333 			}
1334 		}
1335 	}
1336 
1337 	/* Check out the dir to be renamed */
1338 
1339 	if (dir_rename) {
1340 		error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1341 		if (error)
1342 			goto out_gunlock;
1343 	}
1344 
1345 	if (nip == NULL)
1346 		alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1347 	error = alloc_required;
1348 	if (error < 0)
1349 		goto out_gunlock;
1350 	error = 0;
1351 
1352 	if (alloc_required) {
1353 		struct gfs2_alloc *al = gfs2_alloc_get(ndip);
1354 		if (!al) {
1355 			error = -ENOMEM;
1356 			goto out_gunlock;
1357 		}
1358 
1359 		error = gfs2_quota_lock_check(ndip);
1360 		if (error)
1361 			goto out_alloc;
1362 
1363 		al->al_requested = sdp->sd_max_dirres;
1364 
1365 		error = gfs2_inplace_reserve(ndip);
1366 		if (error)
1367 			goto out_gunlock_q;
1368 
1369 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1370 					 gfs2_rg_blocks(ndip) +
1371 					 4 * RES_DINODE + 4 * RES_LEAF +
1372 					 RES_STATFS + RES_QUOTA + 4, 0);
1373 		if (error)
1374 			goto out_ipreserv;
1375 	} else {
1376 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1377 					 5 * RES_LEAF + 4, 0);
1378 		if (error)
1379 			goto out_gunlock;
1380 	}
1381 
1382 	/* Remove the target file, if it exists */
1383 
1384 	if (nip) {
1385 		struct buffer_head *bh;
1386 		error = gfs2_meta_inode_buffer(nip, &bh);
1387 		if (error)
1388 			goto out_end_trans;
1389 		error = gfs2_unlink_inode(ndip, ndentry, bh);
1390 		brelse(bh);
1391 	}
1392 
1393 	if (dir_rename) {
1394 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1395 		if (error)
1396 			goto out_end_trans;
1397 	} else {
1398 		struct buffer_head *dibh;
1399 		error = gfs2_meta_inode_buffer(ip, &dibh);
1400 		if (error)
1401 			goto out_end_trans;
1402 		ip->i_inode.i_ctime = CURRENT_TIME;
1403 		gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1404 		gfs2_dinode_out(ip, dibh->b_data);
1405 		brelse(dibh);
1406 	}
1407 
1408 	error = gfs2_dir_del(odip, odentry);
1409 	if (error)
1410 		goto out_end_trans;
1411 
1412 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1413 	if (error)
1414 		goto out_end_trans;
1415 
1416 out_end_trans:
1417 	gfs2_trans_end(sdp);
1418 out_ipreserv:
1419 	if (alloc_required)
1420 		gfs2_inplace_release(ndip);
1421 out_gunlock_q:
1422 	if (alloc_required)
1423 		gfs2_quota_unlock(ndip);
1424 out_alloc:
1425 	if (alloc_required)
1426 		gfs2_alloc_put(ndip);
1427 out_gunlock:
1428 	while (x--) {
1429 		gfs2_glock_dq(ghs + x);
1430 		gfs2_holder_uninit(ghs + x);
1431 	}
1432 out_gunlock_r:
1433 	if (r_gh.gh_gl)
1434 		gfs2_glock_dq_uninit(&r_gh);
1435 out:
1436 	return error;
1437 }
1438 
1439 /**
1440  * gfs2_follow_link - Follow a symbolic link
1441  * @dentry: The dentry of the link
1442  * @nd: Data that we pass to vfs_follow_link()
1443  *
1444  * This can handle symlinks of any size.
1445  *
1446  * Returns: 0 on success or error code
1447  */
1448 
1449 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1450 {
1451 	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1452 	struct gfs2_holder i_gh;
1453 	struct buffer_head *dibh;
1454 	unsigned int size;
1455 	char *buf;
1456 	int error;
1457 
1458 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1459 	error = gfs2_glock_nq(&i_gh);
1460 	if (error) {
1461 		gfs2_holder_uninit(&i_gh);
1462 		nd_set_link(nd, ERR_PTR(error));
1463 		return NULL;
1464 	}
1465 
1466 	size = (unsigned int)i_size_read(&ip->i_inode);
1467 	if (size == 0) {
1468 		gfs2_consist_inode(ip);
1469 		buf = ERR_PTR(-EIO);
1470 		goto out;
1471 	}
1472 
1473 	error = gfs2_meta_inode_buffer(ip, &dibh);
1474 	if (error) {
1475 		buf = ERR_PTR(error);
1476 		goto out;
1477 	}
1478 
1479 	buf = kzalloc(size + 1, GFP_NOFS);
1480 	if (!buf)
1481 		buf = ERR_PTR(-ENOMEM);
1482 	else
1483 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1484 	brelse(dibh);
1485 out:
1486 	gfs2_glock_dq_uninit(&i_gh);
1487 	nd_set_link(nd, buf);
1488 	return NULL;
1489 }
1490 
1491 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1492 {
1493 	char *s = nd_get_link(nd);
1494 	if (!IS_ERR(s))
1495 		kfree(s);
1496 }
1497 
1498 /**
1499  * gfs2_permission -
1500  * @inode: The inode
1501  * @mask: The mask to be tested
1502  * @flags: Indicates whether this is an RCU path walk or not
1503  *
1504  * This may be called from the VFS directly, or from within GFS2 with the
1505  * inode locked, so we look to see if the glock is already locked and only
1506  * lock the glock if its not already been done.
1507  *
1508  * Returns: errno
1509  */
1510 
1511 int gfs2_permission(struct inode *inode, int mask)
1512 {
1513 	struct gfs2_inode *ip;
1514 	struct gfs2_holder i_gh;
1515 	int error;
1516 	int unlock = 0;
1517 
1518 
1519 	ip = GFS2_I(inode);
1520 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1521 		if (mask & MAY_NOT_BLOCK)
1522 			return -ECHILD;
1523 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1524 		if (error)
1525 			return error;
1526 		unlock = 1;
1527 	}
1528 
1529 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1530 		error = -EACCES;
1531 	else
1532 		error = generic_permission(inode, mask);
1533 	if (unlock)
1534 		gfs2_glock_dq_uninit(&i_gh);
1535 
1536 	return error;
1537 }
1538 
1539 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1540 {
1541 	setattr_copy(inode, attr);
1542 	mark_inode_dirty(inode);
1543 	return 0;
1544 }
1545 
1546 /**
1547  * gfs2_setattr_simple -
1548  * @ip:
1549  * @attr:
1550  *
1551  * Returns: errno
1552  */
1553 
1554 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1555 {
1556 	int error;
1557 
1558 	if (current->journal_info)
1559 		return __gfs2_setattr_simple(inode, attr);
1560 
1561 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1562 	if (error)
1563 		return error;
1564 
1565 	error = __gfs2_setattr_simple(inode, attr);
1566 	gfs2_trans_end(GFS2_SB(inode));
1567 	return error;
1568 }
1569 
1570 static int setattr_chown(struct inode *inode, struct iattr *attr)
1571 {
1572 	struct gfs2_inode *ip = GFS2_I(inode);
1573 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1574 	u32 ouid, ogid, nuid, ngid;
1575 	int error;
1576 
1577 	ouid = inode->i_uid;
1578 	ogid = inode->i_gid;
1579 	nuid = attr->ia_uid;
1580 	ngid = attr->ia_gid;
1581 
1582 	if (!(attr->ia_valid & ATTR_UID) || ouid == nuid)
1583 		ouid = nuid = NO_QUOTA_CHANGE;
1584 	if (!(attr->ia_valid & ATTR_GID) || ogid == ngid)
1585 		ogid = ngid = NO_QUOTA_CHANGE;
1586 
1587 	if (!gfs2_alloc_get(ip))
1588 		return -ENOMEM;
1589 
1590 	error = gfs2_quota_lock(ip, nuid, ngid);
1591 	if (error)
1592 		goto out_alloc;
1593 
1594 	if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1595 		error = gfs2_quota_check(ip, nuid, ngid);
1596 		if (error)
1597 			goto out_gunlock_q;
1598 	}
1599 
1600 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1601 	if (error)
1602 		goto out_gunlock_q;
1603 
1604 	error = gfs2_setattr_simple(inode, attr);
1605 	if (error)
1606 		goto out_end_trans;
1607 
1608 	if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1609 		u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1610 		gfs2_quota_change(ip, -blocks, ouid, ogid);
1611 		gfs2_quota_change(ip, blocks, nuid, ngid);
1612 	}
1613 
1614 out_end_trans:
1615 	gfs2_trans_end(sdp);
1616 out_gunlock_q:
1617 	gfs2_quota_unlock(ip);
1618 out_alloc:
1619 	gfs2_alloc_put(ip);
1620 	return error;
1621 }
1622 
1623 /**
1624  * gfs2_setattr - Change attributes on an inode
1625  * @dentry: The dentry which is changing
1626  * @attr: The structure describing the change
1627  *
1628  * The VFS layer wants to change one or more of an inodes attributes.  Write
1629  * that change out to disk.
1630  *
1631  * Returns: errno
1632  */
1633 
1634 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1635 {
1636 	struct inode *inode = dentry->d_inode;
1637 	struct gfs2_inode *ip = GFS2_I(inode);
1638 	struct gfs2_holder i_gh;
1639 	int error;
1640 
1641 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1642 	if (error)
1643 		return error;
1644 
1645 	error = -EPERM;
1646 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1647 		goto out;
1648 
1649 	error = inode_change_ok(inode, attr);
1650 	if (error)
1651 		goto out;
1652 
1653 	if (attr->ia_valid & ATTR_SIZE)
1654 		error = gfs2_setattr_size(inode, attr->ia_size);
1655 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1656 		error = setattr_chown(inode, attr);
1657 	else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1658 		error = gfs2_acl_chmod(ip, attr);
1659 	else
1660 		error = gfs2_setattr_simple(inode, attr);
1661 
1662 out:
1663 	if (!error)
1664 		mark_inode_dirty(inode);
1665 	gfs2_glock_dq_uninit(&i_gh);
1666 	return error;
1667 }
1668 
1669 /**
1670  * gfs2_getattr - Read out an inode's attributes
1671  * @mnt: The vfsmount the inode is being accessed from
1672  * @dentry: The dentry to stat
1673  * @stat: The inode's stats
1674  *
1675  * This may be called from the VFS directly, or from within GFS2 with the
1676  * inode locked, so we look to see if the glock is already locked and only
1677  * lock the glock if its not already been done. Note that its the NFS
1678  * readdirplus operation which causes this to be called (from filldir)
1679  * with the glock already held.
1680  *
1681  * Returns: errno
1682  */
1683 
1684 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1685 			struct kstat *stat)
1686 {
1687 	struct inode *inode = dentry->d_inode;
1688 	struct gfs2_inode *ip = GFS2_I(inode);
1689 	struct gfs2_holder gh;
1690 	int error;
1691 	int unlock = 0;
1692 
1693 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1694 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1695 		if (error)
1696 			return error;
1697 		unlock = 1;
1698 	}
1699 
1700 	generic_fillattr(inode, stat);
1701 	if (unlock)
1702 		gfs2_glock_dq_uninit(&gh);
1703 
1704 	return 0;
1705 }
1706 
1707 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1708 			 const void *data, size_t size, int flags)
1709 {
1710 	struct inode *inode = dentry->d_inode;
1711 	struct gfs2_inode *ip = GFS2_I(inode);
1712 	struct gfs2_holder gh;
1713 	int ret;
1714 
1715 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1716 	ret = gfs2_glock_nq(&gh);
1717 	if (ret == 0) {
1718 		ret = generic_setxattr(dentry, name, data, size, flags);
1719 		gfs2_glock_dq(&gh);
1720 	}
1721 	gfs2_holder_uninit(&gh);
1722 	return ret;
1723 }
1724 
1725 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1726 			     void *data, size_t size)
1727 {
1728 	struct inode *inode = dentry->d_inode;
1729 	struct gfs2_inode *ip = GFS2_I(inode);
1730 	struct gfs2_holder gh;
1731 	int ret;
1732 
1733 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1734 	ret = gfs2_glock_nq(&gh);
1735 	if (ret == 0) {
1736 		ret = generic_getxattr(dentry, name, data, size);
1737 		gfs2_glock_dq(&gh);
1738 	}
1739 	gfs2_holder_uninit(&gh);
1740 	return ret;
1741 }
1742 
1743 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1744 {
1745 	struct inode *inode = dentry->d_inode;
1746 	struct gfs2_inode *ip = GFS2_I(inode);
1747 	struct gfs2_holder gh;
1748 	int ret;
1749 
1750 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1751 	ret = gfs2_glock_nq(&gh);
1752 	if (ret == 0) {
1753 		ret = generic_removexattr(dentry, name);
1754 		gfs2_glock_dq(&gh);
1755 	}
1756 	gfs2_holder_uninit(&gh);
1757 	return ret;
1758 }
1759 
1760 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1761 		       u64 start, u64 len)
1762 {
1763 	struct gfs2_inode *ip = GFS2_I(inode);
1764 	struct gfs2_holder gh;
1765 	int ret;
1766 
1767 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1768 	if (ret)
1769 		return ret;
1770 
1771 	mutex_lock(&inode->i_mutex);
1772 
1773 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1774 	if (ret)
1775 		goto out;
1776 
1777 	if (gfs2_is_stuffed(ip)) {
1778 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1779 		u64 size = i_size_read(inode);
1780 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1781 			    FIEMAP_EXTENT_DATA_INLINE;
1782 		phys += sizeof(struct gfs2_dinode);
1783 		phys += start;
1784 		if (start + len > size)
1785 			len = size - start;
1786 		if (start < size)
1787 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1788 						      len, flags);
1789 		if (ret == 1)
1790 			ret = 0;
1791 	} else {
1792 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1793 					     gfs2_block_map);
1794 	}
1795 
1796 	gfs2_glock_dq_uninit(&gh);
1797 out:
1798 	mutex_unlock(&inode->i_mutex);
1799 	return ret;
1800 }
1801 
1802 const struct inode_operations gfs2_file_iops = {
1803 	.permission = gfs2_permission,
1804 	.setattr = gfs2_setattr,
1805 	.getattr = gfs2_getattr,
1806 	.setxattr = gfs2_setxattr,
1807 	.getxattr = gfs2_getxattr,
1808 	.listxattr = gfs2_listxattr,
1809 	.removexattr = gfs2_removexattr,
1810 	.fiemap = gfs2_fiemap,
1811 	.get_acl = gfs2_get_acl,
1812 };
1813 
1814 const struct inode_operations gfs2_dir_iops = {
1815 	.create = gfs2_create,
1816 	.lookup = gfs2_lookup,
1817 	.link = gfs2_link,
1818 	.unlink = gfs2_unlink,
1819 	.symlink = gfs2_symlink,
1820 	.mkdir = gfs2_mkdir,
1821 	.rmdir = gfs2_unlink,
1822 	.mknod = gfs2_mknod,
1823 	.rename = gfs2_rename,
1824 	.permission = gfs2_permission,
1825 	.setattr = gfs2_setattr,
1826 	.getattr = gfs2_getattr,
1827 	.setxattr = gfs2_setxattr,
1828 	.getxattr = gfs2_getxattr,
1829 	.listxattr = gfs2_listxattr,
1830 	.removexattr = gfs2_removexattr,
1831 	.fiemap = gfs2_fiemap,
1832 	.get_acl = gfs2_get_acl,
1833 };
1834 
1835 const struct inode_operations gfs2_symlink_iops = {
1836 	.readlink = generic_readlink,
1837 	.follow_link = gfs2_follow_link,
1838 	.put_link = gfs2_put_link,
1839 	.permission = gfs2_permission,
1840 	.setattr = gfs2_setattr,
1841 	.getattr = gfs2_getattr,
1842 	.setxattr = gfs2_setxattr,
1843 	.getxattr = gfs2_getxattr,
1844 	.listxattr = gfs2_listxattr,
1845 	.removexattr = gfs2_removexattr,
1846 	.fiemap = gfs2_fiemap,
1847 	.get_acl = gfs2_get_acl,
1848 };
1849 
1850