xref: /linux/fs/gfs2/inode.c (revision e8d235d4d8fb8957bae5f6ed4521115203a00d8b)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2011 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/namei.h>
15 #include <linux/mm.h>
16 #include <linux/xattr.h>
17 #include <linux/posix_acl.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/crc32.h>
20 #include <linux/fiemap.h>
21 #include <linux/security.h>
22 #include <asm/uaccess.h>
23 
24 #include "gfs2.h"
25 #include "incore.h"
26 #include "acl.h"
27 #include "bmap.h"
28 #include "dir.h"
29 #include "xattr.h"
30 #include "glock.h"
31 #include "inode.h"
32 #include "meta_io.h"
33 #include "quota.h"
34 #include "rgrp.h"
35 #include "trans.h"
36 #include "util.h"
37 #include "super.h"
38 #include "glops.h"
39 
40 struct gfs2_skip_data {
41 	u64 no_addr;
42 	int skipped;
43 	int non_block;
44 };
45 
46 static int iget_test(struct inode *inode, void *opaque)
47 {
48 	struct gfs2_inode *ip = GFS2_I(inode);
49 	struct gfs2_skip_data *data = opaque;
50 
51 	if (ip->i_no_addr == data->no_addr) {
52 		if (data->non_block &&
53 		    inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
54 			data->skipped = 1;
55 			return 0;
56 		}
57 		return 1;
58 	}
59 	return 0;
60 }
61 
62 static int iget_set(struct inode *inode, void *opaque)
63 {
64 	struct gfs2_inode *ip = GFS2_I(inode);
65 	struct gfs2_skip_data *data = opaque;
66 
67 	if (data->skipped)
68 		return -ENOENT;
69 	inode->i_ino = (unsigned long)(data->no_addr);
70 	ip->i_no_addr = data->no_addr;
71 	return 0;
72 }
73 
74 struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr, int non_block)
75 {
76 	unsigned long hash = (unsigned long)no_addr;
77 	struct gfs2_skip_data data;
78 
79 	data.no_addr = no_addr;
80 	data.skipped = 0;
81 	data.non_block = non_block;
82 	return ilookup5(sb, hash, iget_test, &data);
83 }
84 
85 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr,
86 			       int non_block)
87 {
88 	struct gfs2_skip_data data;
89 	unsigned long hash = (unsigned long)no_addr;
90 
91 	data.no_addr = no_addr;
92 	data.skipped = 0;
93 	data.non_block = non_block;
94 	return iget5_locked(sb, hash, iget_test, iget_set, &data);
95 }
96 
97 /**
98  * gfs2_set_iop - Sets inode operations
99  * @inode: The inode with correct i_mode filled in
100  *
101  * GFS2 lookup code fills in vfs inode contents based on info obtained
102  * from directory entry inside gfs2_inode_lookup().
103  */
104 
105 static void gfs2_set_iop(struct inode *inode)
106 {
107 	struct gfs2_sbd *sdp = GFS2_SB(inode);
108 	umode_t mode = inode->i_mode;
109 
110 	if (S_ISREG(mode)) {
111 		inode->i_op = &gfs2_file_iops;
112 		if (gfs2_localflocks(sdp))
113 			inode->i_fop = &gfs2_file_fops_nolock;
114 		else
115 			inode->i_fop = &gfs2_file_fops;
116 	} else if (S_ISDIR(mode)) {
117 		inode->i_op = &gfs2_dir_iops;
118 		if (gfs2_localflocks(sdp))
119 			inode->i_fop = &gfs2_dir_fops_nolock;
120 		else
121 			inode->i_fop = &gfs2_dir_fops;
122 	} else if (S_ISLNK(mode)) {
123 		inode->i_op = &gfs2_symlink_iops;
124 	} else {
125 		inode->i_op = &gfs2_file_iops;
126 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
127 	}
128 }
129 
130 /**
131  * gfs2_inode_lookup - Lookup an inode
132  * @sb: The super block
133  * @no_addr: The inode number
134  * @type: The type of the inode
135  * non_block: Can we block on inodes that are being freed?
136  *
137  * Returns: A VFS inode, or an error
138  */
139 
140 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
141 				u64 no_addr, u64 no_formal_ino, int non_block)
142 {
143 	struct inode *inode;
144 	struct gfs2_inode *ip;
145 	struct gfs2_glock *io_gl = NULL;
146 	int error;
147 
148 	inode = gfs2_iget(sb, no_addr, non_block);
149 	ip = GFS2_I(inode);
150 
151 	if (!inode)
152 		return ERR_PTR(-ENOMEM);
153 
154 	if (inode->i_state & I_NEW) {
155 		struct gfs2_sbd *sdp = GFS2_SB(inode);
156 		ip->i_no_formal_ino = no_formal_ino;
157 
158 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
159 		if (unlikely(error))
160 			goto fail;
161 		ip->i_gl->gl_object = ip;
162 
163 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
164 		if (unlikely(error))
165 			goto fail_put;
166 
167 		set_bit(GIF_INVALID, &ip->i_flags);
168 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
169 		if (unlikely(error))
170 			goto fail_iopen;
171 
172 		ip->i_iopen_gh.gh_gl->gl_object = ip;
173 		gfs2_glock_put(io_gl);
174 		io_gl = NULL;
175 
176 		if (type == DT_UNKNOWN) {
177 			/* Inode glock must be locked already */
178 			error = gfs2_inode_refresh(GFS2_I(inode));
179 			if (error)
180 				goto fail_refresh;
181 		} else {
182 			inode->i_mode = DT2IF(type);
183 		}
184 
185 		gfs2_set_iop(inode);
186 		unlock_new_inode(inode);
187 	}
188 
189 	return inode;
190 
191 fail_refresh:
192 	ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
193 	ip->i_iopen_gh.gh_gl->gl_object = NULL;
194 	gfs2_glock_dq_uninit(&ip->i_iopen_gh);
195 fail_iopen:
196 	if (io_gl)
197 		gfs2_glock_put(io_gl);
198 fail_put:
199 	ip->i_gl->gl_object = NULL;
200 	gfs2_glock_put(ip->i_gl);
201 fail:
202 	iget_failed(inode);
203 	return ERR_PTR(error);
204 }
205 
206 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
207 				  u64 *no_formal_ino, unsigned int blktype)
208 {
209 	struct super_block *sb = sdp->sd_vfs;
210 	struct gfs2_holder i_gh;
211 	struct inode *inode = NULL;
212 	int error;
213 
214 	/* Must not read in block until block type is verified */
215 	error = gfs2_glock_nq_num(sdp, no_addr, &gfs2_inode_glops,
216 				  LM_ST_EXCLUSIVE, GL_SKIP, &i_gh);
217 	if (error)
218 		return ERR_PTR(error);
219 
220 	error = gfs2_check_blk_type(sdp, no_addr, blktype);
221 	if (error)
222 		goto fail;
223 
224 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0, 1);
225 	if (IS_ERR(inode))
226 		goto fail;
227 
228 	/* Two extra checks for NFS only */
229 	if (no_formal_ino) {
230 		error = -ESTALE;
231 		if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
232 			goto fail_iput;
233 
234 		error = -EIO;
235 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
236 			goto fail_iput;
237 
238 		error = 0;
239 	}
240 
241 fail:
242 	gfs2_glock_dq_uninit(&i_gh);
243 	return error ? ERR_PTR(error) : inode;
244 fail_iput:
245 	iput(inode);
246 	goto fail;
247 }
248 
249 
250 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
251 {
252 	struct qstr qstr;
253 	struct inode *inode;
254 	gfs2_str2qstr(&qstr, name);
255 	inode = gfs2_lookupi(dip, &qstr, 1);
256 	/* gfs2_lookupi has inconsistent callers: vfs
257 	 * related routines expect NULL for no entry found,
258 	 * gfs2_lookup_simple callers expect ENOENT
259 	 * and do not check for NULL.
260 	 */
261 	if (inode == NULL)
262 		return ERR_PTR(-ENOENT);
263 	else
264 		return inode;
265 }
266 
267 
268 /**
269  * gfs2_lookupi - Look up a filename in a directory and return its inode
270  * @d_gh: An initialized holder for the directory glock
271  * @name: The name of the inode to look for
272  * @is_root: If 1, ignore the caller's permissions
273  * @i_gh: An uninitialized holder for the new inode glock
274  *
275  * This can be called via the VFS filldir function when NFS is doing
276  * a readdirplus and the inode which its intending to stat isn't
277  * already in cache. In this case we must not take the directory glock
278  * again, since the readdir call will have already taken that lock.
279  *
280  * Returns: errno
281  */
282 
283 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
284 			   int is_root)
285 {
286 	struct super_block *sb = dir->i_sb;
287 	struct gfs2_inode *dip = GFS2_I(dir);
288 	struct gfs2_holder d_gh;
289 	int error = 0;
290 	struct inode *inode = NULL;
291 	int unlock = 0;
292 
293 	if (!name->len || name->len > GFS2_FNAMESIZE)
294 		return ERR_PTR(-ENAMETOOLONG);
295 
296 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
297 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
298 	     dir == d_inode(sb->s_root))) {
299 		igrab(dir);
300 		return dir;
301 	}
302 
303 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
304 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
305 		if (error)
306 			return ERR_PTR(error);
307 		unlock = 1;
308 	}
309 
310 	if (!is_root) {
311 		error = gfs2_permission(dir, MAY_EXEC);
312 		if (error)
313 			goto out;
314 	}
315 
316 	inode = gfs2_dir_search(dir, name, false);
317 	if (IS_ERR(inode))
318 		error = PTR_ERR(inode);
319 out:
320 	if (unlock)
321 		gfs2_glock_dq_uninit(&d_gh);
322 	if (error == -ENOENT)
323 		return NULL;
324 	return inode ? inode : ERR_PTR(error);
325 }
326 
327 /**
328  * create_ok - OK to create a new on-disk inode here?
329  * @dip:  Directory in which dinode is to be created
330  * @name:  Name of new dinode
331  * @mode:
332  *
333  * Returns: errno
334  */
335 
336 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
337 		     umode_t mode)
338 {
339 	int error;
340 
341 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
342 	if (error)
343 		return error;
344 
345 	/*  Don't create entries in an unlinked directory  */
346 	if (!dip->i_inode.i_nlink)
347 		return -ENOENT;
348 
349 	if (dip->i_entries == (u32)-1)
350 		return -EFBIG;
351 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
352 		return -EMLINK;
353 
354 	return 0;
355 }
356 
357 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
358 			       struct inode *inode)
359 {
360 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
361 	    (dip->i_inode.i_mode & S_ISUID) &&
362 	    !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
363 		if (S_ISDIR(inode->i_mode))
364 			inode->i_mode |= S_ISUID;
365 		else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
366 			inode->i_mode &= ~07111;
367 		inode->i_uid = dip->i_inode.i_uid;
368 	} else
369 		inode->i_uid = current_fsuid();
370 
371 	if (dip->i_inode.i_mode & S_ISGID) {
372 		if (S_ISDIR(inode->i_mode))
373 			inode->i_mode |= S_ISGID;
374 		inode->i_gid = dip->i_inode.i_gid;
375 	} else
376 		inode->i_gid = current_fsgid();
377 }
378 
379 static int alloc_dinode(struct gfs2_inode *ip, u32 flags, unsigned *dblocks)
380 {
381 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
382 	struct gfs2_alloc_parms ap = { .target = *dblocks, .aflags = flags, };
383 	int error;
384 
385 	error = gfs2_quota_lock_check(ip, &ap);
386 	if (error)
387 		goto out;
388 
389 	error = gfs2_inplace_reserve(ip, &ap);
390 	if (error)
391 		goto out_quota;
392 
393 	error = gfs2_trans_begin(sdp, (*dblocks * RES_RG_BIT) + RES_STATFS + RES_QUOTA, 0);
394 	if (error)
395 		goto out_ipreserv;
396 
397 	error = gfs2_alloc_blocks(ip, &ip->i_no_addr, dblocks, 1, &ip->i_generation);
398 	ip->i_no_formal_ino = ip->i_generation;
399 	ip->i_inode.i_ino = ip->i_no_addr;
400 	ip->i_goal = ip->i_no_addr;
401 
402 	gfs2_trans_end(sdp);
403 
404 out_ipreserv:
405 	gfs2_inplace_release(ip);
406 out_quota:
407 	gfs2_quota_unlock(ip);
408 out:
409 	return error;
410 }
411 
412 static void gfs2_init_dir(struct buffer_head *dibh,
413 			  const struct gfs2_inode *parent)
414 {
415 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
416 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
417 
418 	gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
419 	dent->de_inum = di->di_num; /* already GFS2 endian */
420 	dent->de_type = cpu_to_be16(DT_DIR);
421 
422 	dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
423 	gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
424 	gfs2_inum_out(parent, dent);
425 	dent->de_type = cpu_to_be16(DT_DIR);
426 
427 }
428 
429 /**
430  * gfs2_init_xattr - Initialise an xattr block for a new inode
431  * @ip: The inode in question
432  *
433  * This sets up an empty xattr block for a new inode, ready to
434  * take any ACLs, LSM xattrs, etc.
435  */
436 
437 static void gfs2_init_xattr(struct gfs2_inode *ip)
438 {
439 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
440 	struct buffer_head *bh;
441 	struct gfs2_ea_header *ea;
442 
443 	bh = gfs2_meta_new(ip->i_gl, ip->i_eattr);
444 	gfs2_trans_add_meta(ip->i_gl, bh);
445 	gfs2_metatype_set(bh, GFS2_METATYPE_EA, GFS2_FORMAT_EA);
446 	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
447 
448 	ea = GFS2_EA_BH2FIRST(bh);
449 	ea->ea_rec_len = cpu_to_be32(sdp->sd_jbsize);
450 	ea->ea_type = GFS2_EATYPE_UNUSED;
451 	ea->ea_flags = GFS2_EAFLAG_LAST;
452 
453 	brelse(bh);
454 }
455 
456 /**
457  * init_dinode - Fill in a new dinode structure
458  * @dip: The directory this inode is being created in
459  * @ip: The inode
460  * @symname: The symlink destination (if a symlink)
461  * @bhp: The buffer head (returned to caller)
462  *
463  */
464 
465 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
466 			const char *symname)
467 {
468 	struct gfs2_dinode *di;
469 	struct buffer_head *dibh;
470 
471 	dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
472 	gfs2_trans_add_meta(ip->i_gl, dibh);
473 	di = (struct gfs2_dinode *)dibh->b_data;
474 	gfs2_dinode_out(ip, di);
475 
476 	di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
477 	di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
478 	di->__pad1 = 0;
479 	di->__pad2 = 0;
480 	di->__pad3 = 0;
481 	memset(&di->__pad4, 0, sizeof(di->__pad4));
482 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
483 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
484 
485 	switch(ip->i_inode.i_mode & S_IFMT) {
486 	case S_IFDIR:
487 		gfs2_init_dir(dibh, dip);
488 		break;
489 	case S_IFLNK:
490 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
491 		break;
492 	}
493 
494 	set_buffer_uptodate(dibh);
495 	brelse(dibh);
496 }
497 
498 /**
499  * gfs2_trans_da_blocks - Calculate number of blocks to link inode
500  * @dip: The directory we are linking into
501  * @da: The dir add information
502  * @nr_inodes: The number of inodes involved
503  *
504  * This calculate the number of blocks we need to reserve in a
505  * transaction to link @nr_inodes into a directory. In most cases
506  * @nr_inodes will be 2 (the directory plus the inode being linked in)
507  * but in case of rename, 4 may be required.
508  *
509  * Returns: Number of blocks
510  */
511 
512 static unsigned gfs2_trans_da_blks(const struct gfs2_inode *dip,
513 				   const struct gfs2_diradd *da,
514 				   unsigned nr_inodes)
515 {
516 	return da->nr_blocks + gfs2_rg_blocks(dip, da->nr_blocks) +
517 	       (nr_inodes * RES_DINODE) + RES_QUOTA + RES_STATFS;
518 }
519 
520 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
521 		       struct gfs2_inode *ip, struct gfs2_diradd *da)
522 {
523 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
524 	struct gfs2_alloc_parms ap = { .target = da->nr_blocks, };
525 	int error;
526 
527 	if (da->nr_blocks) {
528 		error = gfs2_quota_lock_check(dip, &ap);
529 		if (error)
530 			goto fail_quota_locks;
531 
532 		error = gfs2_inplace_reserve(dip, &ap);
533 		if (error)
534 			goto fail_quota_locks;
535 
536 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, da, 2), 0);
537 		if (error)
538 			goto fail_ipreserv;
539 	} else {
540 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
541 		if (error)
542 			goto fail_quota_locks;
543 	}
544 
545 	error = gfs2_dir_add(&dip->i_inode, name, ip, da);
546 
547 	gfs2_trans_end(sdp);
548 fail_ipreserv:
549 	gfs2_inplace_release(dip);
550 fail_quota_locks:
551 	gfs2_quota_unlock(dip);
552 	return error;
553 }
554 
555 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
556 		    void *fs_info)
557 {
558 	const struct xattr *xattr;
559 	int err = 0;
560 
561 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
562 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
563 				       xattr->value_len, 0,
564 				       GFS2_EATYPE_SECURITY);
565 		if (err < 0)
566 			break;
567 	}
568 	return err;
569 }
570 
571 /**
572  * gfs2_create_inode - Create a new inode
573  * @dir: The parent directory
574  * @dentry: The new dentry
575  * @file: If non-NULL, the file which is being opened
576  * @mode: The permissions on the new inode
577  * @dev: For device nodes, this is the device number
578  * @symname: For symlinks, this is the link destination
579  * @size: The initial size of the inode (ignored for directories)
580  *
581  * Returns: 0 on success, or error code
582  */
583 
584 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
585 			     struct file *file,
586 			     umode_t mode, dev_t dev, const char *symname,
587 			     unsigned int size, int excl, int *opened)
588 {
589 	const struct qstr *name = &dentry->d_name;
590 	struct posix_acl *default_acl, *acl;
591 	struct gfs2_holder ghs[2];
592 	struct inode *inode = NULL;
593 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
594 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
595 	struct gfs2_glock *io_gl;
596 	int error, free_vfs_inode = 0;
597 	u32 aflags = 0;
598 	unsigned blocks = 1;
599 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
600 
601 	if (!name->len || name->len > GFS2_FNAMESIZE)
602 		return -ENAMETOOLONG;
603 
604 	error = gfs2_rs_alloc(dip);
605 	if (error)
606 		return error;
607 
608 	error = gfs2_rindex_update(sdp);
609 	if (error)
610 		return error;
611 
612 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
613 	if (error)
614 		goto fail;
615 
616 	error = create_ok(dip, name, mode);
617 	if (error)
618 		goto fail_gunlock;
619 
620 	inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
621 	error = PTR_ERR(inode);
622 	if (!IS_ERR(inode)) {
623 		if (S_ISDIR(inode->i_mode)) {
624 			iput(inode);
625 			inode = ERR_PTR(-EISDIR);
626 			goto fail_gunlock;
627 		}
628 		d_instantiate(dentry, inode);
629 		error = 0;
630 		if (file) {
631 			if (S_ISREG(inode->i_mode))
632 				error = finish_open(file, dentry, gfs2_open_common, opened);
633 			else
634 				error = finish_no_open(file, NULL);
635 		}
636 		gfs2_glock_dq_uninit(ghs);
637 		return error;
638 	} else if (error != -ENOENT) {
639 		goto fail_gunlock;
640 	}
641 
642 	error = gfs2_diradd_alloc_required(dir, name, &da);
643 	if (error < 0)
644 		goto fail_gunlock;
645 
646 	inode = new_inode(sdp->sd_vfs);
647 	error = -ENOMEM;
648 	if (!inode)
649 		goto fail_gunlock;
650 
651 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
652 	if (error)
653 		goto fail_free_vfs_inode;
654 
655 	ip = GFS2_I(inode);
656 	error = gfs2_rs_alloc(ip);
657 	if (error)
658 		goto fail_free_acls;
659 
660 	inode->i_mode = mode;
661 	set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
662 	inode->i_rdev = dev;
663 	inode->i_size = size;
664 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
665 	gfs2_set_inode_blocks(inode, 1);
666 	munge_mode_uid_gid(dip, inode);
667 	check_and_update_goal(dip);
668 	ip->i_goal = dip->i_goal;
669 	ip->i_diskflags = 0;
670 	ip->i_eattr = 0;
671 	ip->i_height = 0;
672 	ip->i_depth = 0;
673 	ip->i_entries = 0;
674 
675 	switch(mode & S_IFMT) {
676 	case S_IFREG:
677 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
678 		    gfs2_tune_get(sdp, gt_new_files_jdata))
679 			ip->i_diskflags |= GFS2_DIF_JDATA;
680 		gfs2_set_aops(inode);
681 		break;
682 	case S_IFDIR:
683 		ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
684 		ip->i_diskflags |= GFS2_DIF_JDATA;
685 		ip->i_entries = 2;
686 		break;
687 	}
688 	gfs2_set_inode_flags(inode);
689 
690 	if ((GFS2_I(d_inode(sdp->sd_root_dir)) == dip) ||
691 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
692 		aflags |= GFS2_AF_ORLOV;
693 
694 	if (default_acl || acl)
695 		blocks++;
696 
697 	error = alloc_dinode(ip, aflags, &blocks);
698 	if (error)
699 		goto fail_free_inode;
700 
701 	gfs2_set_inode_blocks(inode, blocks);
702 
703 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
704 	if (error)
705 		goto fail_free_inode;
706 
707 	ip->i_gl->gl_object = ip;
708 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
709 	if (error)
710 		goto fail_free_inode;
711 
712 	error = gfs2_trans_begin(sdp, blocks, 0);
713 	if (error)
714 		goto fail_gunlock2;
715 
716 	if (blocks > 1) {
717 		ip->i_eattr = ip->i_no_addr + 1;
718 		gfs2_init_xattr(ip);
719 	}
720 	init_dinode(dip, ip, symname);
721 	gfs2_trans_end(sdp);
722 
723 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
724 	if (error)
725 		goto fail_gunlock2;
726 
727 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
728 	if (error)
729 		goto fail_gunlock2;
730 
731 	ip->i_iopen_gh.gh_gl->gl_object = ip;
732 	gfs2_glock_put(io_gl);
733 	gfs2_set_iop(inode);
734 	insert_inode_hash(inode);
735 
736 	if (default_acl) {
737 		error = gfs2_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
738 		posix_acl_release(default_acl);
739 	}
740 	if (acl) {
741 		if (!error)
742 			error = gfs2_set_acl(inode, acl, ACL_TYPE_ACCESS);
743 		posix_acl_release(acl);
744 	}
745 
746 	if (error)
747 		goto fail_gunlock3;
748 
749 	error = security_inode_init_security(&ip->i_inode, &dip->i_inode, name,
750 					     &gfs2_initxattrs, NULL);
751 	if (error)
752 		goto fail_gunlock3;
753 
754 	error = link_dinode(dip, name, ip, &da);
755 	if (error)
756 		goto fail_gunlock3;
757 
758 	mark_inode_dirty(inode);
759 	d_instantiate(dentry, inode);
760 	if (file) {
761 		*opened |= FILE_CREATED;
762 		error = finish_open(file, dentry, gfs2_open_common, opened);
763 	}
764 	gfs2_glock_dq_uninit(ghs);
765 	gfs2_glock_dq_uninit(ghs + 1);
766 	return error;
767 
768 fail_gunlock3:
769 	gfs2_glock_dq_uninit(ghs + 1);
770 	if (ip->i_gl)
771 		gfs2_glock_put(ip->i_gl);
772 	goto fail_gunlock;
773 
774 fail_gunlock2:
775 	gfs2_glock_dq_uninit(ghs + 1);
776 fail_free_inode:
777 	if (ip->i_gl)
778 		gfs2_glock_put(ip->i_gl);
779 	gfs2_rs_delete(ip, NULL);
780 fail_free_acls:
781 	if (default_acl)
782 		posix_acl_release(default_acl);
783 	if (acl)
784 		posix_acl_release(acl);
785 fail_free_vfs_inode:
786 	free_vfs_inode = 1;
787 fail_gunlock:
788 	gfs2_dir_no_add(&da);
789 	gfs2_glock_dq_uninit(ghs);
790 	if (inode && !IS_ERR(inode)) {
791 		clear_nlink(inode);
792 		if (!free_vfs_inode)
793 			mark_inode_dirty(inode);
794 		set_bit(free_vfs_inode ? GIF_FREE_VFS_INODE : GIF_ALLOC_FAILED,
795 			&GFS2_I(inode)->i_flags);
796 		iput(inode);
797 	}
798 fail:
799 	return error;
800 }
801 
802 /**
803  * gfs2_create - Create a file
804  * @dir: The directory in which to create the file
805  * @dentry: The dentry of the new file
806  * @mode: The mode of the new file
807  *
808  * Returns: errno
809  */
810 
811 static int gfs2_create(struct inode *dir, struct dentry *dentry,
812 		       umode_t mode, bool excl)
813 {
814 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl, NULL);
815 }
816 
817 /**
818  * __gfs2_lookup - Look up a filename in a directory and return its inode
819  * @dir: The directory inode
820  * @dentry: The dentry of the new inode
821  * @file: File to be opened
822  * @opened: atomic_open flags
823  *
824  *
825  * Returns: errno
826  */
827 
828 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
829 				    struct file *file, int *opened)
830 {
831 	struct inode *inode;
832 	struct dentry *d;
833 	struct gfs2_holder gh;
834 	struct gfs2_glock *gl;
835 	int error;
836 
837 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
838 	if (inode == NULL) {
839 		d_add(dentry, NULL);
840 		return NULL;
841 	}
842 	if (IS_ERR(inode))
843 		return ERR_CAST(inode);
844 
845 	gl = GFS2_I(inode)->i_gl;
846 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
847 	if (error) {
848 		iput(inode);
849 		return ERR_PTR(error);
850 	}
851 
852 	d = d_splice_alias(inode, dentry);
853 	if (IS_ERR(d)) {
854 		gfs2_glock_dq_uninit(&gh);
855 		return d;
856 	}
857 	if (file && S_ISREG(inode->i_mode))
858 		error = finish_open(file, dentry, gfs2_open_common, opened);
859 
860 	gfs2_glock_dq_uninit(&gh);
861 	if (error) {
862 		dput(d);
863 		return ERR_PTR(error);
864 	}
865 	return d;
866 }
867 
868 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
869 				  unsigned flags)
870 {
871 	return __gfs2_lookup(dir, dentry, NULL, NULL);
872 }
873 
874 /**
875  * gfs2_link - Link to a file
876  * @old_dentry: The inode to link
877  * @dir: Add link to this directory
878  * @dentry: The name of the link
879  *
880  * Link the inode in "old_dentry" into the directory "dir" with the
881  * name in "dentry".
882  *
883  * Returns: errno
884  */
885 
886 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
887 		     struct dentry *dentry)
888 {
889 	struct gfs2_inode *dip = GFS2_I(dir);
890 	struct gfs2_sbd *sdp = GFS2_SB(dir);
891 	struct inode *inode = d_inode(old_dentry);
892 	struct gfs2_inode *ip = GFS2_I(inode);
893 	struct gfs2_holder ghs[2];
894 	struct buffer_head *dibh;
895 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
896 	int error;
897 
898 	if (S_ISDIR(inode->i_mode))
899 		return -EPERM;
900 
901 	error = gfs2_rs_alloc(dip);
902 	if (error)
903 		return error;
904 
905 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
906 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
907 
908 	error = gfs2_glock_nq(ghs); /* parent */
909 	if (error)
910 		goto out_parent;
911 
912 	error = gfs2_glock_nq(ghs + 1); /* child */
913 	if (error)
914 		goto out_child;
915 
916 	error = -ENOENT;
917 	if (inode->i_nlink == 0)
918 		goto out_gunlock;
919 
920 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
921 	if (error)
922 		goto out_gunlock;
923 
924 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
925 	switch (error) {
926 	case -ENOENT:
927 		break;
928 	case 0:
929 		error = -EEXIST;
930 	default:
931 		goto out_gunlock;
932 	}
933 
934 	error = -EINVAL;
935 	if (!dip->i_inode.i_nlink)
936 		goto out_gunlock;
937 	error = -EFBIG;
938 	if (dip->i_entries == (u32)-1)
939 		goto out_gunlock;
940 	error = -EPERM;
941 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
942 		goto out_gunlock;
943 	error = -EINVAL;
944 	if (!ip->i_inode.i_nlink)
945 		goto out_gunlock;
946 	error = -EMLINK;
947 	if (ip->i_inode.i_nlink == (u32)-1)
948 		goto out_gunlock;
949 
950 	error = gfs2_diradd_alloc_required(dir, &dentry->d_name, &da);
951 	if (error < 0)
952 		goto out_gunlock;
953 
954 	if (da.nr_blocks) {
955 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
956 		error = gfs2_quota_lock_check(dip, &ap);
957 		if (error)
958 			goto out_gunlock;
959 
960 		error = gfs2_inplace_reserve(dip, &ap);
961 		if (error)
962 			goto out_gunlock_q;
963 
964 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, &da, 2), 0);
965 		if (error)
966 			goto out_ipres;
967 	} else {
968 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
969 		if (error)
970 			goto out_ipres;
971 	}
972 
973 	error = gfs2_meta_inode_buffer(ip, &dibh);
974 	if (error)
975 		goto out_end_trans;
976 
977 	error = gfs2_dir_add(dir, &dentry->d_name, ip, &da);
978 	if (error)
979 		goto out_brelse;
980 
981 	gfs2_trans_add_meta(ip->i_gl, dibh);
982 	inc_nlink(&ip->i_inode);
983 	ip->i_inode.i_ctime = CURRENT_TIME;
984 	ihold(inode);
985 	d_instantiate(dentry, inode);
986 	mark_inode_dirty(inode);
987 
988 out_brelse:
989 	brelse(dibh);
990 out_end_trans:
991 	gfs2_trans_end(sdp);
992 out_ipres:
993 	if (da.nr_blocks)
994 		gfs2_inplace_release(dip);
995 out_gunlock_q:
996 	if (da.nr_blocks)
997 		gfs2_quota_unlock(dip);
998 out_gunlock:
999 	gfs2_dir_no_add(&da);
1000 	gfs2_glock_dq(ghs + 1);
1001 out_child:
1002 	gfs2_glock_dq(ghs);
1003 out_parent:
1004 	gfs2_holder_uninit(ghs);
1005 	gfs2_holder_uninit(ghs + 1);
1006 	return error;
1007 }
1008 
1009 /*
1010  * gfs2_unlink_ok - check to see that a inode is still in a directory
1011  * @dip: the directory
1012  * @name: the name of the file
1013  * @ip: the inode
1014  *
1015  * Assumes that the lock on (at least) @dip is held.
1016  *
1017  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1018  */
1019 
1020 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
1021 			  const struct gfs2_inode *ip)
1022 {
1023 	int error;
1024 
1025 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
1026 		return -EPERM;
1027 
1028 	if ((dip->i_inode.i_mode & S_ISVTX) &&
1029 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
1030 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
1031 		return -EPERM;
1032 
1033 	if (IS_APPEND(&dip->i_inode))
1034 		return -EPERM;
1035 
1036 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
1037 	if (error)
1038 		return error;
1039 
1040 	return gfs2_dir_check(&dip->i_inode, name, ip);
1041 }
1042 
1043 /**
1044  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1045  * @dip: The parent directory
1046  * @name: The name of the entry in the parent directory
1047  * @inode: The inode to be removed
1048  *
1049  * Called with all the locks and in a transaction. This will only be
1050  * called for a directory after it has been checked to ensure it is empty.
1051  *
1052  * Returns: 0 on success, or an error
1053  */
1054 
1055 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1056 			     const struct dentry *dentry)
1057 {
1058 	struct inode *inode = d_inode(dentry);
1059 	struct gfs2_inode *ip = GFS2_I(inode);
1060 	int error;
1061 
1062 	error = gfs2_dir_del(dip, dentry);
1063 	if (error)
1064 		return error;
1065 
1066 	ip->i_entries = 0;
1067 	inode->i_ctime = CURRENT_TIME;
1068 	if (S_ISDIR(inode->i_mode))
1069 		clear_nlink(inode);
1070 	else
1071 		drop_nlink(inode);
1072 	mark_inode_dirty(inode);
1073 	if (inode->i_nlink == 0)
1074 		gfs2_unlink_di(inode);
1075 	return 0;
1076 }
1077 
1078 
1079 /**
1080  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1081  * @dir: The inode of the directory containing the inode to unlink
1082  * @dentry: The file itself
1083  *
1084  * This routine uses the type of the inode as a flag to figure out
1085  * whether this is an unlink or an rmdir.
1086  *
1087  * Returns: errno
1088  */
1089 
1090 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1091 {
1092 	struct gfs2_inode *dip = GFS2_I(dir);
1093 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1094 	struct inode *inode = d_inode(dentry);
1095 	struct gfs2_inode *ip = GFS2_I(inode);
1096 	struct gfs2_holder ghs[3];
1097 	struct gfs2_rgrpd *rgd;
1098 	int error;
1099 
1100 	error = gfs2_rindex_update(sdp);
1101 	if (error)
1102 		return error;
1103 
1104 	error = -EROFS;
1105 
1106 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1107 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1108 
1109 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1110 	if (!rgd)
1111 		goto out_inodes;
1112 
1113 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1114 
1115 
1116 	error = gfs2_glock_nq(ghs); /* parent */
1117 	if (error)
1118 		goto out_parent;
1119 
1120 	error = gfs2_glock_nq(ghs + 1); /* child */
1121 	if (error)
1122 		goto out_child;
1123 
1124 	error = -ENOENT;
1125 	if (inode->i_nlink == 0)
1126 		goto out_rgrp;
1127 
1128 	if (S_ISDIR(inode->i_mode)) {
1129 		error = -ENOTEMPTY;
1130 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1131 			goto out_rgrp;
1132 	}
1133 
1134 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1135 	if (error)
1136 		goto out_rgrp;
1137 
1138 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1139 	if (error)
1140 		goto out_gunlock;
1141 
1142 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1143 	if (error)
1144 		goto out_end_trans;
1145 
1146 	error = gfs2_unlink_inode(dip, dentry);
1147 
1148 out_end_trans:
1149 	gfs2_trans_end(sdp);
1150 out_gunlock:
1151 	gfs2_glock_dq(ghs + 2);
1152 out_rgrp:
1153 	gfs2_glock_dq(ghs + 1);
1154 out_child:
1155 	gfs2_glock_dq(ghs);
1156 out_parent:
1157 	gfs2_holder_uninit(ghs + 2);
1158 out_inodes:
1159 	gfs2_holder_uninit(ghs + 1);
1160 	gfs2_holder_uninit(ghs);
1161 	return error;
1162 }
1163 
1164 /**
1165  * gfs2_symlink - Create a symlink
1166  * @dir: The directory to create the symlink in
1167  * @dentry: The dentry to put the symlink in
1168  * @symname: The thing which the link points to
1169  *
1170  * Returns: errno
1171  */
1172 
1173 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1174 			const char *symname)
1175 {
1176 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1177 	unsigned int size;
1178 
1179 	size = strlen(symname);
1180 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1181 		return -ENAMETOOLONG;
1182 
1183 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0, NULL);
1184 }
1185 
1186 /**
1187  * gfs2_mkdir - Make a directory
1188  * @dir: The parent directory of the new one
1189  * @dentry: The dentry of the new directory
1190  * @mode: The mode of the new directory
1191  *
1192  * Returns: errno
1193  */
1194 
1195 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1196 {
1197 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1198 	unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1199 	return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0, NULL);
1200 }
1201 
1202 /**
1203  * gfs2_mknod - Make a special file
1204  * @dir: The directory in which the special file will reside
1205  * @dentry: The dentry of the special file
1206  * @mode: The mode of the special file
1207  * @dev: The device specification of the special file
1208  *
1209  */
1210 
1211 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1212 		      dev_t dev)
1213 {
1214 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0, NULL);
1215 }
1216 
1217 /**
1218  * gfs2_atomic_open - Atomically open a file
1219  * @dir: The directory
1220  * @dentry: The proposed new entry
1221  * @file: The proposed new struct file
1222  * @flags: open flags
1223  * @mode: File mode
1224  * @opened: Flag to say whether the file has been opened or not
1225  *
1226  * Returns: error code or 0 for success
1227  */
1228 
1229 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1230                             struct file *file, unsigned flags,
1231                             umode_t mode, int *opened)
1232 {
1233 	struct dentry *d;
1234 	bool excl = !!(flags & O_EXCL);
1235 
1236 	if (!d_unhashed(dentry))
1237 		goto skip_lookup;
1238 
1239 	d = __gfs2_lookup(dir, dentry, file, opened);
1240 	if (IS_ERR(d))
1241 		return PTR_ERR(d);
1242 	if (d != NULL)
1243 		dentry = d;
1244 	if (d_really_is_positive(dentry)) {
1245 		if (!(*opened & FILE_OPENED))
1246 			return finish_no_open(file, d);
1247 		dput(d);
1248 		return 0;
1249 	}
1250 
1251 	BUG_ON(d != NULL);
1252 
1253 skip_lookup:
1254 	if (!(flags & O_CREAT))
1255 		return -ENOENT;
1256 
1257 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl, opened);
1258 }
1259 
1260 /*
1261  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1262  * @this: move this
1263  * @to: to here
1264  *
1265  * Follow @to back to the root and make sure we don't encounter @this
1266  * Assumes we already hold the rename lock.
1267  *
1268  * Returns: errno
1269  */
1270 
1271 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1272 {
1273 	struct inode *dir = &to->i_inode;
1274 	struct super_block *sb = dir->i_sb;
1275 	struct inode *tmp;
1276 	int error = 0;
1277 
1278 	igrab(dir);
1279 
1280 	for (;;) {
1281 		if (dir == &this->i_inode) {
1282 			error = -EINVAL;
1283 			break;
1284 		}
1285 		if (dir == d_inode(sb->s_root)) {
1286 			error = 0;
1287 			break;
1288 		}
1289 
1290 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1291 		if (!tmp) {
1292 			error = -ENOENT;
1293 			break;
1294 		}
1295 		if (IS_ERR(tmp)) {
1296 			error = PTR_ERR(tmp);
1297 			break;
1298 		}
1299 
1300 		iput(dir);
1301 		dir = tmp;
1302 	}
1303 
1304 	iput(dir);
1305 
1306 	return error;
1307 }
1308 
1309 /**
1310  * gfs2_rename - Rename a file
1311  * @odir: Parent directory of old file name
1312  * @odentry: The old dentry of the file
1313  * @ndir: Parent directory of new file name
1314  * @ndentry: The new dentry of the file
1315  *
1316  * Returns: errno
1317  */
1318 
1319 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1320 		       struct inode *ndir, struct dentry *ndentry)
1321 {
1322 	struct gfs2_inode *odip = GFS2_I(odir);
1323 	struct gfs2_inode *ndip = GFS2_I(ndir);
1324 	struct gfs2_inode *ip = GFS2_I(d_inode(odentry));
1325 	struct gfs2_inode *nip = NULL;
1326 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1327 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1328 	struct gfs2_rgrpd *nrgd;
1329 	unsigned int num_gh;
1330 	int dir_rename = 0;
1331 	struct gfs2_diradd da = { .nr_blocks = 0, .save_loc = 0, };
1332 	unsigned int x;
1333 	int error;
1334 
1335 	if (d_really_is_positive(ndentry)) {
1336 		nip = GFS2_I(d_inode(ndentry));
1337 		if (ip == nip)
1338 			return 0;
1339 	}
1340 
1341 	error = gfs2_rindex_update(sdp);
1342 	if (error)
1343 		return error;
1344 
1345 	error = gfs2_rs_alloc(ndip);
1346 	if (error)
1347 		return error;
1348 
1349 	if (odip != ndip) {
1350 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1351 					   0, &r_gh);
1352 		if (error)
1353 			goto out;
1354 
1355 		if (S_ISDIR(ip->i_inode.i_mode)) {
1356 			dir_rename = 1;
1357 			/* don't move a dirctory into it's subdir */
1358 			error = gfs2_ok_to_move(ip, ndip);
1359 			if (error)
1360 				goto out_gunlock_r;
1361 		}
1362 	}
1363 
1364 	num_gh = 1;
1365 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1366 	if (odip != ndip) {
1367 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1368 		num_gh++;
1369 	}
1370 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1371 	num_gh++;
1372 
1373 	if (nip) {
1374 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1375 		num_gh++;
1376 		/* grab the resource lock for unlink flag twiddling
1377 		 * this is the case of the target file already existing
1378 		 * so we unlink before doing the rename
1379 		 */
1380 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1381 		if (nrgd)
1382 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1383 	}
1384 
1385 	for (x = 0; x < num_gh; x++) {
1386 		error = gfs2_glock_nq(ghs + x);
1387 		if (error)
1388 			goto out_gunlock;
1389 	}
1390 
1391 	error = -ENOENT;
1392 	if (ip->i_inode.i_nlink == 0)
1393 		goto out_gunlock;
1394 
1395 	/* Check out the old directory */
1396 
1397 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1398 	if (error)
1399 		goto out_gunlock;
1400 
1401 	/* Check out the new directory */
1402 
1403 	if (nip) {
1404 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1405 		if (error)
1406 			goto out_gunlock;
1407 
1408 		if (nip->i_inode.i_nlink == 0) {
1409 			error = -EAGAIN;
1410 			goto out_gunlock;
1411 		}
1412 
1413 		if (S_ISDIR(nip->i_inode.i_mode)) {
1414 			if (nip->i_entries < 2) {
1415 				gfs2_consist_inode(nip);
1416 				error = -EIO;
1417 				goto out_gunlock;
1418 			}
1419 			if (nip->i_entries > 2) {
1420 				error = -ENOTEMPTY;
1421 				goto out_gunlock;
1422 			}
1423 		}
1424 	} else {
1425 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1426 		if (error)
1427 			goto out_gunlock;
1428 
1429 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1430 		switch (error) {
1431 		case -ENOENT:
1432 			error = 0;
1433 			break;
1434 		case 0:
1435 			error = -EEXIST;
1436 		default:
1437 			goto out_gunlock;
1438 		};
1439 
1440 		if (odip != ndip) {
1441 			if (!ndip->i_inode.i_nlink) {
1442 				error = -ENOENT;
1443 				goto out_gunlock;
1444 			}
1445 			if (ndip->i_entries == (u32)-1) {
1446 				error = -EFBIG;
1447 				goto out_gunlock;
1448 			}
1449 			if (S_ISDIR(ip->i_inode.i_mode) &&
1450 			    ndip->i_inode.i_nlink == (u32)-1) {
1451 				error = -EMLINK;
1452 				goto out_gunlock;
1453 			}
1454 		}
1455 	}
1456 
1457 	/* Check out the dir to be renamed */
1458 
1459 	if (dir_rename) {
1460 		error = gfs2_permission(d_inode(odentry), MAY_WRITE);
1461 		if (error)
1462 			goto out_gunlock;
1463 	}
1464 
1465 	if (nip == NULL) {
1466 		error = gfs2_diradd_alloc_required(ndir, &ndentry->d_name, &da);
1467 		if (error)
1468 			goto out_gunlock;
1469 	}
1470 
1471 	if (da.nr_blocks) {
1472 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1473 		error = gfs2_quota_lock_check(ndip, &ap);
1474 		if (error)
1475 			goto out_gunlock;
1476 
1477 		error = gfs2_inplace_reserve(ndip, &ap);
1478 		if (error)
1479 			goto out_gunlock_q;
1480 
1481 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(ndip, &da, 4) +
1482 					 4 * RES_LEAF + 4, 0);
1483 		if (error)
1484 			goto out_ipreserv;
1485 	} else {
1486 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1487 					 5 * RES_LEAF + 4, 0);
1488 		if (error)
1489 			goto out_gunlock;
1490 	}
1491 
1492 	/* Remove the target file, if it exists */
1493 
1494 	if (nip)
1495 		error = gfs2_unlink_inode(ndip, ndentry);
1496 
1497 	if (dir_rename) {
1498 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1499 		if (error)
1500 			goto out_end_trans;
1501 	} else {
1502 		struct buffer_head *dibh;
1503 		error = gfs2_meta_inode_buffer(ip, &dibh);
1504 		if (error)
1505 			goto out_end_trans;
1506 		ip->i_inode.i_ctime = CURRENT_TIME;
1507 		gfs2_trans_add_meta(ip->i_gl, dibh);
1508 		gfs2_dinode_out(ip, dibh->b_data);
1509 		brelse(dibh);
1510 	}
1511 
1512 	error = gfs2_dir_del(odip, odentry);
1513 	if (error)
1514 		goto out_end_trans;
1515 
1516 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip, &da);
1517 	if (error)
1518 		goto out_end_trans;
1519 
1520 out_end_trans:
1521 	gfs2_trans_end(sdp);
1522 out_ipreserv:
1523 	if (da.nr_blocks)
1524 		gfs2_inplace_release(ndip);
1525 out_gunlock_q:
1526 	if (da.nr_blocks)
1527 		gfs2_quota_unlock(ndip);
1528 out_gunlock:
1529 	gfs2_dir_no_add(&da);
1530 	while (x--) {
1531 		gfs2_glock_dq(ghs + x);
1532 		gfs2_holder_uninit(ghs + x);
1533 	}
1534 out_gunlock_r:
1535 	if (r_gh.gh_gl)
1536 		gfs2_glock_dq_uninit(&r_gh);
1537 out:
1538 	return error;
1539 }
1540 
1541 /**
1542  * gfs2_follow_link - Follow a symbolic link
1543  * @dentry: The dentry of the link
1544  * @nd: Data that we pass to vfs_follow_link()
1545  *
1546  * This can handle symlinks of any size.
1547  *
1548  * Returns: 0 on success or error code
1549  */
1550 
1551 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1552 {
1553 	struct gfs2_inode *ip = GFS2_I(d_inode(dentry));
1554 	struct gfs2_holder i_gh;
1555 	struct buffer_head *dibh;
1556 	unsigned int size;
1557 	char *buf;
1558 	int error;
1559 
1560 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1561 	error = gfs2_glock_nq(&i_gh);
1562 	if (error) {
1563 		gfs2_holder_uninit(&i_gh);
1564 		nd_set_link(nd, ERR_PTR(error));
1565 		return NULL;
1566 	}
1567 
1568 	size = (unsigned int)i_size_read(&ip->i_inode);
1569 	if (size == 0) {
1570 		gfs2_consist_inode(ip);
1571 		buf = ERR_PTR(-EIO);
1572 		goto out;
1573 	}
1574 
1575 	error = gfs2_meta_inode_buffer(ip, &dibh);
1576 	if (error) {
1577 		buf = ERR_PTR(error);
1578 		goto out;
1579 	}
1580 
1581 	buf = kzalloc(size + 1, GFP_NOFS);
1582 	if (!buf)
1583 		buf = ERR_PTR(-ENOMEM);
1584 	else
1585 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1586 	brelse(dibh);
1587 out:
1588 	gfs2_glock_dq_uninit(&i_gh);
1589 	nd_set_link(nd, buf);
1590 	return NULL;
1591 }
1592 
1593 /**
1594  * gfs2_permission -
1595  * @inode: The inode
1596  * @mask: The mask to be tested
1597  * @flags: Indicates whether this is an RCU path walk or not
1598  *
1599  * This may be called from the VFS directly, or from within GFS2 with the
1600  * inode locked, so we look to see if the glock is already locked and only
1601  * lock the glock if its not already been done.
1602  *
1603  * Returns: errno
1604  */
1605 
1606 int gfs2_permission(struct inode *inode, int mask)
1607 {
1608 	struct gfs2_inode *ip;
1609 	struct gfs2_holder i_gh;
1610 	int error;
1611 	int unlock = 0;
1612 
1613 
1614 	ip = GFS2_I(inode);
1615 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1616 		if (mask & MAY_NOT_BLOCK)
1617 			return -ECHILD;
1618 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1619 		if (error)
1620 			return error;
1621 		unlock = 1;
1622 	}
1623 
1624 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1625 		error = -EACCES;
1626 	else
1627 		error = generic_permission(inode, mask);
1628 	if (unlock)
1629 		gfs2_glock_dq_uninit(&i_gh);
1630 
1631 	return error;
1632 }
1633 
1634 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1635 {
1636 	setattr_copy(inode, attr);
1637 	mark_inode_dirty(inode);
1638 	return 0;
1639 }
1640 
1641 /**
1642  * gfs2_setattr_simple -
1643  * @ip:
1644  * @attr:
1645  *
1646  * Returns: errno
1647  */
1648 
1649 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1650 {
1651 	int error;
1652 
1653 	if (current->journal_info)
1654 		return __gfs2_setattr_simple(inode, attr);
1655 
1656 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1657 	if (error)
1658 		return error;
1659 
1660 	error = __gfs2_setattr_simple(inode, attr);
1661 	gfs2_trans_end(GFS2_SB(inode));
1662 	return error;
1663 }
1664 
1665 static int setattr_chown(struct inode *inode, struct iattr *attr)
1666 {
1667 	struct gfs2_inode *ip = GFS2_I(inode);
1668 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1669 	kuid_t ouid, nuid;
1670 	kgid_t ogid, ngid;
1671 	int error;
1672 	struct gfs2_alloc_parms ap;
1673 
1674 	ouid = inode->i_uid;
1675 	ogid = inode->i_gid;
1676 	nuid = attr->ia_uid;
1677 	ngid = attr->ia_gid;
1678 
1679 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1680 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1681 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1682 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1683 
1684 	error = get_write_access(inode);
1685 	if (error)
1686 		return error;
1687 
1688 	error = gfs2_rs_alloc(ip);
1689 	if (error)
1690 		goto out;
1691 
1692 	error = gfs2_rindex_update(sdp);
1693 	if (error)
1694 		goto out;
1695 
1696 	error = gfs2_quota_lock(ip, nuid, ngid);
1697 	if (error)
1698 		goto out;
1699 
1700 	ap.target = gfs2_get_inode_blocks(&ip->i_inode);
1701 
1702 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1703 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1704 		error = gfs2_quota_check(ip, nuid, ngid, &ap);
1705 		if (error)
1706 			goto out_gunlock_q;
1707 	}
1708 
1709 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1710 	if (error)
1711 		goto out_gunlock_q;
1712 
1713 	error = gfs2_setattr_simple(inode, attr);
1714 	if (error)
1715 		goto out_end_trans;
1716 
1717 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1718 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1719 		gfs2_quota_change(ip, -ap.target, ouid, ogid);
1720 		gfs2_quota_change(ip, ap.target, nuid, ngid);
1721 	}
1722 
1723 out_end_trans:
1724 	gfs2_trans_end(sdp);
1725 out_gunlock_q:
1726 	gfs2_quota_unlock(ip);
1727 out:
1728 	put_write_access(inode);
1729 	return error;
1730 }
1731 
1732 /**
1733  * gfs2_setattr - Change attributes on an inode
1734  * @dentry: The dentry which is changing
1735  * @attr: The structure describing the change
1736  *
1737  * The VFS layer wants to change one or more of an inodes attributes.  Write
1738  * that change out to disk.
1739  *
1740  * Returns: errno
1741  */
1742 
1743 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1744 {
1745 	struct inode *inode = d_inode(dentry);
1746 	struct gfs2_inode *ip = GFS2_I(inode);
1747 	struct gfs2_holder i_gh;
1748 	int error;
1749 
1750 	error = gfs2_rs_alloc(ip);
1751 	if (error)
1752 		return error;
1753 
1754 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1755 	if (error)
1756 		return error;
1757 
1758 	error = -EPERM;
1759 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1760 		goto out;
1761 
1762 	error = inode_change_ok(inode, attr);
1763 	if (error)
1764 		goto out;
1765 
1766 	if (attr->ia_valid & ATTR_SIZE)
1767 		error = gfs2_setattr_size(inode, attr->ia_size);
1768 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1769 		error = setattr_chown(inode, attr);
1770 	else {
1771 		error = gfs2_setattr_simple(inode, attr);
1772 		if (!error && attr->ia_valid & ATTR_MODE)
1773 			error = posix_acl_chmod(inode, inode->i_mode);
1774 	}
1775 
1776 out:
1777 	if (!error)
1778 		mark_inode_dirty(inode);
1779 	gfs2_glock_dq_uninit(&i_gh);
1780 	return error;
1781 }
1782 
1783 /**
1784  * gfs2_getattr - Read out an inode's attributes
1785  * @mnt: The vfsmount the inode is being accessed from
1786  * @dentry: The dentry to stat
1787  * @stat: The inode's stats
1788  *
1789  * This may be called from the VFS directly, or from within GFS2 with the
1790  * inode locked, so we look to see if the glock is already locked and only
1791  * lock the glock if its not already been done. Note that its the NFS
1792  * readdirplus operation which causes this to be called (from filldir)
1793  * with the glock already held.
1794  *
1795  * Returns: errno
1796  */
1797 
1798 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1799 			struct kstat *stat)
1800 {
1801 	struct inode *inode = d_inode(dentry);
1802 	struct gfs2_inode *ip = GFS2_I(inode);
1803 	struct gfs2_holder gh;
1804 	int error;
1805 	int unlock = 0;
1806 
1807 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1808 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1809 		if (error)
1810 			return error;
1811 		unlock = 1;
1812 	}
1813 
1814 	generic_fillattr(inode, stat);
1815 	if (unlock)
1816 		gfs2_glock_dq_uninit(&gh);
1817 
1818 	return 0;
1819 }
1820 
1821 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1822 			 const void *data, size_t size, int flags)
1823 {
1824 	struct inode *inode = d_inode(dentry);
1825 	struct gfs2_inode *ip = GFS2_I(inode);
1826 	struct gfs2_holder gh;
1827 	int ret;
1828 
1829 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1830 	ret = gfs2_glock_nq(&gh);
1831 	if (ret == 0) {
1832 		ret = gfs2_rs_alloc(ip);
1833 		if (ret == 0)
1834 			ret = generic_setxattr(dentry, name, data, size, flags);
1835 		gfs2_glock_dq(&gh);
1836 	}
1837 	gfs2_holder_uninit(&gh);
1838 	return ret;
1839 }
1840 
1841 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1842 			     void *data, size_t size)
1843 {
1844 	struct inode *inode = d_inode(dentry);
1845 	struct gfs2_inode *ip = GFS2_I(inode);
1846 	struct gfs2_holder gh;
1847 	int ret;
1848 
1849 	/* For selinux during lookup */
1850 	if (gfs2_glock_is_locked_by_me(ip->i_gl))
1851 		return generic_getxattr(dentry, name, data, size);
1852 
1853 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1854 	ret = gfs2_glock_nq(&gh);
1855 	if (ret == 0) {
1856 		ret = generic_getxattr(dentry, name, data, size);
1857 		gfs2_glock_dq(&gh);
1858 	}
1859 	gfs2_holder_uninit(&gh);
1860 	return ret;
1861 }
1862 
1863 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1864 {
1865 	struct inode *inode = d_inode(dentry);
1866 	struct gfs2_inode *ip = GFS2_I(inode);
1867 	struct gfs2_holder gh;
1868 	int ret;
1869 
1870 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1871 	ret = gfs2_glock_nq(&gh);
1872 	if (ret == 0) {
1873 		ret = gfs2_rs_alloc(ip);
1874 		if (ret == 0)
1875 			ret = generic_removexattr(dentry, name);
1876 		gfs2_glock_dq(&gh);
1877 	}
1878 	gfs2_holder_uninit(&gh);
1879 	return ret;
1880 }
1881 
1882 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1883 		       u64 start, u64 len)
1884 {
1885 	struct gfs2_inode *ip = GFS2_I(inode);
1886 	struct gfs2_holder gh;
1887 	int ret;
1888 
1889 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1890 	if (ret)
1891 		return ret;
1892 
1893 	mutex_lock(&inode->i_mutex);
1894 
1895 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1896 	if (ret)
1897 		goto out;
1898 
1899 	if (gfs2_is_stuffed(ip)) {
1900 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1901 		u64 size = i_size_read(inode);
1902 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1903 			    FIEMAP_EXTENT_DATA_INLINE;
1904 		phys += sizeof(struct gfs2_dinode);
1905 		phys += start;
1906 		if (start + len > size)
1907 			len = size - start;
1908 		if (start < size)
1909 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1910 						      len, flags);
1911 		if (ret == 1)
1912 			ret = 0;
1913 	} else {
1914 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1915 					     gfs2_block_map);
1916 	}
1917 
1918 	gfs2_glock_dq_uninit(&gh);
1919 out:
1920 	mutex_unlock(&inode->i_mutex);
1921 	return ret;
1922 }
1923 
1924 const struct inode_operations gfs2_file_iops = {
1925 	.permission = gfs2_permission,
1926 	.setattr = gfs2_setattr,
1927 	.getattr = gfs2_getattr,
1928 	.setxattr = gfs2_setxattr,
1929 	.getxattr = gfs2_getxattr,
1930 	.listxattr = gfs2_listxattr,
1931 	.removexattr = gfs2_removexattr,
1932 	.fiemap = gfs2_fiemap,
1933 	.get_acl = gfs2_get_acl,
1934 	.set_acl = gfs2_set_acl,
1935 };
1936 
1937 const struct inode_operations gfs2_dir_iops = {
1938 	.create = gfs2_create,
1939 	.lookup = gfs2_lookup,
1940 	.link = gfs2_link,
1941 	.unlink = gfs2_unlink,
1942 	.symlink = gfs2_symlink,
1943 	.mkdir = gfs2_mkdir,
1944 	.rmdir = gfs2_unlink,
1945 	.mknod = gfs2_mknod,
1946 	.rename = gfs2_rename,
1947 	.permission = gfs2_permission,
1948 	.setattr = gfs2_setattr,
1949 	.getattr = gfs2_getattr,
1950 	.setxattr = gfs2_setxattr,
1951 	.getxattr = gfs2_getxattr,
1952 	.listxattr = gfs2_listxattr,
1953 	.removexattr = gfs2_removexattr,
1954 	.fiemap = gfs2_fiemap,
1955 	.get_acl = gfs2_get_acl,
1956 	.set_acl = gfs2_set_acl,
1957 	.atomic_open = gfs2_atomic_open,
1958 };
1959 
1960 const struct inode_operations gfs2_symlink_iops = {
1961 	.readlink = generic_readlink,
1962 	.follow_link = gfs2_follow_link,
1963 	.put_link = kfree_put_link,
1964 	.permission = gfs2_permission,
1965 	.setattr = gfs2_setattr,
1966 	.getattr = gfs2_getattr,
1967 	.setxattr = gfs2_setxattr,
1968 	.getxattr = gfs2_getxattr,
1969 	.listxattr = gfs2_listxattr,
1970 	.removexattr = gfs2_removexattr,
1971 	.fiemap = gfs2_fiemap,
1972 };
1973 
1974