xref: /openbmc/linux/fs/gfs2/inode.c (revision 97da55fc)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2011 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/namei.h>
15 #include <linux/mm.h>
16 #include <linux/xattr.h>
17 #include <linux/posix_acl.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/crc32.h>
20 #include <linux/fiemap.h>
21 #include <linux/security.h>
22 #include <asm/uaccess.h>
23 
24 #include "gfs2.h"
25 #include "incore.h"
26 #include "acl.h"
27 #include "bmap.h"
28 #include "dir.h"
29 #include "xattr.h"
30 #include "glock.h"
31 #include "inode.h"
32 #include "meta_io.h"
33 #include "quota.h"
34 #include "rgrp.h"
35 #include "trans.h"
36 #include "util.h"
37 #include "super.h"
38 #include "glops.h"
39 
40 struct gfs2_skip_data {
41 	u64 no_addr;
42 	int skipped;
43 	int non_block;
44 };
45 
46 static int iget_test(struct inode *inode, void *opaque)
47 {
48 	struct gfs2_inode *ip = GFS2_I(inode);
49 	struct gfs2_skip_data *data = opaque;
50 
51 	if (ip->i_no_addr == data->no_addr) {
52 		if (data->non_block &&
53 		    inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
54 			data->skipped = 1;
55 			return 0;
56 		}
57 		return 1;
58 	}
59 	return 0;
60 }
61 
62 static int iget_set(struct inode *inode, void *opaque)
63 {
64 	struct gfs2_inode *ip = GFS2_I(inode);
65 	struct gfs2_skip_data *data = opaque;
66 
67 	if (data->skipped)
68 		return -ENOENT;
69 	inode->i_ino = (unsigned long)(data->no_addr);
70 	ip->i_no_addr = data->no_addr;
71 	return 0;
72 }
73 
74 struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr, int non_block)
75 {
76 	unsigned long hash = (unsigned long)no_addr;
77 	struct gfs2_skip_data data;
78 
79 	data.no_addr = no_addr;
80 	data.skipped = 0;
81 	data.non_block = non_block;
82 	return ilookup5(sb, hash, iget_test, &data);
83 }
84 
85 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr,
86 			       int non_block)
87 {
88 	struct gfs2_skip_data data;
89 	unsigned long hash = (unsigned long)no_addr;
90 
91 	data.no_addr = no_addr;
92 	data.skipped = 0;
93 	data.non_block = non_block;
94 	return iget5_locked(sb, hash, iget_test, iget_set, &data);
95 }
96 
97 /**
98  * gfs2_set_iop - Sets inode operations
99  * @inode: The inode with correct i_mode filled in
100  *
101  * GFS2 lookup code fills in vfs inode contents based on info obtained
102  * from directory entry inside gfs2_inode_lookup().
103  */
104 
105 static void gfs2_set_iop(struct inode *inode)
106 {
107 	struct gfs2_sbd *sdp = GFS2_SB(inode);
108 	umode_t mode = inode->i_mode;
109 
110 	if (S_ISREG(mode)) {
111 		inode->i_op = &gfs2_file_iops;
112 		if (gfs2_localflocks(sdp))
113 			inode->i_fop = &gfs2_file_fops_nolock;
114 		else
115 			inode->i_fop = &gfs2_file_fops;
116 	} else if (S_ISDIR(mode)) {
117 		inode->i_op = &gfs2_dir_iops;
118 		if (gfs2_localflocks(sdp))
119 			inode->i_fop = &gfs2_dir_fops_nolock;
120 		else
121 			inode->i_fop = &gfs2_dir_fops;
122 	} else if (S_ISLNK(mode)) {
123 		inode->i_op = &gfs2_symlink_iops;
124 	} else {
125 		inode->i_op = &gfs2_file_iops;
126 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
127 	}
128 }
129 
130 /**
131  * gfs2_inode_lookup - Lookup an inode
132  * @sb: The super block
133  * @no_addr: The inode number
134  * @type: The type of the inode
135  * non_block: Can we block on inodes that are being freed?
136  *
137  * Returns: A VFS inode, or an error
138  */
139 
140 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
141 				u64 no_addr, u64 no_formal_ino, int non_block)
142 {
143 	struct inode *inode;
144 	struct gfs2_inode *ip;
145 	struct gfs2_glock *io_gl = NULL;
146 	int error;
147 
148 	inode = gfs2_iget(sb, no_addr, non_block);
149 	ip = GFS2_I(inode);
150 
151 	if (!inode)
152 		return ERR_PTR(-ENOBUFS);
153 
154 	if (inode->i_state & I_NEW) {
155 		struct gfs2_sbd *sdp = GFS2_SB(inode);
156 		ip->i_no_formal_ino = no_formal_ino;
157 
158 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
159 		if (unlikely(error))
160 			goto fail;
161 		ip->i_gl->gl_object = ip;
162 
163 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
164 		if (unlikely(error))
165 			goto fail_put;
166 
167 		set_bit(GIF_INVALID, &ip->i_flags);
168 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
169 		if (unlikely(error))
170 			goto fail_iopen;
171 
172 		ip->i_iopen_gh.gh_gl->gl_object = ip;
173 		gfs2_glock_put(io_gl);
174 		io_gl = NULL;
175 
176 		if (type == DT_UNKNOWN) {
177 			/* Inode glock must be locked already */
178 			error = gfs2_inode_refresh(GFS2_I(inode));
179 			if (error)
180 				goto fail_refresh;
181 		} else {
182 			inode->i_mode = DT2IF(type);
183 		}
184 
185 		gfs2_set_iop(inode);
186 		unlock_new_inode(inode);
187 	}
188 
189 	return inode;
190 
191 fail_refresh:
192 	ip->i_iopen_gh.gh_gl->gl_object = NULL;
193 	gfs2_glock_dq_uninit(&ip->i_iopen_gh);
194 fail_iopen:
195 	if (io_gl)
196 		gfs2_glock_put(io_gl);
197 fail_put:
198 	ip->i_gl->gl_object = NULL;
199 	gfs2_glock_put(ip->i_gl);
200 fail:
201 	iget_failed(inode);
202 	return ERR_PTR(error);
203 }
204 
205 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
206 				  u64 *no_formal_ino, unsigned int blktype)
207 {
208 	struct super_block *sb = sdp->sd_vfs;
209 	struct gfs2_holder i_gh;
210 	struct inode *inode = NULL;
211 	int error;
212 
213 	/* Must not read in block until block type is verified */
214 	error = gfs2_glock_nq_num(sdp, no_addr, &gfs2_inode_glops,
215 				  LM_ST_EXCLUSIVE, GL_SKIP, &i_gh);
216 	if (error)
217 		return ERR_PTR(error);
218 
219 	error = gfs2_check_blk_type(sdp, no_addr, blktype);
220 	if (error)
221 		goto fail;
222 
223 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0, 1);
224 	if (IS_ERR(inode))
225 		goto fail;
226 
227 	/* Two extra checks for NFS only */
228 	if (no_formal_ino) {
229 		error = -ESTALE;
230 		if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
231 			goto fail_iput;
232 
233 		error = -EIO;
234 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
235 			goto fail_iput;
236 
237 		error = 0;
238 	}
239 
240 fail:
241 	gfs2_glock_dq_uninit(&i_gh);
242 	return error ? ERR_PTR(error) : inode;
243 fail_iput:
244 	iput(inode);
245 	goto fail;
246 }
247 
248 
249 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
250 {
251 	struct qstr qstr;
252 	struct inode *inode;
253 	gfs2_str2qstr(&qstr, name);
254 	inode = gfs2_lookupi(dip, &qstr, 1);
255 	/* gfs2_lookupi has inconsistent callers: vfs
256 	 * related routines expect NULL for no entry found,
257 	 * gfs2_lookup_simple callers expect ENOENT
258 	 * and do not check for NULL.
259 	 */
260 	if (inode == NULL)
261 		return ERR_PTR(-ENOENT);
262 	else
263 		return inode;
264 }
265 
266 
267 /**
268  * gfs2_lookupi - Look up a filename in a directory and return its inode
269  * @d_gh: An initialized holder for the directory glock
270  * @name: The name of the inode to look for
271  * @is_root: If 1, ignore the caller's permissions
272  * @i_gh: An uninitialized holder for the new inode glock
273  *
274  * This can be called via the VFS filldir function when NFS is doing
275  * a readdirplus and the inode which its intending to stat isn't
276  * already in cache. In this case we must not take the directory glock
277  * again, since the readdir call will have already taken that lock.
278  *
279  * Returns: errno
280  */
281 
282 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
283 			   int is_root)
284 {
285 	struct super_block *sb = dir->i_sb;
286 	struct gfs2_inode *dip = GFS2_I(dir);
287 	struct gfs2_holder d_gh;
288 	int error = 0;
289 	struct inode *inode = NULL;
290 	int unlock = 0;
291 
292 	if (!name->len || name->len > GFS2_FNAMESIZE)
293 		return ERR_PTR(-ENAMETOOLONG);
294 
295 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
296 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
297 	     dir == sb->s_root->d_inode)) {
298 		igrab(dir);
299 		return dir;
300 	}
301 
302 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
303 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
304 		if (error)
305 			return ERR_PTR(error);
306 		unlock = 1;
307 	}
308 
309 	if (!is_root) {
310 		error = gfs2_permission(dir, MAY_EXEC);
311 		if (error)
312 			goto out;
313 	}
314 
315 	inode = gfs2_dir_search(dir, name);
316 	if (IS_ERR(inode))
317 		error = PTR_ERR(inode);
318 out:
319 	if (unlock)
320 		gfs2_glock_dq_uninit(&d_gh);
321 	if (error == -ENOENT)
322 		return NULL;
323 	return inode ? inode : ERR_PTR(error);
324 }
325 
326 /**
327  * create_ok - OK to create a new on-disk inode here?
328  * @dip:  Directory in which dinode is to be created
329  * @name:  Name of new dinode
330  * @mode:
331  *
332  * Returns: errno
333  */
334 
335 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
336 		     umode_t mode)
337 {
338 	int error;
339 
340 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
341 	if (error)
342 		return error;
343 
344 	/*  Don't create entries in an unlinked directory  */
345 	if (!dip->i_inode.i_nlink)
346 		return -ENOENT;
347 
348 	error = gfs2_dir_check(&dip->i_inode, name, NULL);
349 	switch (error) {
350 	case -ENOENT:
351 		error = 0;
352 		break;
353 	case 0:
354 		return -EEXIST;
355 	default:
356 		return error;
357 	}
358 
359 	if (dip->i_entries == (u32)-1)
360 		return -EFBIG;
361 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
362 		return -EMLINK;
363 
364 	return 0;
365 }
366 
367 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
368 			       struct inode *inode)
369 {
370 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
371 	    (dip->i_inode.i_mode & S_ISUID) &&
372 	    !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
373 		if (S_ISDIR(inode->i_mode))
374 			inode->i_mode |= S_ISUID;
375 		else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
376 			inode->i_mode &= ~07111;
377 		inode->i_uid = dip->i_inode.i_uid;
378 	} else
379 		inode->i_uid = current_fsuid();
380 
381 	if (dip->i_inode.i_mode & S_ISGID) {
382 		if (S_ISDIR(inode->i_mode))
383 			inode->i_mode |= S_ISGID;
384 		inode->i_gid = dip->i_inode.i_gid;
385 	} else
386 		inode->i_gid = current_fsgid();
387 }
388 
389 static int alloc_dinode(struct gfs2_inode *ip, u32 flags)
390 {
391 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
392 	int error;
393 	int dblocks = 1;
394 
395 	error = gfs2_inplace_reserve(ip, RES_DINODE, flags);
396 	if (error)
397 		goto out;
398 
399 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
400 	if (error)
401 		goto out_ipreserv;
402 
403 	error = gfs2_alloc_blocks(ip, &ip->i_no_addr, &dblocks, 1, &ip->i_generation);
404 	ip->i_no_formal_ino = ip->i_generation;
405 	ip->i_inode.i_ino = ip->i_no_addr;
406 	ip->i_goal = ip->i_no_addr;
407 
408 	gfs2_trans_end(sdp);
409 
410 out_ipreserv:
411 	gfs2_inplace_release(ip);
412 out:
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  * @ip: The inode
437  * @symname: The symlink destination (if a symlink)
438  * @bhp: The buffer head (returned to caller)
439  *
440  */
441 
442 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
443 			const char *symname, struct buffer_head **bhp)
444 {
445 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
446 	struct gfs2_dinode *di;
447 	struct buffer_head *dibh;
448 	struct timespec tv = CURRENT_TIME;
449 
450 	dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
451 	gfs2_trans_add_meta(ip->i_gl, dibh);
452 	gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
453 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
454 	di = (struct gfs2_dinode *)dibh->b_data;
455 
456 	di->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
457 	di->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
458 	di->di_mode = cpu_to_be32(ip->i_inode.i_mode);
459 	di->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
460 	di->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
461 	di->di_nlink = 0;
462 	di->di_size = cpu_to_be64(ip->i_inode.i_size);
463 	di->di_blocks = cpu_to_be64(1);
464 	di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(tv.tv_sec);
465 	di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
466 	di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
467 	di->di_goal_meta = di->di_goal_data = cpu_to_be64(ip->i_no_addr);
468 	di->di_generation = cpu_to_be64(ip->i_generation);
469 	di->di_flags = 0;
470 	di->__pad1 = 0;
471 	di->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) ? GFS2_FORMAT_DE : 0);
472 	di->di_height = 0;
473 	di->__pad2 = 0;
474 	di->__pad3 = 0;
475 	di->di_depth = 0;
476 	di->di_entries = 0;
477 	memset(&di->__pad4, 0, sizeof(di->__pad4));
478 	di->di_eattr = 0;
479 	di->di_atime_nsec = cpu_to_be32(tv.tv_nsec);
480 	di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec);
481 	di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec);
482 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
483 
484 	switch(ip->i_inode.i_mode & S_IFMT) {
485 	case S_IFREG:
486 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
487 		    gfs2_tune_get(sdp, gt_new_files_jdata))
488 			di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
489 		break;
490 	case S_IFDIR:
491 		di->di_flags |= cpu_to_be32(dip->i_diskflags &
492 					    GFS2_DIF_INHERIT_JDATA);
493 		di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
494 		di->di_size = cpu_to_be64(sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode));
495 		di->di_entries = cpu_to_be32(2);
496 		gfs2_init_dir(dibh, dip);
497 		break;
498 	case S_IFLNK:
499 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
500 		break;
501 	}
502 
503 	set_buffer_uptodate(dibh);
504 
505 	*bhp = dibh;
506 }
507 
508 static int make_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
509 		       const char *symname, struct buffer_head **bhp)
510 {
511 	struct inode *inode = &ip->i_inode;
512 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
513 	int error;
514 
515 	error = gfs2_rindex_update(sdp);
516 	if (error)
517 		return error;
518 
519 	error = gfs2_quota_lock(dip, inode->i_uid, inode->i_gid);
520 	if (error)
521 		return error;
522 
523 	error = gfs2_quota_check(dip, inode->i_uid, inode->i_gid);
524 	if (error)
525 		goto out_quota;
526 
527 	error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
528 	if (error)
529 		goto out_quota;
530 
531 	init_dinode(dip, ip, symname, bhp);
532 	gfs2_quota_change(dip, +1, inode->i_uid, inode->i_gid);
533 	gfs2_trans_end(sdp);
534 
535 out_quota:
536 	gfs2_quota_unlock(dip);
537 	return error;
538 }
539 
540 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
541 		       struct gfs2_inode *ip)
542 {
543 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
544 	int alloc_required;
545 	struct buffer_head *dibh;
546 	int error;
547 
548 	error = gfs2_rindex_update(sdp);
549 	if (error)
550 		return error;
551 
552 	error = gfs2_quota_lock(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
553 	if (error)
554 		goto fail;
555 
556 	error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
557 	if (alloc_required < 0)
558 		goto fail_quota_locks;
559 	if (alloc_required) {
560 		error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
561 		if (error)
562 			goto fail_quota_locks;
563 
564 		error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
565 		if (error)
566 			goto fail_quota_locks;
567 
568 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
569 					 dip->i_rgd->rd_length +
570 					 2 * RES_DINODE +
571 					 RES_STATFS + RES_QUOTA, 0);
572 		if (error)
573 			goto fail_ipreserv;
574 	} else {
575 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
576 		if (error)
577 			goto fail_quota_locks;
578 	}
579 
580 	error = gfs2_dir_add(&dip->i_inode, name, ip);
581 	if (error)
582 		goto fail_end_trans;
583 
584 	error = gfs2_meta_inode_buffer(ip, &dibh);
585 	if (error)
586 		goto fail_end_trans;
587 	set_nlink(&ip->i_inode, S_ISDIR(ip->i_inode.i_mode) ? 2 : 1);
588 	gfs2_trans_add_meta(ip->i_gl, dibh);
589 	gfs2_dinode_out(ip, dibh->b_data);
590 	brelse(dibh);
591 	return 0;
592 
593 fail_end_trans:
594 	gfs2_trans_end(sdp);
595 
596 fail_ipreserv:
597 	if (alloc_required)
598 		gfs2_inplace_release(dip);
599 
600 fail_quota_locks:
601 	gfs2_quota_unlock(dip);
602 
603 fail:
604 	return error;
605 }
606 
607 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
608 		    void *fs_info)
609 {
610 	const struct xattr *xattr;
611 	int err = 0;
612 
613 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
614 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
615 				       xattr->value_len, 0,
616 				       GFS2_EATYPE_SECURITY);
617 		if (err < 0)
618 			break;
619 	}
620 	return err;
621 }
622 
623 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
624 			      const struct qstr *qstr)
625 {
626 	return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
627 					    &gfs2_initxattrs, NULL);
628 }
629 
630 /**
631  * gfs2_create_inode - Create a new inode
632  * @dir: The parent directory
633  * @dentry: The new dentry
634  * @mode: The permissions on the new inode
635  * @dev: For device nodes, this is the device number
636  * @symname: For symlinks, this is the link destination
637  * @size: The initial size of the inode (ignored for directories)
638  *
639  * Returns: 0 on success, or error code
640  */
641 
642 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
643 			     umode_t mode, dev_t dev, const char *symname,
644 			     unsigned int size, int excl)
645 {
646 	const struct qstr *name = &dentry->d_name;
647 	struct gfs2_holder ghs[2];
648 	struct inode *inode = NULL;
649 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
650 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
651 	struct gfs2_glock *io_gl;
652 	int error;
653 	struct buffer_head *bh = NULL;
654 	u32 aflags = 0;
655 
656 	if (!name->len || name->len > GFS2_FNAMESIZE)
657 		return -ENAMETOOLONG;
658 
659 	error = gfs2_rs_alloc(dip);
660 	if (error)
661 		return error;
662 
663 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
664 	if (error)
665 		goto fail;
666 
667 	error = create_ok(dip, name, mode);
668 	if ((error == -EEXIST) && S_ISREG(mode) && !excl) {
669 		inode = gfs2_lookupi(dir, &dentry->d_name, 0);
670 		gfs2_glock_dq_uninit(ghs);
671 		d_instantiate(dentry, inode);
672 		return IS_ERR(inode) ? PTR_ERR(inode) : 0;
673 	}
674 	if (error)
675 		goto fail_gunlock;
676 
677 	inode = new_inode(sdp->sd_vfs);
678 	if (!inode) {
679 		gfs2_glock_dq_uninit(ghs);
680 		return -ENOMEM;
681 	}
682 	ip = GFS2_I(inode);
683 	error = gfs2_rs_alloc(ip);
684 	if (error)
685 		goto fail_free_inode;
686 
687 	set_bit(GIF_INVALID, &ip->i_flags);
688 	inode->i_mode = mode;
689 	inode->i_rdev = dev;
690 	inode->i_size = size;
691 	munge_mode_uid_gid(dip, inode);
692 	ip->i_goal = dip->i_goal;
693 
694 	if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
695 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
696 		aflags |= GFS2_AF_ORLOV;
697 
698 	error = alloc_dinode(ip, aflags);
699 	if (error)
700 		goto fail_free_inode;
701 
702 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
703 	if (error)
704 		goto fail_free_inode;
705 
706 	ip->i_gl->gl_object = ip;
707 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
708 	if (error)
709 		goto fail_free_inode;
710 
711 	error = make_dinode(dip, ip, symname, &bh);
712 	if (error)
713 		goto fail_gunlock2;
714 
715 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
716 	if (error)
717 		goto fail_gunlock2;
718 
719 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
720 	if (error)
721 		goto fail_gunlock2;
722 
723 	ip->i_iopen_gh.gh_gl->gl_object = ip;
724 	gfs2_glock_put(io_gl);
725 	gfs2_set_iop(inode);
726 	insert_inode_hash(inode);
727 
728 	error = gfs2_inode_refresh(ip);
729 	if (error)
730 		goto fail_gunlock3;
731 
732 	error = gfs2_acl_create(dip, inode);
733 	if (error)
734 		goto fail_gunlock3;
735 
736 	error = gfs2_security_init(dip, ip, name);
737 	if (error)
738 		goto fail_gunlock3;
739 
740 	error = link_dinode(dip, name, ip);
741 	if (error)
742 		goto fail_gunlock3;
743 
744 	if (bh)
745 		brelse(bh);
746 
747 	gfs2_trans_end(sdp);
748 	gfs2_inplace_release(dip);
749 	gfs2_quota_unlock(dip);
750 	mark_inode_dirty(inode);
751 	gfs2_glock_dq_uninit_m(2, ghs);
752 	d_instantiate(dentry, inode);
753 	return 0;
754 
755 fail_gunlock3:
756 	gfs2_glock_dq_uninit(ghs + 1);
757 	if (ip->i_gl)
758 		gfs2_glock_put(ip->i_gl);
759 	goto fail_gunlock;
760 
761 fail_gunlock2:
762 	gfs2_glock_dq_uninit(ghs + 1);
763 fail_free_inode:
764 	if (ip->i_gl)
765 		gfs2_glock_put(ip->i_gl);
766 	gfs2_rs_delete(ip);
767 	free_inode_nonrcu(inode);
768 	inode = NULL;
769 fail_gunlock:
770 	gfs2_glock_dq_uninit(ghs);
771 	if (inode && !IS_ERR(inode)) {
772 		set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
773 		iput(inode);
774 	}
775 fail:
776 	if (bh)
777 		brelse(bh);
778 	return error;
779 }
780 
781 /**
782  * gfs2_create - Create a file
783  * @dir: The directory in which to create the file
784  * @dentry: The dentry of the new file
785  * @mode: The mode of the new file
786  *
787  * Returns: errno
788  */
789 
790 static int gfs2_create(struct inode *dir, struct dentry *dentry,
791 		       umode_t mode, bool excl)
792 {
793 	return gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0, excl);
794 }
795 
796 /**
797  * gfs2_lookup - Look up a filename in a directory and return its inode
798  * @dir: The directory inode
799  * @dentry: The dentry of the new inode
800  * @nd: passed from Linux VFS, ignored by us
801  *
802  * Called by the VFS layer. Lock dir and call gfs2_lookupi()
803  *
804  * Returns: errno
805  */
806 
807 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
808 				  unsigned int flags)
809 {
810 	struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
811 	if (inode && !IS_ERR(inode)) {
812 		struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
813 		struct gfs2_holder gh;
814 		int error;
815 		error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
816 		if (error) {
817 			iput(inode);
818 			return ERR_PTR(error);
819 		}
820 		gfs2_glock_dq_uninit(&gh);
821 	}
822 	return d_splice_alias(inode, dentry);
823 }
824 
825 /**
826  * gfs2_link - Link to a file
827  * @old_dentry: The inode to link
828  * @dir: Add link to this directory
829  * @dentry: The name of the link
830  *
831  * Link the inode in "old_dentry" into the directory "dir" with the
832  * name in "dentry".
833  *
834  * Returns: errno
835  */
836 
837 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
838 		     struct dentry *dentry)
839 {
840 	struct gfs2_inode *dip = GFS2_I(dir);
841 	struct gfs2_sbd *sdp = GFS2_SB(dir);
842 	struct inode *inode = old_dentry->d_inode;
843 	struct gfs2_inode *ip = GFS2_I(inode);
844 	struct gfs2_holder ghs[2];
845 	struct buffer_head *dibh;
846 	int alloc_required;
847 	int error;
848 
849 	if (S_ISDIR(inode->i_mode))
850 		return -EPERM;
851 
852 	error = gfs2_rs_alloc(dip);
853 	if (error)
854 		return error;
855 
856 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
857 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
858 
859 	error = gfs2_glock_nq(ghs); /* parent */
860 	if (error)
861 		goto out_parent;
862 
863 	error = gfs2_glock_nq(ghs + 1); /* child */
864 	if (error)
865 		goto out_child;
866 
867 	error = -ENOENT;
868 	if (inode->i_nlink == 0)
869 		goto out_gunlock;
870 
871 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
872 	if (error)
873 		goto out_gunlock;
874 
875 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
876 	switch (error) {
877 	case -ENOENT:
878 		break;
879 	case 0:
880 		error = -EEXIST;
881 	default:
882 		goto out_gunlock;
883 	}
884 
885 	error = -EINVAL;
886 	if (!dip->i_inode.i_nlink)
887 		goto out_gunlock;
888 	error = -EFBIG;
889 	if (dip->i_entries == (u32)-1)
890 		goto out_gunlock;
891 	error = -EPERM;
892 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
893 		goto out_gunlock;
894 	error = -EINVAL;
895 	if (!ip->i_inode.i_nlink)
896 		goto out_gunlock;
897 	error = -EMLINK;
898 	if (ip->i_inode.i_nlink == (u32)-1)
899 		goto out_gunlock;
900 
901 	alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
902 	if (error < 0)
903 		goto out_gunlock;
904 	error = 0;
905 
906 	if (alloc_required) {
907 		error = gfs2_quota_lock_check(dip);
908 		if (error)
909 			goto out_gunlock;
910 
911 		error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
912 		if (error)
913 			goto out_gunlock_q;
914 
915 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
916 					 gfs2_rg_blocks(dip, sdp->sd_max_dirres) +
917 					 2 * RES_DINODE + RES_STATFS +
918 					 RES_QUOTA, 0);
919 		if (error)
920 			goto out_ipres;
921 	} else {
922 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
923 		if (error)
924 			goto out_ipres;
925 	}
926 
927 	error = gfs2_meta_inode_buffer(ip, &dibh);
928 	if (error)
929 		goto out_end_trans;
930 
931 	error = gfs2_dir_add(dir, &dentry->d_name, ip);
932 	if (error)
933 		goto out_brelse;
934 
935 	gfs2_trans_add_meta(ip->i_gl, dibh);
936 	inc_nlink(&ip->i_inode);
937 	ip->i_inode.i_ctime = CURRENT_TIME;
938 	ihold(inode);
939 	d_instantiate(dentry, inode);
940 	mark_inode_dirty(inode);
941 
942 out_brelse:
943 	brelse(dibh);
944 out_end_trans:
945 	gfs2_trans_end(sdp);
946 out_ipres:
947 	if (alloc_required)
948 		gfs2_inplace_release(dip);
949 out_gunlock_q:
950 	if (alloc_required)
951 		gfs2_quota_unlock(dip);
952 out_gunlock:
953 	gfs2_glock_dq(ghs + 1);
954 out_child:
955 	gfs2_glock_dq(ghs);
956 out_parent:
957 	gfs2_holder_uninit(ghs);
958 	gfs2_holder_uninit(ghs + 1);
959 	return error;
960 }
961 
962 /*
963  * gfs2_unlink_ok - check to see that a inode is still in a directory
964  * @dip: the directory
965  * @name: the name of the file
966  * @ip: the inode
967  *
968  * Assumes that the lock on (at least) @dip is held.
969  *
970  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
971  */
972 
973 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
974 			  const struct gfs2_inode *ip)
975 {
976 	int error;
977 
978 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
979 		return -EPERM;
980 
981 	if ((dip->i_inode.i_mode & S_ISVTX) &&
982 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
983 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
984 		return -EPERM;
985 
986 	if (IS_APPEND(&dip->i_inode))
987 		return -EPERM;
988 
989 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
990 	if (error)
991 		return error;
992 
993 	error = gfs2_dir_check(&dip->i_inode, name, ip);
994 	if (error)
995 		return error;
996 
997 	return 0;
998 }
999 
1000 /**
1001  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1002  * @dip: The parent directory
1003  * @name: The name of the entry in the parent directory
1004  * @inode: The inode to be removed
1005  *
1006  * Called with all the locks and in a transaction. This will only be
1007  * called for a directory after it has been checked to ensure it is empty.
1008  *
1009  * Returns: 0 on success, or an error
1010  */
1011 
1012 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1013 			     const struct dentry *dentry)
1014 {
1015 	struct inode *inode = dentry->d_inode;
1016 	struct gfs2_inode *ip = GFS2_I(inode);
1017 	int error;
1018 
1019 	error = gfs2_dir_del(dip, dentry);
1020 	if (error)
1021 		return error;
1022 
1023 	ip->i_entries = 0;
1024 	inode->i_ctime = CURRENT_TIME;
1025 	if (S_ISDIR(inode->i_mode))
1026 		clear_nlink(inode);
1027 	else
1028 		drop_nlink(inode);
1029 	mark_inode_dirty(inode);
1030 	if (inode->i_nlink == 0)
1031 		gfs2_unlink_di(inode);
1032 	return 0;
1033 }
1034 
1035 
1036 /**
1037  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1038  * @dir: The inode of the directory containing the inode to unlink
1039  * @dentry: The file itself
1040  *
1041  * This routine uses the type of the inode as a flag to figure out
1042  * whether this is an unlink or an rmdir.
1043  *
1044  * Returns: errno
1045  */
1046 
1047 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1048 {
1049 	struct gfs2_inode *dip = GFS2_I(dir);
1050 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1051 	struct inode *inode = dentry->d_inode;
1052 	struct gfs2_inode *ip = GFS2_I(inode);
1053 	struct gfs2_holder ghs[3];
1054 	struct gfs2_rgrpd *rgd;
1055 	int error;
1056 
1057 	error = gfs2_rindex_update(sdp);
1058 	if (error)
1059 		return error;
1060 
1061 	error = -EROFS;
1062 
1063 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1064 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1065 
1066 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1067 	if (!rgd)
1068 		goto out_inodes;
1069 
1070 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1071 
1072 
1073 	error = gfs2_glock_nq(ghs); /* parent */
1074 	if (error)
1075 		goto out_parent;
1076 
1077 	error = gfs2_glock_nq(ghs + 1); /* child */
1078 	if (error)
1079 		goto out_child;
1080 
1081 	error = -ENOENT;
1082 	if (inode->i_nlink == 0)
1083 		goto out_rgrp;
1084 
1085 	if (S_ISDIR(inode->i_mode)) {
1086 		error = -ENOTEMPTY;
1087 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1088 			goto out_rgrp;
1089 	}
1090 
1091 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1092 	if (error)
1093 		goto out_rgrp;
1094 
1095 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1096 	if (error)
1097 		goto out_gunlock;
1098 
1099 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1100 	if (error)
1101 		goto out_end_trans;
1102 
1103 	error = gfs2_unlink_inode(dip, dentry);
1104 
1105 out_end_trans:
1106 	gfs2_trans_end(sdp);
1107 out_gunlock:
1108 	gfs2_glock_dq(ghs + 2);
1109 out_rgrp:
1110 	gfs2_glock_dq(ghs + 1);
1111 out_child:
1112 	gfs2_glock_dq(ghs);
1113 out_parent:
1114 	gfs2_holder_uninit(ghs + 2);
1115 out_inodes:
1116 	gfs2_holder_uninit(ghs + 1);
1117 	gfs2_holder_uninit(ghs);
1118 	return error;
1119 }
1120 
1121 /**
1122  * gfs2_symlink - Create a symlink
1123  * @dir: The directory to create the symlink in
1124  * @dentry: The dentry to put the symlink in
1125  * @symname: The thing which the link points to
1126  *
1127  * Returns: errno
1128  */
1129 
1130 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1131 			const char *symname)
1132 {
1133 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1134 	unsigned int size;
1135 
1136 	size = strlen(symname);
1137 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1138 		return -ENAMETOOLONG;
1139 
1140 	return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1141 }
1142 
1143 /**
1144  * gfs2_mkdir - Make a directory
1145  * @dir: The parent directory of the new one
1146  * @dentry: The dentry of the new directory
1147  * @mode: The mode of the new directory
1148  *
1149  * Returns: errno
1150  */
1151 
1152 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1153 {
1154 	return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, 0, 0);
1155 }
1156 
1157 /**
1158  * gfs2_mknod - Make a special file
1159  * @dir: The directory in which the special file will reside
1160  * @dentry: The dentry of the special file
1161  * @mode: The mode of the special file
1162  * @dev: The device specification of the special file
1163  *
1164  */
1165 
1166 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1167 		      dev_t dev)
1168 {
1169 	return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0, 0);
1170 }
1171 
1172 /*
1173  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1174  * @this: move this
1175  * @to: to here
1176  *
1177  * Follow @to back to the root and make sure we don't encounter @this
1178  * Assumes we already hold the rename lock.
1179  *
1180  * Returns: errno
1181  */
1182 
1183 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1184 {
1185 	struct inode *dir = &to->i_inode;
1186 	struct super_block *sb = dir->i_sb;
1187 	struct inode *tmp;
1188 	int error = 0;
1189 
1190 	igrab(dir);
1191 
1192 	for (;;) {
1193 		if (dir == &this->i_inode) {
1194 			error = -EINVAL;
1195 			break;
1196 		}
1197 		if (dir == sb->s_root->d_inode) {
1198 			error = 0;
1199 			break;
1200 		}
1201 
1202 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1203 		if (IS_ERR(tmp)) {
1204 			error = PTR_ERR(tmp);
1205 			break;
1206 		}
1207 
1208 		iput(dir);
1209 		dir = tmp;
1210 	}
1211 
1212 	iput(dir);
1213 
1214 	return error;
1215 }
1216 
1217 /**
1218  * gfs2_rename - Rename a file
1219  * @odir: Parent directory of old file name
1220  * @odentry: The old dentry of the file
1221  * @ndir: Parent directory of new file name
1222  * @ndentry: The new dentry of the file
1223  *
1224  * Returns: errno
1225  */
1226 
1227 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1228 		       struct inode *ndir, struct dentry *ndentry)
1229 {
1230 	struct gfs2_inode *odip = GFS2_I(odir);
1231 	struct gfs2_inode *ndip = GFS2_I(ndir);
1232 	struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1233 	struct gfs2_inode *nip = NULL;
1234 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1235 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1236 	struct gfs2_rgrpd *nrgd;
1237 	unsigned int num_gh;
1238 	int dir_rename = 0;
1239 	int alloc_required = 0;
1240 	unsigned int x;
1241 	int error;
1242 
1243 	if (ndentry->d_inode) {
1244 		nip = GFS2_I(ndentry->d_inode);
1245 		if (ip == nip)
1246 			return 0;
1247 	}
1248 
1249 	error = gfs2_rindex_update(sdp);
1250 	if (error)
1251 		return error;
1252 
1253 	error = gfs2_rs_alloc(ndip);
1254 	if (error)
1255 		return error;
1256 
1257 	if (odip != ndip) {
1258 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1259 					   0, &r_gh);
1260 		if (error)
1261 			goto out;
1262 
1263 		if (S_ISDIR(ip->i_inode.i_mode)) {
1264 			dir_rename = 1;
1265 			/* don't move a dirctory into it's subdir */
1266 			error = gfs2_ok_to_move(ip, ndip);
1267 			if (error)
1268 				goto out_gunlock_r;
1269 		}
1270 	}
1271 
1272 	num_gh = 1;
1273 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1274 	if (odip != ndip) {
1275 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1276 		num_gh++;
1277 	}
1278 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1279 	num_gh++;
1280 
1281 	if (nip) {
1282 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1283 		num_gh++;
1284 		/* grab the resource lock for unlink flag twiddling
1285 		 * this is the case of the target file already existing
1286 		 * so we unlink before doing the rename
1287 		 */
1288 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1289 		if (nrgd)
1290 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1291 	}
1292 
1293 	for (x = 0; x < num_gh; x++) {
1294 		error = gfs2_glock_nq(ghs + x);
1295 		if (error)
1296 			goto out_gunlock;
1297 	}
1298 
1299 	error = -ENOENT;
1300 	if (ip->i_inode.i_nlink == 0)
1301 		goto out_gunlock;
1302 
1303 	/* Check out the old directory */
1304 
1305 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1306 	if (error)
1307 		goto out_gunlock;
1308 
1309 	/* Check out the new directory */
1310 
1311 	if (nip) {
1312 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1313 		if (error)
1314 			goto out_gunlock;
1315 
1316 		if (nip->i_inode.i_nlink == 0) {
1317 			error = -EAGAIN;
1318 			goto out_gunlock;
1319 		}
1320 
1321 		if (S_ISDIR(nip->i_inode.i_mode)) {
1322 			if (nip->i_entries < 2) {
1323 				gfs2_consist_inode(nip);
1324 				error = -EIO;
1325 				goto out_gunlock;
1326 			}
1327 			if (nip->i_entries > 2) {
1328 				error = -ENOTEMPTY;
1329 				goto out_gunlock;
1330 			}
1331 		}
1332 	} else {
1333 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1334 		if (error)
1335 			goto out_gunlock;
1336 
1337 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1338 		switch (error) {
1339 		case -ENOENT:
1340 			error = 0;
1341 			break;
1342 		case 0:
1343 			error = -EEXIST;
1344 		default:
1345 			goto out_gunlock;
1346 		};
1347 
1348 		if (odip != ndip) {
1349 			if (!ndip->i_inode.i_nlink) {
1350 				error = -ENOENT;
1351 				goto out_gunlock;
1352 			}
1353 			if (ndip->i_entries == (u32)-1) {
1354 				error = -EFBIG;
1355 				goto out_gunlock;
1356 			}
1357 			if (S_ISDIR(ip->i_inode.i_mode) &&
1358 			    ndip->i_inode.i_nlink == (u32)-1) {
1359 				error = -EMLINK;
1360 				goto out_gunlock;
1361 			}
1362 		}
1363 	}
1364 
1365 	/* Check out the dir to be renamed */
1366 
1367 	if (dir_rename) {
1368 		error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1369 		if (error)
1370 			goto out_gunlock;
1371 	}
1372 
1373 	if (nip == NULL)
1374 		alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1375 	error = alloc_required;
1376 	if (error < 0)
1377 		goto out_gunlock;
1378 
1379 	if (alloc_required) {
1380 		error = gfs2_quota_lock_check(ndip);
1381 		if (error)
1382 			goto out_gunlock;
1383 
1384 		error = gfs2_inplace_reserve(ndip, sdp->sd_max_dirres, 0);
1385 		if (error)
1386 			goto out_gunlock_q;
1387 
1388 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1389 					 gfs2_rg_blocks(ndip, sdp->sd_max_dirres) +
1390 					 4 * RES_DINODE + 4 * RES_LEAF +
1391 					 RES_STATFS + RES_QUOTA + 4, 0);
1392 		if (error)
1393 			goto out_ipreserv;
1394 	} else {
1395 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1396 					 5 * RES_LEAF + 4, 0);
1397 		if (error)
1398 			goto out_gunlock;
1399 	}
1400 
1401 	/* Remove the target file, if it exists */
1402 
1403 	if (nip)
1404 		error = gfs2_unlink_inode(ndip, ndentry);
1405 
1406 	if (dir_rename) {
1407 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1408 		if (error)
1409 			goto out_end_trans;
1410 	} else {
1411 		struct buffer_head *dibh;
1412 		error = gfs2_meta_inode_buffer(ip, &dibh);
1413 		if (error)
1414 			goto out_end_trans;
1415 		ip->i_inode.i_ctime = CURRENT_TIME;
1416 		gfs2_trans_add_meta(ip->i_gl, dibh);
1417 		gfs2_dinode_out(ip, dibh->b_data);
1418 		brelse(dibh);
1419 	}
1420 
1421 	error = gfs2_dir_del(odip, odentry);
1422 	if (error)
1423 		goto out_end_trans;
1424 
1425 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1426 	if (error)
1427 		goto out_end_trans;
1428 
1429 out_end_trans:
1430 	gfs2_trans_end(sdp);
1431 out_ipreserv:
1432 	if (alloc_required)
1433 		gfs2_inplace_release(ndip);
1434 out_gunlock_q:
1435 	if (alloc_required)
1436 		gfs2_quota_unlock(ndip);
1437 out_gunlock:
1438 	while (x--) {
1439 		gfs2_glock_dq(ghs + x);
1440 		gfs2_holder_uninit(ghs + x);
1441 	}
1442 out_gunlock_r:
1443 	if (r_gh.gh_gl)
1444 		gfs2_glock_dq_uninit(&r_gh);
1445 out:
1446 	return error;
1447 }
1448 
1449 /**
1450  * gfs2_follow_link - Follow a symbolic link
1451  * @dentry: The dentry of the link
1452  * @nd: Data that we pass to vfs_follow_link()
1453  *
1454  * This can handle symlinks of any size.
1455  *
1456  * Returns: 0 on success or error code
1457  */
1458 
1459 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1460 {
1461 	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1462 	struct gfs2_holder i_gh;
1463 	struct buffer_head *dibh;
1464 	unsigned int size;
1465 	char *buf;
1466 	int error;
1467 
1468 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1469 	error = gfs2_glock_nq(&i_gh);
1470 	if (error) {
1471 		gfs2_holder_uninit(&i_gh);
1472 		nd_set_link(nd, ERR_PTR(error));
1473 		return NULL;
1474 	}
1475 
1476 	size = (unsigned int)i_size_read(&ip->i_inode);
1477 	if (size == 0) {
1478 		gfs2_consist_inode(ip);
1479 		buf = ERR_PTR(-EIO);
1480 		goto out;
1481 	}
1482 
1483 	error = gfs2_meta_inode_buffer(ip, &dibh);
1484 	if (error) {
1485 		buf = ERR_PTR(error);
1486 		goto out;
1487 	}
1488 
1489 	buf = kzalloc(size + 1, GFP_NOFS);
1490 	if (!buf)
1491 		buf = ERR_PTR(-ENOMEM);
1492 	else
1493 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1494 	brelse(dibh);
1495 out:
1496 	gfs2_glock_dq_uninit(&i_gh);
1497 	nd_set_link(nd, buf);
1498 	return NULL;
1499 }
1500 
1501 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1502 {
1503 	char *s = nd_get_link(nd);
1504 	if (!IS_ERR(s))
1505 		kfree(s);
1506 }
1507 
1508 /**
1509  * gfs2_permission -
1510  * @inode: The inode
1511  * @mask: The mask to be tested
1512  * @flags: Indicates whether this is an RCU path walk or not
1513  *
1514  * This may be called from the VFS directly, or from within GFS2 with the
1515  * inode locked, so we look to see if the glock is already locked and only
1516  * lock the glock if its not already been done.
1517  *
1518  * Returns: errno
1519  */
1520 
1521 int gfs2_permission(struct inode *inode, int mask)
1522 {
1523 	struct gfs2_inode *ip;
1524 	struct gfs2_holder i_gh;
1525 	int error;
1526 	int unlock = 0;
1527 
1528 
1529 	ip = GFS2_I(inode);
1530 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1531 		if (mask & MAY_NOT_BLOCK)
1532 			return -ECHILD;
1533 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1534 		if (error)
1535 			return error;
1536 		unlock = 1;
1537 	}
1538 
1539 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1540 		error = -EACCES;
1541 	else
1542 		error = generic_permission(inode, mask);
1543 	if (unlock)
1544 		gfs2_glock_dq_uninit(&i_gh);
1545 
1546 	return error;
1547 }
1548 
1549 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1550 {
1551 	setattr_copy(inode, attr);
1552 	mark_inode_dirty(inode);
1553 	return 0;
1554 }
1555 
1556 /**
1557  * gfs2_setattr_simple -
1558  * @ip:
1559  * @attr:
1560  *
1561  * Returns: errno
1562  */
1563 
1564 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1565 {
1566 	int error;
1567 
1568 	if (current->journal_info)
1569 		return __gfs2_setattr_simple(inode, attr);
1570 
1571 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1572 	if (error)
1573 		return error;
1574 
1575 	error = __gfs2_setattr_simple(inode, attr);
1576 	gfs2_trans_end(GFS2_SB(inode));
1577 	return error;
1578 }
1579 
1580 static int setattr_chown(struct inode *inode, struct iattr *attr)
1581 {
1582 	struct gfs2_inode *ip = GFS2_I(inode);
1583 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1584 	kuid_t ouid, nuid;
1585 	kgid_t ogid, ngid;
1586 	int error;
1587 
1588 	ouid = inode->i_uid;
1589 	ogid = inode->i_gid;
1590 	nuid = attr->ia_uid;
1591 	ngid = attr->ia_gid;
1592 
1593 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1594 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1595 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1596 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1597 
1598 	error = gfs2_quota_lock(ip, nuid, ngid);
1599 	if (error)
1600 		return error;
1601 
1602 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1603 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1604 		error = gfs2_quota_check(ip, nuid, ngid);
1605 		if (error)
1606 			goto out_gunlock_q;
1607 	}
1608 
1609 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1610 	if (error)
1611 		goto out_gunlock_q;
1612 
1613 	error = gfs2_setattr_simple(inode, attr);
1614 	if (error)
1615 		goto out_end_trans;
1616 
1617 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1618 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1619 		u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1620 		gfs2_quota_change(ip, -blocks, ouid, ogid);
1621 		gfs2_quota_change(ip, blocks, nuid, ngid);
1622 	}
1623 
1624 out_end_trans:
1625 	gfs2_trans_end(sdp);
1626 out_gunlock_q:
1627 	gfs2_quota_unlock(ip);
1628 	return error;
1629 }
1630 
1631 /**
1632  * gfs2_setattr - Change attributes on an inode
1633  * @dentry: The dentry which is changing
1634  * @attr: The structure describing the change
1635  *
1636  * The VFS layer wants to change one or more of an inodes attributes.  Write
1637  * that change out to disk.
1638  *
1639  * Returns: errno
1640  */
1641 
1642 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1643 {
1644 	struct inode *inode = dentry->d_inode;
1645 	struct gfs2_inode *ip = GFS2_I(inode);
1646 	struct gfs2_holder i_gh;
1647 	int error;
1648 
1649 	error = gfs2_rs_alloc(ip);
1650 	if (error)
1651 		return error;
1652 
1653 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1654 	if (error)
1655 		return error;
1656 
1657 	error = -EPERM;
1658 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1659 		goto out;
1660 
1661 	error = inode_change_ok(inode, attr);
1662 	if (error)
1663 		goto out;
1664 
1665 	if (attr->ia_valid & ATTR_SIZE)
1666 		error = gfs2_setattr_size(inode, attr->ia_size);
1667 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1668 		error = setattr_chown(inode, attr);
1669 	else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1670 		error = gfs2_acl_chmod(ip, attr);
1671 	else
1672 		error = gfs2_setattr_simple(inode, attr);
1673 
1674 out:
1675 	if (!error)
1676 		mark_inode_dirty(inode);
1677 	gfs2_glock_dq_uninit(&i_gh);
1678 	return error;
1679 }
1680 
1681 /**
1682  * gfs2_getattr - Read out an inode's attributes
1683  * @mnt: The vfsmount the inode is being accessed from
1684  * @dentry: The dentry to stat
1685  * @stat: The inode's stats
1686  *
1687  * This may be called from the VFS directly, or from within GFS2 with the
1688  * inode locked, so we look to see if the glock is already locked and only
1689  * lock the glock if its not already been done. Note that its the NFS
1690  * readdirplus operation which causes this to be called (from filldir)
1691  * with the glock already held.
1692  *
1693  * Returns: errno
1694  */
1695 
1696 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1697 			struct kstat *stat)
1698 {
1699 	struct inode *inode = dentry->d_inode;
1700 	struct gfs2_inode *ip = GFS2_I(inode);
1701 	struct gfs2_holder gh;
1702 	int error;
1703 	int unlock = 0;
1704 
1705 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1706 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1707 		if (error)
1708 			return error;
1709 		unlock = 1;
1710 	}
1711 
1712 	generic_fillattr(inode, stat);
1713 	if (unlock)
1714 		gfs2_glock_dq_uninit(&gh);
1715 
1716 	return 0;
1717 }
1718 
1719 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1720 			 const void *data, size_t size, int flags)
1721 {
1722 	struct inode *inode = dentry->d_inode;
1723 	struct gfs2_inode *ip = GFS2_I(inode);
1724 	struct gfs2_holder gh;
1725 	int ret;
1726 
1727 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1728 	ret = gfs2_glock_nq(&gh);
1729 	if (ret == 0) {
1730 		ret = gfs2_rs_alloc(ip);
1731 		if (ret == 0)
1732 			ret = generic_setxattr(dentry, name, data, size, flags);
1733 		gfs2_glock_dq(&gh);
1734 	}
1735 	gfs2_holder_uninit(&gh);
1736 	return ret;
1737 }
1738 
1739 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1740 			     void *data, size_t size)
1741 {
1742 	struct inode *inode = dentry->d_inode;
1743 	struct gfs2_inode *ip = GFS2_I(inode);
1744 	struct gfs2_holder gh;
1745 	int ret;
1746 
1747 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1748 	ret = gfs2_glock_nq(&gh);
1749 	if (ret == 0) {
1750 		ret = generic_getxattr(dentry, name, data, size);
1751 		gfs2_glock_dq(&gh);
1752 	}
1753 	gfs2_holder_uninit(&gh);
1754 	return ret;
1755 }
1756 
1757 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1758 {
1759 	struct inode *inode = dentry->d_inode;
1760 	struct gfs2_inode *ip = GFS2_I(inode);
1761 	struct gfs2_holder gh;
1762 	int ret;
1763 
1764 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1765 	ret = gfs2_glock_nq(&gh);
1766 	if (ret == 0) {
1767 		ret = gfs2_rs_alloc(ip);
1768 		if (ret == 0)
1769 			ret = generic_removexattr(dentry, name);
1770 		gfs2_glock_dq(&gh);
1771 	}
1772 	gfs2_holder_uninit(&gh);
1773 	return ret;
1774 }
1775 
1776 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1777 		       u64 start, u64 len)
1778 {
1779 	struct gfs2_inode *ip = GFS2_I(inode);
1780 	struct gfs2_holder gh;
1781 	int ret;
1782 
1783 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1784 	if (ret)
1785 		return ret;
1786 
1787 	mutex_lock(&inode->i_mutex);
1788 
1789 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1790 	if (ret)
1791 		goto out;
1792 
1793 	if (gfs2_is_stuffed(ip)) {
1794 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1795 		u64 size = i_size_read(inode);
1796 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1797 			    FIEMAP_EXTENT_DATA_INLINE;
1798 		phys += sizeof(struct gfs2_dinode);
1799 		phys += start;
1800 		if (start + len > size)
1801 			len = size - start;
1802 		if (start < size)
1803 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1804 						      len, flags);
1805 		if (ret == 1)
1806 			ret = 0;
1807 	} else {
1808 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1809 					     gfs2_block_map);
1810 	}
1811 
1812 	gfs2_glock_dq_uninit(&gh);
1813 out:
1814 	mutex_unlock(&inode->i_mutex);
1815 	return ret;
1816 }
1817 
1818 const struct inode_operations gfs2_file_iops = {
1819 	.permission = gfs2_permission,
1820 	.setattr = gfs2_setattr,
1821 	.getattr = gfs2_getattr,
1822 	.setxattr = gfs2_setxattr,
1823 	.getxattr = gfs2_getxattr,
1824 	.listxattr = gfs2_listxattr,
1825 	.removexattr = gfs2_removexattr,
1826 	.fiemap = gfs2_fiemap,
1827 	.get_acl = gfs2_get_acl,
1828 };
1829 
1830 const struct inode_operations gfs2_dir_iops = {
1831 	.create = gfs2_create,
1832 	.lookup = gfs2_lookup,
1833 	.link = gfs2_link,
1834 	.unlink = gfs2_unlink,
1835 	.symlink = gfs2_symlink,
1836 	.mkdir = gfs2_mkdir,
1837 	.rmdir = gfs2_unlink,
1838 	.mknod = gfs2_mknod,
1839 	.rename = gfs2_rename,
1840 	.permission = gfs2_permission,
1841 	.setattr = gfs2_setattr,
1842 	.getattr = gfs2_getattr,
1843 	.setxattr = gfs2_setxattr,
1844 	.getxattr = gfs2_getxattr,
1845 	.listxattr = gfs2_listxattr,
1846 	.removexattr = gfs2_removexattr,
1847 	.fiemap = gfs2_fiemap,
1848 	.get_acl = gfs2_get_acl,
1849 };
1850 
1851 const struct inode_operations gfs2_symlink_iops = {
1852 	.readlink = generic_readlink,
1853 	.follow_link = gfs2_follow_link,
1854 	.put_link = gfs2_put_link,
1855 	.permission = gfs2_permission,
1856 	.setattr = gfs2_setattr,
1857 	.getattr = gfs2_getattr,
1858 	.setxattr = gfs2_setxattr,
1859 	.getxattr = gfs2_getxattr,
1860 	.listxattr = gfs2_listxattr,
1861 	.removexattr = gfs2_removexattr,
1862 	.fiemap = gfs2_fiemap,
1863 	.get_acl = gfs2_get_acl,
1864 };
1865 
1866