xref: /linux/fs/gfs2/inode.c (revision a1c3be890440a1769ed6f822376a3e3ab0d42994)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2011 Red Hat, Inc.  All rights reserved.
5  */
6 
7 #include <linux/slab.h>
8 #include <linux/spinlock.h>
9 #include <linux/completion.h>
10 #include <linux/buffer_head.h>
11 #include <linux/namei.h>
12 #include <linux/mm.h>
13 #include <linux/cred.h>
14 #include <linux/xattr.h>
15 #include <linux/posix_acl.h>
16 #include <linux/gfs2_ondisk.h>
17 #include <linux/crc32.h>
18 #include <linux/iomap.h>
19 #include <linux/security.h>
20 #include <linux/fiemap.h>
21 #include <linux/uaccess.h>
22 
23 #include "gfs2.h"
24 #include "incore.h"
25 #include "acl.h"
26 #include "bmap.h"
27 #include "dir.h"
28 #include "xattr.h"
29 #include "glock.h"
30 #include "inode.h"
31 #include "meta_io.h"
32 #include "quota.h"
33 #include "rgrp.h"
34 #include "trans.h"
35 #include "util.h"
36 #include "super.h"
37 #include "glops.h"
38 
39 static const struct inode_operations gfs2_file_iops;
40 static const struct inode_operations gfs2_dir_iops;
41 static const struct inode_operations gfs2_symlink_iops;
42 
43 static int iget_test(struct inode *inode, void *opaque)
44 {
45 	u64 no_addr = *(u64 *)opaque;
46 
47 	return GFS2_I(inode)->i_no_addr == no_addr;
48 }
49 
50 static int iget_set(struct inode *inode, void *opaque)
51 {
52 	u64 no_addr = *(u64 *)opaque;
53 
54 	GFS2_I(inode)->i_no_addr = no_addr;
55 	inode->i_ino = no_addr;
56 	return 0;
57 }
58 
59 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr)
60 {
61 	struct inode *inode;
62 
63 repeat:
64 	inode = iget5_locked(sb, no_addr, iget_test, iget_set, &no_addr);
65 	if (!inode)
66 		return inode;
67 	if (is_bad_inode(inode)) {
68 		iput(inode);
69 		goto repeat;
70 	}
71 	return inode;
72 }
73 
74 /**
75  * gfs2_set_iop - Sets inode operations
76  * @inode: The inode with correct i_mode filled in
77  *
78  * GFS2 lookup code fills in vfs inode contents based on info obtained
79  * from directory entry inside gfs2_inode_lookup().
80  */
81 
82 static void gfs2_set_iop(struct inode *inode)
83 {
84 	struct gfs2_sbd *sdp = GFS2_SB(inode);
85 	umode_t mode = inode->i_mode;
86 
87 	if (S_ISREG(mode)) {
88 		inode->i_op = &gfs2_file_iops;
89 		if (gfs2_localflocks(sdp))
90 			inode->i_fop = &gfs2_file_fops_nolock;
91 		else
92 			inode->i_fop = &gfs2_file_fops;
93 	} else if (S_ISDIR(mode)) {
94 		inode->i_op = &gfs2_dir_iops;
95 		if (gfs2_localflocks(sdp))
96 			inode->i_fop = &gfs2_dir_fops_nolock;
97 		else
98 			inode->i_fop = &gfs2_dir_fops;
99 	} else if (S_ISLNK(mode)) {
100 		inode->i_op = &gfs2_symlink_iops;
101 	} else {
102 		inode->i_op = &gfs2_file_iops;
103 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
104 	}
105 }
106 
107 /**
108  * gfs2_inode_lookup - Lookup an inode
109  * @sb: The super block
110  * @type: The type of the inode
111  * @no_addr: The inode number
112  * @no_formal_ino: The inode generation number
113  * @blktype: Requested block type (GFS2_BLKST_DINODE or GFS2_BLKST_UNLINKED;
114  *           GFS2_BLKST_FREE to indicate not to verify)
115  *
116  * If @type is DT_UNKNOWN, the inode type is fetched from disk.
117  *
118  * If @blktype is anything other than GFS2_BLKST_FREE (which is used as a
119  * placeholder because it doesn't otherwise make sense), the on-disk block type
120  * is verified to be @blktype.
121  *
122  * When @no_formal_ino is non-zero, this function will return ERR_PTR(-ESTALE)
123  * if it detects that @no_formal_ino doesn't match the actual inode generation
124  * number.  However, it doesn't always know unless @type is DT_UNKNOWN.
125  *
126  * Returns: A VFS inode, or an error
127  */
128 
129 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
130 				u64 no_addr, u64 no_formal_ino,
131 				unsigned int blktype)
132 {
133 	struct inode *inode;
134 	struct gfs2_inode *ip;
135 	struct gfs2_glock *io_gl = NULL;
136 	struct gfs2_holder i_gh;
137 	int error;
138 
139 	gfs2_holder_mark_uninitialized(&i_gh);
140 	inode = gfs2_iget(sb, no_addr);
141 	if (!inode)
142 		return ERR_PTR(-ENOMEM);
143 
144 	ip = GFS2_I(inode);
145 
146 	if (inode->i_state & I_NEW) {
147 		struct gfs2_sbd *sdp = GFS2_SB(inode);
148 
149 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
150 		if (unlikely(error))
151 			goto fail;
152 		flush_delayed_work(&ip->i_gl->gl_work);
153 
154 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
155 		if (unlikely(error))
156 			goto fail;
157 		if (blktype != GFS2_BLKST_UNLINKED)
158 			gfs2_cancel_delete_work(io_gl);
159 
160 		if (type == DT_UNKNOWN || blktype != GFS2_BLKST_FREE) {
161 			/*
162 			 * The GL_SKIP flag indicates to skip reading the inode
163 			 * block.  We read the inode with gfs2_inode_refresh
164 			 * after possibly checking the block type.
165 			 */
166 			error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE,
167 						   GL_SKIP, &i_gh);
168 			if (error)
169 				goto fail;
170 
171 			error = -ESTALE;
172 			if (no_formal_ino &&
173 			    gfs2_inode_already_deleted(ip->i_gl, no_formal_ino))
174 				goto fail;
175 
176 			if (blktype != GFS2_BLKST_FREE) {
177 				error = gfs2_check_blk_type(sdp, no_addr,
178 							    blktype);
179 				if (error)
180 					goto fail;
181 			}
182 		}
183 
184 		glock_set_object(ip->i_gl, ip);
185 		set_bit(GIF_INVALID, &ip->i_flags);
186 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
187 		if (unlikely(error))
188 			goto fail;
189 		glock_set_object(ip->i_iopen_gh.gh_gl, ip);
190 		gfs2_glock_put(io_gl);
191 		io_gl = NULL;
192 
193 		/* Lowest possible timestamp; will be overwritten in gfs2_dinode_in. */
194 		inode->i_atime.tv_sec = 1LL << (8 * sizeof(inode->i_atime.tv_sec) - 1);
195 		inode->i_atime.tv_nsec = 0;
196 
197 		if (type == DT_UNKNOWN) {
198 			/* Inode glock must be locked already */
199 			error = gfs2_inode_refresh(GFS2_I(inode));
200 			if (error)
201 				goto fail;
202 		} else {
203 			ip->i_no_formal_ino = no_formal_ino;
204 			inode->i_mode = DT2IF(type);
205 		}
206 
207 		if (gfs2_holder_initialized(&i_gh))
208 			gfs2_glock_dq_uninit(&i_gh);
209 
210 		gfs2_set_iop(inode);
211 	}
212 
213 	if (no_formal_ino && ip->i_no_formal_ino &&
214 	    no_formal_ino != ip->i_no_formal_ino) {
215 		error = -ESTALE;
216 		if (inode->i_state & I_NEW)
217 			goto fail;
218 		iput(inode);
219 		return ERR_PTR(error);
220 	}
221 
222 	if (inode->i_state & I_NEW)
223 		unlock_new_inode(inode);
224 
225 	return inode;
226 
227 fail:
228 	if (io_gl)
229 		gfs2_glock_put(io_gl);
230 	if (gfs2_holder_initialized(&i_gh))
231 		gfs2_glock_dq_uninit(&i_gh);
232 	iget_failed(inode);
233 	return ERR_PTR(error);
234 }
235 
236 /**
237  * gfs2_lookup_by_inum - look up an inode by inode number
238  * @sdp: The super block
239  * @no_addr: The inode number
240  * @no_formal_ino: The inode generation number (0 for any)
241  * @blktype: Requested block type (see gfs2_inode_lookup)
242  */
243 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
244 				  u64 no_formal_ino, unsigned int blktype)
245 {
246 	struct super_block *sb = sdp->sd_vfs;
247 	struct inode *inode;
248 	int error;
249 
250 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, no_formal_ino,
251 				  blktype);
252 	if (IS_ERR(inode))
253 		return inode;
254 
255 	if (no_formal_ino) {
256 		error = -EIO;
257 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
258 			goto fail_iput;
259 	}
260 	return inode;
261 
262 fail_iput:
263 	iput(inode);
264 	return ERR_PTR(error);
265 }
266 
267 
268 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
269 {
270 	struct qstr qstr;
271 	struct inode *inode;
272 	gfs2_str2qstr(&qstr, name);
273 	inode = gfs2_lookupi(dip, &qstr, 1);
274 	/* gfs2_lookupi has inconsistent callers: vfs
275 	 * related routines expect NULL for no entry found,
276 	 * gfs2_lookup_simple callers expect ENOENT
277 	 * and do not check for NULL.
278 	 */
279 	if (inode == NULL)
280 		return ERR_PTR(-ENOENT);
281 	else
282 		return inode;
283 }
284 
285 
286 /**
287  * gfs2_lookupi - Look up a filename in a directory and return its inode
288  * @d_gh: An initialized holder for the directory glock
289  * @name: The name of the inode to look for
290  * @is_root: If 1, ignore the caller's permissions
291  * @i_gh: An uninitialized holder for the new inode glock
292  *
293  * This can be called via the VFS filldir function when NFS is doing
294  * a readdirplus and the inode which its intending to stat isn't
295  * already in cache. In this case we must not take the directory glock
296  * again, since the readdir call will have already taken that lock.
297  *
298  * Returns: errno
299  */
300 
301 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
302 			   int is_root)
303 {
304 	struct super_block *sb = dir->i_sb;
305 	struct gfs2_inode *dip = GFS2_I(dir);
306 	struct gfs2_holder d_gh;
307 	int error = 0;
308 	struct inode *inode = NULL;
309 
310 	gfs2_holder_mark_uninitialized(&d_gh);
311 	if (!name->len || name->len > GFS2_FNAMESIZE)
312 		return ERR_PTR(-ENAMETOOLONG);
313 
314 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
315 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
316 	     dir == d_inode(sb->s_root))) {
317 		igrab(dir);
318 		return dir;
319 	}
320 
321 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
322 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
323 		if (error)
324 			return ERR_PTR(error);
325 	}
326 
327 	if (!is_root) {
328 		error = gfs2_permission(&init_user_ns, dir, MAY_EXEC);
329 		if (error)
330 			goto out;
331 	}
332 
333 	inode = gfs2_dir_search(dir, name, false);
334 	if (IS_ERR(inode))
335 		error = PTR_ERR(inode);
336 out:
337 	if (gfs2_holder_initialized(&d_gh))
338 		gfs2_glock_dq_uninit(&d_gh);
339 	if (error == -ENOENT)
340 		return NULL;
341 	return inode ? inode : ERR_PTR(error);
342 }
343 
344 /**
345  * create_ok - OK to create a new on-disk inode here?
346  * @dip:  Directory in which dinode is to be created
347  * @name:  Name of new dinode
348  * @mode:
349  *
350  * Returns: errno
351  */
352 
353 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
354 		     umode_t mode)
355 {
356 	int error;
357 
358 	error = gfs2_permission(&init_user_ns, &dip->i_inode,
359 				MAY_WRITE | MAY_EXEC);
360 	if (error)
361 		return error;
362 
363 	/*  Don't create entries in an unlinked directory  */
364 	if (!dip->i_inode.i_nlink)
365 		return -ENOENT;
366 
367 	if (dip->i_entries == (u32)-1)
368 		return -EFBIG;
369 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
370 		return -EMLINK;
371 
372 	return 0;
373 }
374 
375 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
376 			       struct inode *inode)
377 {
378 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
379 	    (dip->i_inode.i_mode & S_ISUID) &&
380 	    !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
381 		if (S_ISDIR(inode->i_mode))
382 			inode->i_mode |= S_ISUID;
383 		else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
384 			inode->i_mode &= ~07111;
385 		inode->i_uid = dip->i_inode.i_uid;
386 	} else
387 		inode->i_uid = current_fsuid();
388 
389 	if (dip->i_inode.i_mode & S_ISGID) {
390 		if (S_ISDIR(inode->i_mode))
391 			inode->i_mode |= S_ISGID;
392 		inode->i_gid = dip->i_inode.i_gid;
393 	} else
394 		inode->i_gid = current_fsgid();
395 }
396 
397 static int alloc_dinode(struct gfs2_inode *ip, u32 flags, unsigned *dblocks)
398 {
399 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
400 	struct gfs2_alloc_parms ap = { .target = *dblocks, .aflags = flags, };
401 	int error;
402 
403 	error = gfs2_quota_lock_check(ip, &ap);
404 	if (error)
405 		goto out;
406 
407 	error = gfs2_inplace_reserve(ip, &ap);
408 	if (error)
409 		goto out_quota;
410 
411 	error = gfs2_trans_begin(sdp, (*dblocks * RES_RG_BIT) + RES_STATFS + RES_QUOTA, 0);
412 	if (error)
413 		goto out_ipreserv;
414 
415 	error = gfs2_alloc_blocks(ip, &ip->i_no_addr, dblocks, 1, &ip->i_generation);
416 	ip->i_no_formal_ino = ip->i_generation;
417 	ip->i_inode.i_ino = ip->i_no_addr;
418 	ip->i_goal = ip->i_no_addr;
419 
420 	gfs2_trans_end(sdp);
421 
422 out_ipreserv:
423 	gfs2_inplace_release(ip);
424 out_quota:
425 	gfs2_quota_unlock(ip);
426 out:
427 	return error;
428 }
429 
430 static void gfs2_init_dir(struct buffer_head *dibh,
431 			  const struct gfs2_inode *parent)
432 {
433 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
434 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
435 
436 	gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
437 	dent->de_inum = di->di_num; /* already GFS2 endian */
438 	dent->de_type = cpu_to_be16(DT_DIR);
439 
440 	dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
441 	gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
442 	gfs2_inum_out(parent, dent);
443 	dent->de_type = cpu_to_be16(DT_DIR);
444 
445 }
446 
447 /**
448  * gfs2_init_xattr - Initialise an xattr block for a new inode
449  * @ip: The inode in question
450  *
451  * This sets up an empty xattr block for a new inode, ready to
452  * take any ACLs, LSM xattrs, etc.
453  */
454 
455 static void gfs2_init_xattr(struct gfs2_inode *ip)
456 {
457 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
458 	struct buffer_head *bh;
459 	struct gfs2_ea_header *ea;
460 
461 	bh = gfs2_meta_new(ip->i_gl, ip->i_eattr);
462 	gfs2_trans_add_meta(ip->i_gl, bh);
463 	gfs2_metatype_set(bh, GFS2_METATYPE_EA, GFS2_FORMAT_EA);
464 	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
465 
466 	ea = GFS2_EA_BH2FIRST(bh);
467 	ea->ea_rec_len = cpu_to_be32(sdp->sd_jbsize);
468 	ea->ea_type = GFS2_EATYPE_UNUSED;
469 	ea->ea_flags = GFS2_EAFLAG_LAST;
470 
471 	brelse(bh);
472 }
473 
474 /**
475  * init_dinode - Fill in a new dinode structure
476  * @dip: The directory this inode is being created in
477  * @ip: The inode
478  * @symname: The symlink destination (if a symlink)
479  * @bhp: The buffer head (returned to caller)
480  *
481  */
482 
483 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
484 			const char *symname)
485 {
486 	struct gfs2_dinode *di;
487 	struct buffer_head *dibh;
488 
489 	dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
490 	gfs2_trans_add_meta(ip->i_gl, dibh);
491 	di = (struct gfs2_dinode *)dibh->b_data;
492 	gfs2_dinode_out(ip, di);
493 
494 	di->di_major = cpu_to_be32(imajor(&ip->i_inode));
495 	di->di_minor = cpu_to_be32(iminor(&ip->i_inode));
496 	di->__pad1 = 0;
497 	di->__pad2 = 0;
498 	di->__pad3 = 0;
499 	memset(&di->__pad4, 0, sizeof(di->__pad4));
500 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
501 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
502 
503 	switch(ip->i_inode.i_mode & S_IFMT) {
504 	case S_IFDIR:
505 		gfs2_init_dir(dibh, dip);
506 		break;
507 	case S_IFLNK:
508 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
509 		break;
510 	}
511 
512 	set_buffer_uptodate(dibh);
513 	brelse(dibh);
514 }
515 
516 /**
517  * gfs2_trans_da_blocks - Calculate number of blocks to link inode
518  * @dip: The directory we are linking into
519  * @da: The dir add information
520  * @nr_inodes: The number of inodes involved
521  *
522  * This calculate the number of blocks we need to reserve in a
523  * transaction to link @nr_inodes into a directory. In most cases
524  * @nr_inodes will be 2 (the directory plus the inode being linked in)
525  * but in case of rename, 4 may be required.
526  *
527  * Returns: Number of blocks
528  */
529 
530 static unsigned gfs2_trans_da_blks(const struct gfs2_inode *dip,
531 				   const struct gfs2_diradd *da,
532 				   unsigned nr_inodes)
533 {
534 	return da->nr_blocks + gfs2_rg_blocks(dip, da->nr_blocks) +
535 	       (nr_inodes * RES_DINODE) + RES_QUOTA + RES_STATFS;
536 }
537 
538 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
539 		       struct gfs2_inode *ip, struct gfs2_diradd *da)
540 {
541 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
542 	struct gfs2_alloc_parms ap = { .target = da->nr_blocks, };
543 	int error;
544 
545 	if (da->nr_blocks) {
546 		error = gfs2_quota_lock_check(dip, &ap);
547 		if (error)
548 			goto fail_quota_locks;
549 
550 		error = gfs2_inplace_reserve(dip, &ap);
551 		if (error)
552 			goto fail_quota_locks;
553 
554 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, da, 2), 0);
555 		if (error)
556 			goto fail_ipreserv;
557 	} else {
558 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
559 		if (error)
560 			goto fail_quota_locks;
561 	}
562 
563 	error = gfs2_dir_add(&dip->i_inode, name, ip, da);
564 
565 	gfs2_trans_end(sdp);
566 fail_ipreserv:
567 	gfs2_inplace_release(dip);
568 fail_quota_locks:
569 	gfs2_quota_unlock(dip);
570 	return error;
571 }
572 
573 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
574 		    void *fs_info)
575 {
576 	const struct xattr *xattr;
577 	int err = 0;
578 
579 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
580 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
581 				       xattr->value_len, 0,
582 				       GFS2_EATYPE_SECURITY);
583 		if (err < 0)
584 			break;
585 	}
586 	return err;
587 }
588 
589 /**
590  * gfs2_create_inode - Create a new inode
591  * @dir: The parent directory
592  * @dentry: The new dentry
593  * @file: If non-NULL, the file which is being opened
594  * @mode: The permissions on the new inode
595  * @dev: For device nodes, this is the device number
596  * @symname: For symlinks, this is the link destination
597  * @size: The initial size of the inode (ignored for directories)
598  *
599  * Returns: 0 on success, or error code
600  */
601 
602 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
603 			     struct file *file,
604 			     umode_t mode, dev_t dev, const char *symname,
605 			     unsigned int size, int excl)
606 {
607 	const struct qstr *name = &dentry->d_name;
608 	struct posix_acl *default_acl, *acl;
609 	struct gfs2_holder ghs[2];
610 	struct inode *inode = NULL;
611 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
612 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
613 	struct gfs2_glock *io_gl;
614 	int error, free_vfs_inode = 1;
615 	u32 aflags = 0;
616 	unsigned blocks = 1;
617 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
618 
619 	if (!name->len || name->len > GFS2_FNAMESIZE)
620 		return -ENAMETOOLONG;
621 
622 	error = gfs2_qa_get(dip);
623 	if (error)
624 		return error;
625 
626 	error = gfs2_rindex_update(sdp);
627 	if (error)
628 		goto fail;
629 
630 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
631 	if (error)
632 		goto fail;
633 	gfs2_holder_mark_uninitialized(ghs + 1);
634 
635 	error = create_ok(dip, name, mode);
636 	if (error)
637 		goto fail_gunlock;
638 
639 	inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
640 	error = PTR_ERR(inode);
641 	if (!IS_ERR(inode)) {
642 		if (S_ISDIR(inode->i_mode)) {
643 			iput(inode);
644 			inode = ERR_PTR(-EISDIR);
645 			goto fail_gunlock;
646 		}
647 		d_instantiate(dentry, inode);
648 		error = 0;
649 		if (file) {
650 			if (S_ISREG(inode->i_mode))
651 				error = finish_open(file, dentry, gfs2_open_common);
652 			else
653 				error = finish_no_open(file, NULL);
654 		}
655 		gfs2_glock_dq_uninit(ghs);
656 		goto fail;
657 	} else if (error != -ENOENT) {
658 		goto fail_gunlock;
659 	}
660 
661 	error = gfs2_diradd_alloc_required(dir, name, &da);
662 	if (error < 0)
663 		goto fail_gunlock;
664 
665 	inode = new_inode(sdp->sd_vfs);
666 	error = -ENOMEM;
667 	if (!inode)
668 		goto fail_gunlock;
669 
670 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
671 	if (error)
672 		goto fail_gunlock;
673 
674 	ip = GFS2_I(inode);
675 	error = gfs2_qa_get(ip);
676 	if (error)
677 		goto fail_free_acls;
678 
679 	inode->i_mode = mode;
680 	set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
681 	inode->i_rdev = dev;
682 	inode->i_size = size;
683 	inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
684 	munge_mode_uid_gid(dip, inode);
685 	check_and_update_goal(dip);
686 	ip->i_goal = dip->i_goal;
687 	ip->i_diskflags = 0;
688 	ip->i_eattr = 0;
689 	ip->i_height = 0;
690 	ip->i_depth = 0;
691 	ip->i_entries = 0;
692 	ip->i_no_addr = 0; /* Temporarily zero until real addr is assigned */
693 
694 	switch(mode & S_IFMT) {
695 	case S_IFREG:
696 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
697 		    gfs2_tune_get(sdp, gt_new_files_jdata))
698 			ip->i_diskflags |= GFS2_DIF_JDATA;
699 		gfs2_set_aops(inode);
700 		break;
701 	case S_IFDIR:
702 		ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
703 		ip->i_diskflags |= GFS2_DIF_JDATA;
704 		ip->i_entries = 2;
705 		break;
706 	}
707 
708 	/* Force SYSTEM flag on all files and subdirs of a SYSTEM directory */
709 	if (dip->i_diskflags & GFS2_DIF_SYSTEM)
710 		ip->i_diskflags |= GFS2_DIF_SYSTEM;
711 
712 	gfs2_set_inode_flags(inode);
713 
714 	if ((GFS2_I(d_inode(sdp->sd_root_dir)) == dip) ||
715 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
716 		aflags |= GFS2_AF_ORLOV;
717 
718 	if (default_acl || acl)
719 		blocks++;
720 
721 	error = alloc_dinode(ip, aflags, &blocks);
722 	if (error)
723 		goto fail_free_inode;
724 
725 	gfs2_set_inode_blocks(inode, blocks);
726 
727 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
728 	if (error)
729 		goto fail_free_inode;
730 	flush_delayed_work(&ip->i_gl->gl_work);
731 	glock_set_object(ip->i_gl, ip);
732 
733 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
734 	if (error)
735 		goto fail_free_inode;
736 	gfs2_cancel_delete_work(io_gl);
737 	glock_set_object(io_gl, ip);
738 
739 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
740 	if (error)
741 		goto fail_gunlock2;
742 
743 	error = gfs2_trans_begin(sdp, blocks, 0);
744 	if (error)
745 		goto fail_gunlock2;
746 
747 	if (blocks > 1) {
748 		ip->i_eattr = ip->i_no_addr + 1;
749 		gfs2_init_xattr(ip);
750 	}
751 	init_dinode(dip, ip, symname);
752 	gfs2_trans_end(sdp);
753 
754 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
755 	if (error)
756 		goto fail_gunlock2;
757 
758 	gfs2_set_iop(inode);
759 	insert_inode_hash(inode);
760 
761 	free_vfs_inode = 0; /* After this point, the inode is no longer
762 			       considered free. Any failures need to undo
763 			       the gfs2 structures. */
764 	if (default_acl) {
765 		error = __gfs2_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
766 		if (error)
767 			goto fail_gunlock3;
768 		posix_acl_release(default_acl);
769 		default_acl = NULL;
770 	}
771 	if (acl) {
772 		error = __gfs2_set_acl(inode, acl, ACL_TYPE_ACCESS);
773 		if (error)
774 			goto fail_gunlock3;
775 		posix_acl_release(acl);
776 		acl = NULL;
777 	}
778 
779 	error = security_inode_init_security(&ip->i_inode, &dip->i_inode, name,
780 					     &gfs2_initxattrs, NULL);
781 	if (error)
782 		goto fail_gunlock3;
783 
784 	error = link_dinode(dip, name, ip, &da);
785 	if (error)
786 		goto fail_gunlock3;
787 
788 	mark_inode_dirty(inode);
789 	d_instantiate(dentry, inode);
790 	/* After instantiate, errors should result in evict which will destroy
791 	 * both inode and iopen glocks properly. */
792 	if (file) {
793 		file->f_mode |= FMODE_CREATED;
794 		error = finish_open(file, dentry, gfs2_open_common);
795 	}
796 	gfs2_glock_dq_uninit(ghs);
797 	gfs2_qa_put(ip);
798 	gfs2_glock_dq_uninit(ghs + 1);
799 	gfs2_glock_put(io_gl);
800 	gfs2_qa_put(dip);
801 	return error;
802 
803 fail_gunlock3:
804 	glock_clear_object(io_gl, ip);
805 	gfs2_glock_dq_uninit(&ip->i_iopen_gh);
806 fail_gunlock2:
807 	glock_clear_object(io_gl, ip);
808 	gfs2_glock_put(io_gl);
809 fail_free_inode:
810 	if (ip->i_gl) {
811 		glock_clear_object(ip->i_gl, ip);
812 		if (free_vfs_inode) /* else evict will do the put for us */
813 			gfs2_glock_put(ip->i_gl);
814 	}
815 	gfs2_rs_delete(ip, NULL);
816 	gfs2_qa_put(ip);
817 fail_free_acls:
818 	posix_acl_release(default_acl);
819 	posix_acl_release(acl);
820 fail_gunlock:
821 	gfs2_dir_no_add(&da);
822 	gfs2_glock_dq_uninit(ghs);
823 	if (!IS_ERR_OR_NULL(inode)) {
824 		clear_nlink(inode);
825 		if (!free_vfs_inode)
826 			mark_inode_dirty(inode);
827 		set_bit(free_vfs_inode ? GIF_FREE_VFS_INODE : GIF_ALLOC_FAILED,
828 			&GFS2_I(inode)->i_flags);
829 		iput(inode);
830 	}
831 	if (gfs2_holder_initialized(ghs + 1))
832 		gfs2_glock_dq_uninit(ghs + 1);
833 fail:
834 	gfs2_qa_put(dip);
835 	return error;
836 }
837 
838 /**
839  * gfs2_create - Create a file
840  * @dir: The directory in which to create the file
841  * @dentry: The dentry of the new file
842  * @mode: The mode of the new file
843  *
844  * Returns: errno
845  */
846 
847 static int gfs2_create(struct user_namespace *mnt_userns, struct inode *dir,
848 		       struct dentry *dentry, umode_t mode, bool excl)
849 {
850 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl);
851 }
852 
853 /**
854  * __gfs2_lookup - Look up a filename in a directory and return its inode
855  * @dir: The directory inode
856  * @dentry: The dentry of the new inode
857  * @file: File to be opened
858  *
859  *
860  * Returns: errno
861  */
862 
863 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
864 				    struct file *file)
865 {
866 	struct inode *inode;
867 	struct dentry *d;
868 	struct gfs2_holder gh;
869 	struct gfs2_glock *gl;
870 	int error;
871 
872 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
873 	if (inode == NULL) {
874 		d_add(dentry, NULL);
875 		return NULL;
876 	}
877 	if (IS_ERR(inode))
878 		return ERR_CAST(inode);
879 
880 	gl = GFS2_I(inode)->i_gl;
881 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
882 	if (error) {
883 		iput(inode);
884 		return ERR_PTR(error);
885 	}
886 
887 	d = d_splice_alias(inode, dentry);
888 	if (IS_ERR(d)) {
889 		gfs2_glock_dq_uninit(&gh);
890 		return d;
891 	}
892 	if (file && S_ISREG(inode->i_mode))
893 		error = finish_open(file, dentry, gfs2_open_common);
894 
895 	gfs2_glock_dq_uninit(&gh);
896 	if (error) {
897 		dput(d);
898 		return ERR_PTR(error);
899 	}
900 	return d;
901 }
902 
903 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
904 				  unsigned flags)
905 {
906 	return __gfs2_lookup(dir, dentry, NULL);
907 }
908 
909 /**
910  * gfs2_link - Link to a file
911  * @old_dentry: The inode to link
912  * @dir: Add link to this directory
913  * @dentry: The name of the link
914  *
915  * Link the inode in "old_dentry" into the directory "dir" with the
916  * name in "dentry".
917  *
918  * Returns: errno
919  */
920 
921 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
922 		     struct dentry *dentry)
923 {
924 	struct gfs2_inode *dip = GFS2_I(dir);
925 	struct gfs2_sbd *sdp = GFS2_SB(dir);
926 	struct inode *inode = d_inode(old_dentry);
927 	struct gfs2_inode *ip = GFS2_I(inode);
928 	struct gfs2_holder ghs[2];
929 	struct buffer_head *dibh;
930 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
931 	int error;
932 
933 	if (S_ISDIR(inode->i_mode))
934 		return -EPERM;
935 
936 	error = gfs2_qa_get(dip);
937 	if (error)
938 		return error;
939 
940 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
941 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
942 
943 	error = gfs2_glock_nq(ghs); /* parent */
944 	if (error)
945 		goto out_parent;
946 
947 	error = gfs2_glock_nq(ghs + 1); /* child */
948 	if (error)
949 		goto out_child;
950 
951 	error = -ENOENT;
952 	if (inode->i_nlink == 0)
953 		goto out_gunlock;
954 
955 	error = gfs2_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
956 	if (error)
957 		goto out_gunlock;
958 
959 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
960 	switch (error) {
961 	case -ENOENT:
962 		break;
963 	case 0:
964 		error = -EEXIST;
965 	default:
966 		goto out_gunlock;
967 	}
968 
969 	error = -EINVAL;
970 	if (!dip->i_inode.i_nlink)
971 		goto out_gunlock;
972 	error = -EFBIG;
973 	if (dip->i_entries == (u32)-1)
974 		goto out_gunlock;
975 	error = -EPERM;
976 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
977 		goto out_gunlock;
978 	error = -EINVAL;
979 	if (!ip->i_inode.i_nlink)
980 		goto out_gunlock;
981 	error = -EMLINK;
982 	if (ip->i_inode.i_nlink == (u32)-1)
983 		goto out_gunlock;
984 
985 	error = gfs2_diradd_alloc_required(dir, &dentry->d_name, &da);
986 	if (error < 0)
987 		goto out_gunlock;
988 
989 	if (da.nr_blocks) {
990 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
991 		error = gfs2_quota_lock_check(dip, &ap);
992 		if (error)
993 			goto out_gunlock;
994 
995 		error = gfs2_inplace_reserve(dip, &ap);
996 		if (error)
997 			goto out_gunlock_q;
998 
999 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, &da, 2), 0);
1000 		if (error)
1001 			goto out_ipres;
1002 	} else {
1003 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
1004 		if (error)
1005 			goto out_ipres;
1006 	}
1007 
1008 	error = gfs2_meta_inode_buffer(ip, &dibh);
1009 	if (error)
1010 		goto out_end_trans;
1011 
1012 	error = gfs2_dir_add(dir, &dentry->d_name, ip, &da);
1013 	if (error)
1014 		goto out_brelse;
1015 
1016 	gfs2_trans_add_meta(ip->i_gl, dibh);
1017 	inc_nlink(&ip->i_inode);
1018 	ip->i_inode.i_ctime = current_time(&ip->i_inode);
1019 	ihold(inode);
1020 	d_instantiate(dentry, inode);
1021 	mark_inode_dirty(inode);
1022 
1023 out_brelse:
1024 	brelse(dibh);
1025 out_end_trans:
1026 	gfs2_trans_end(sdp);
1027 out_ipres:
1028 	if (da.nr_blocks)
1029 		gfs2_inplace_release(dip);
1030 out_gunlock_q:
1031 	if (da.nr_blocks)
1032 		gfs2_quota_unlock(dip);
1033 out_gunlock:
1034 	gfs2_dir_no_add(&da);
1035 	gfs2_glock_dq(ghs + 1);
1036 out_child:
1037 	gfs2_glock_dq(ghs);
1038 out_parent:
1039 	gfs2_qa_put(dip);
1040 	gfs2_holder_uninit(ghs);
1041 	gfs2_holder_uninit(ghs + 1);
1042 	return error;
1043 }
1044 
1045 /*
1046  * gfs2_unlink_ok - check to see that a inode is still in a directory
1047  * @dip: the directory
1048  * @name: the name of the file
1049  * @ip: the inode
1050  *
1051  * Assumes that the lock on (at least) @dip is held.
1052  *
1053  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1054  */
1055 
1056 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
1057 			  const struct gfs2_inode *ip)
1058 {
1059 	int error;
1060 
1061 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
1062 		return -EPERM;
1063 
1064 	if ((dip->i_inode.i_mode & S_ISVTX) &&
1065 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
1066 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
1067 		return -EPERM;
1068 
1069 	if (IS_APPEND(&dip->i_inode))
1070 		return -EPERM;
1071 
1072 	error = gfs2_permission(&init_user_ns, &dip->i_inode,
1073 				MAY_WRITE | MAY_EXEC);
1074 	if (error)
1075 		return error;
1076 
1077 	return gfs2_dir_check(&dip->i_inode, name, ip);
1078 }
1079 
1080 /**
1081  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1082  * @dip: The parent directory
1083  * @name: The name of the entry in the parent directory
1084  * @inode: The inode to be removed
1085  *
1086  * Called with all the locks and in a transaction. This will only be
1087  * called for a directory after it has been checked to ensure it is empty.
1088  *
1089  * Returns: 0 on success, or an error
1090  */
1091 
1092 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1093 			     const struct dentry *dentry)
1094 {
1095 	struct inode *inode = d_inode(dentry);
1096 	struct gfs2_inode *ip = GFS2_I(inode);
1097 	int error;
1098 
1099 	error = gfs2_dir_del(dip, dentry);
1100 	if (error)
1101 		return error;
1102 
1103 	ip->i_entries = 0;
1104 	inode->i_ctime = current_time(inode);
1105 	if (S_ISDIR(inode->i_mode))
1106 		clear_nlink(inode);
1107 	else
1108 		drop_nlink(inode);
1109 	mark_inode_dirty(inode);
1110 	if (inode->i_nlink == 0)
1111 		gfs2_unlink_di(inode);
1112 	return 0;
1113 }
1114 
1115 
1116 /**
1117  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1118  * @dir: The inode of the directory containing the inode to unlink
1119  * @dentry: The file itself
1120  *
1121  * This routine uses the type of the inode as a flag to figure out
1122  * whether this is an unlink or an rmdir.
1123  *
1124  * Returns: errno
1125  */
1126 
1127 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1128 {
1129 	struct gfs2_inode *dip = GFS2_I(dir);
1130 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1131 	struct inode *inode = d_inode(dentry);
1132 	struct gfs2_inode *ip = GFS2_I(inode);
1133 	struct gfs2_holder ghs[3];
1134 	struct gfs2_rgrpd *rgd;
1135 	int error;
1136 
1137 	error = gfs2_rindex_update(sdp);
1138 	if (error)
1139 		return error;
1140 
1141 	error = -EROFS;
1142 
1143 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1144 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1145 
1146 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1147 	if (!rgd)
1148 		goto out_inodes;
1149 
1150 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, LM_FLAG_NODE_SCOPE, ghs + 2);
1151 
1152 
1153 	error = gfs2_glock_nq(ghs); /* parent */
1154 	if (error)
1155 		goto out_parent;
1156 
1157 	error = gfs2_glock_nq(ghs + 1); /* child */
1158 	if (error)
1159 		goto out_child;
1160 
1161 	error = -ENOENT;
1162 	if (inode->i_nlink == 0)
1163 		goto out_rgrp;
1164 
1165 	if (S_ISDIR(inode->i_mode)) {
1166 		error = -ENOTEMPTY;
1167 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1168 			goto out_rgrp;
1169 	}
1170 
1171 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1172 	if (error)
1173 		goto out_rgrp;
1174 
1175 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1176 	if (error)
1177 		goto out_gunlock;
1178 
1179 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1180 	if (error)
1181 		goto out_gunlock;
1182 
1183 	error = gfs2_unlink_inode(dip, dentry);
1184 	gfs2_trans_end(sdp);
1185 
1186 out_gunlock:
1187 	gfs2_glock_dq(ghs + 2);
1188 out_rgrp:
1189 	gfs2_glock_dq(ghs + 1);
1190 out_child:
1191 	gfs2_glock_dq(ghs);
1192 out_parent:
1193 	gfs2_holder_uninit(ghs + 2);
1194 out_inodes:
1195 	gfs2_holder_uninit(ghs + 1);
1196 	gfs2_holder_uninit(ghs);
1197 	return error;
1198 }
1199 
1200 /**
1201  * gfs2_symlink - Create a symlink
1202  * @dir: The directory to create the symlink in
1203  * @dentry: The dentry to put the symlink in
1204  * @symname: The thing which the link points to
1205  *
1206  * Returns: errno
1207  */
1208 
1209 static int gfs2_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1210 			struct dentry *dentry, const char *symname)
1211 {
1212 	unsigned int size;
1213 
1214 	size = strlen(symname);
1215 	if (size >= gfs2_max_stuffed_size(GFS2_I(dir)))
1216 		return -ENAMETOOLONG;
1217 
1218 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1219 }
1220 
1221 /**
1222  * gfs2_mkdir - Make a directory
1223  * @dir: The parent directory of the new one
1224  * @dentry: The dentry of the new directory
1225  * @mode: The mode of the new directory
1226  *
1227  * Returns: errno
1228  */
1229 
1230 static int gfs2_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1231 		      struct dentry *dentry, umode_t mode)
1232 {
1233 	unsigned dsize = gfs2_max_stuffed_size(GFS2_I(dir));
1234 	return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0);
1235 }
1236 
1237 /**
1238  * gfs2_mknod - Make a special file
1239  * @dir: The directory in which the special file will reside
1240  * @dentry: The dentry of the special file
1241  * @mode: The mode of the special file
1242  * @dev: The device specification of the special file
1243  *
1244  */
1245 
1246 static int gfs2_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1247 		      struct dentry *dentry, umode_t mode, dev_t dev)
1248 {
1249 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0);
1250 }
1251 
1252 /**
1253  * gfs2_atomic_open - Atomically open a file
1254  * @dir: The directory
1255  * @dentry: The proposed new entry
1256  * @file: The proposed new struct file
1257  * @flags: open flags
1258  * @mode: File mode
1259  *
1260  * Returns: error code or 0 for success
1261  */
1262 
1263 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1264 			    struct file *file, unsigned flags,
1265 			    umode_t mode)
1266 {
1267 	struct dentry *d;
1268 	bool excl = !!(flags & O_EXCL);
1269 
1270 	if (!d_in_lookup(dentry))
1271 		goto skip_lookup;
1272 
1273 	d = __gfs2_lookup(dir, dentry, file);
1274 	if (IS_ERR(d))
1275 		return PTR_ERR(d);
1276 	if (d != NULL)
1277 		dentry = d;
1278 	if (d_really_is_positive(dentry)) {
1279 		if (!(file->f_mode & FMODE_OPENED))
1280 			return finish_no_open(file, d);
1281 		dput(d);
1282 		return excl && (flags & O_CREAT) ? -EEXIST : 0;
1283 	}
1284 
1285 	BUG_ON(d != NULL);
1286 
1287 skip_lookup:
1288 	if (!(flags & O_CREAT))
1289 		return -ENOENT;
1290 
1291 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl);
1292 }
1293 
1294 /*
1295  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1296  * @this: move this
1297  * @to: to here
1298  *
1299  * Follow @to back to the root and make sure we don't encounter @this
1300  * Assumes we already hold the rename lock.
1301  *
1302  * Returns: errno
1303  */
1304 
1305 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1306 {
1307 	struct inode *dir = &to->i_inode;
1308 	struct super_block *sb = dir->i_sb;
1309 	struct inode *tmp;
1310 	int error = 0;
1311 
1312 	igrab(dir);
1313 
1314 	for (;;) {
1315 		if (dir == &this->i_inode) {
1316 			error = -EINVAL;
1317 			break;
1318 		}
1319 		if (dir == d_inode(sb->s_root)) {
1320 			error = 0;
1321 			break;
1322 		}
1323 
1324 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1325 		if (!tmp) {
1326 			error = -ENOENT;
1327 			break;
1328 		}
1329 		if (IS_ERR(tmp)) {
1330 			error = PTR_ERR(tmp);
1331 			break;
1332 		}
1333 
1334 		iput(dir);
1335 		dir = tmp;
1336 	}
1337 
1338 	iput(dir);
1339 
1340 	return error;
1341 }
1342 
1343 /**
1344  * update_moved_ino - Update an inode that's being moved
1345  * @ip: The inode being moved
1346  * @ndip: The parent directory of the new filename
1347  * @dir_rename: True of ip is a directory
1348  *
1349  * Returns: errno
1350  */
1351 
1352 static int update_moved_ino(struct gfs2_inode *ip, struct gfs2_inode *ndip,
1353 			    int dir_rename)
1354 {
1355 	if (dir_rename)
1356 		return gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1357 
1358 	ip->i_inode.i_ctime = current_time(&ip->i_inode);
1359 	mark_inode_dirty_sync(&ip->i_inode);
1360 	return 0;
1361 }
1362 
1363 
1364 /**
1365  * gfs2_rename - Rename a file
1366  * @odir: Parent directory of old file name
1367  * @odentry: The old dentry of the file
1368  * @ndir: Parent directory of new file name
1369  * @ndentry: The new dentry of the file
1370  *
1371  * Returns: errno
1372  */
1373 
1374 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1375 		       struct inode *ndir, struct dentry *ndentry)
1376 {
1377 	struct gfs2_inode *odip = GFS2_I(odir);
1378 	struct gfs2_inode *ndip = GFS2_I(ndir);
1379 	struct gfs2_inode *ip = GFS2_I(d_inode(odentry));
1380 	struct gfs2_inode *nip = NULL;
1381 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1382 	struct gfs2_holder ghs[4], r_gh, rd_gh;
1383 	struct gfs2_rgrpd *nrgd;
1384 	unsigned int num_gh;
1385 	int dir_rename = 0;
1386 	struct gfs2_diradd da = { .nr_blocks = 0, .save_loc = 0, };
1387 	unsigned int x;
1388 	int error;
1389 
1390 	gfs2_holder_mark_uninitialized(&r_gh);
1391 	gfs2_holder_mark_uninitialized(&rd_gh);
1392 	if (d_really_is_positive(ndentry)) {
1393 		nip = GFS2_I(d_inode(ndentry));
1394 		if (ip == nip)
1395 			return 0;
1396 	}
1397 
1398 	error = gfs2_rindex_update(sdp);
1399 	if (error)
1400 		return error;
1401 
1402 	error = gfs2_qa_get(ndip);
1403 	if (error)
1404 		return error;
1405 
1406 	if (odip != ndip) {
1407 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1408 					   0, &r_gh);
1409 		if (error)
1410 			goto out;
1411 
1412 		if (S_ISDIR(ip->i_inode.i_mode)) {
1413 			dir_rename = 1;
1414 			/* don't move a directory into its subdir */
1415 			error = gfs2_ok_to_move(ip, ndip);
1416 			if (error)
1417 				goto out_gunlock_r;
1418 		}
1419 	}
1420 
1421 	num_gh = 1;
1422 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs);
1423 	if (odip != ndip) {
1424 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE,GL_ASYNC,
1425 				 ghs + num_gh);
1426 		num_gh++;
1427 	}
1428 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1429 	num_gh++;
1430 
1431 	if (nip) {
1432 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC,
1433 				 ghs + num_gh);
1434 		num_gh++;
1435 	}
1436 
1437 	for (x = 0; x < num_gh; x++) {
1438 		error = gfs2_glock_nq(ghs + x);
1439 		if (error)
1440 			goto out_gunlock;
1441 	}
1442 	error = gfs2_glock_async_wait(num_gh, ghs);
1443 	if (error)
1444 		goto out_gunlock;
1445 
1446 	if (nip) {
1447 		/* Grab the resource group glock for unlink flag twiddling.
1448 		 * This is the case where the target dinode already exists
1449 		 * so we unlink before doing the rename.
1450 		 */
1451 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1452 		if (!nrgd) {
1453 			error = -ENOENT;
1454 			goto out_gunlock;
1455 		}
1456 		error = gfs2_glock_nq_init(nrgd->rd_gl, LM_ST_EXCLUSIVE,
1457 					   LM_FLAG_NODE_SCOPE, &rd_gh);
1458 		if (error)
1459 			goto out_gunlock;
1460 	}
1461 
1462 	error = -ENOENT;
1463 	if (ip->i_inode.i_nlink == 0)
1464 		goto out_gunlock;
1465 
1466 	/* Check out the old directory */
1467 
1468 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1469 	if (error)
1470 		goto out_gunlock;
1471 
1472 	/* Check out the new directory */
1473 
1474 	if (nip) {
1475 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1476 		if (error)
1477 			goto out_gunlock;
1478 
1479 		if (nip->i_inode.i_nlink == 0) {
1480 			error = -EAGAIN;
1481 			goto out_gunlock;
1482 		}
1483 
1484 		if (S_ISDIR(nip->i_inode.i_mode)) {
1485 			if (nip->i_entries < 2) {
1486 				gfs2_consist_inode(nip);
1487 				error = -EIO;
1488 				goto out_gunlock;
1489 			}
1490 			if (nip->i_entries > 2) {
1491 				error = -ENOTEMPTY;
1492 				goto out_gunlock;
1493 			}
1494 		}
1495 	} else {
1496 		error = gfs2_permission(&init_user_ns, ndir,
1497 					MAY_WRITE | MAY_EXEC);
1498 		if (error)
1499 			goto out_gunlock;
1500 
1501 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1502 		switch (error) {
1503 		case -ENOENT:
1504 			error = 0;
1505 			break;
1506 		case 0:
1507 			error = -EEXIST;
1508 		default:
1509 			goto out_gunlock;
1510 		}
1511 
1512 		if (odip != ndip) {
1513 			if (!ndip->i_inode.i_nlink) {
1514 				error = -ENOENT;
1515 				goto out_gunlock;
1516 			}
1517 			if (ndip->i_entries == (u32)-1) {
1518 				error = -EFBIG;
1519 				goto out_gunlock;
1520 			}
1521 			if (S_ISDIR(ip->i_inode.i_mode) &&
1522 			    ndip->i_inode.i_nlink == (u32)-1) {
1523 				error = -EMLINK;
1524 				goto out_gunlock;
1525 			}
1526 		}
1527 	}
1528 
1529 	/* Check out the dir to be renamed */
1530 
1531 	if (dir_rename) {
1532 		error = gfs2_permission(&init_user_ns, d_inode(odentry),
1533 					MAY_WRITE);
1534 		if (error)
1535 			goto out_gunlock;
1536 	}
1537 
1538 	if (nip == NULL) {
1539 		error = gfs2_diradd_alloc_required(ndir, &ndentry->d_name, &da);
1540 		if (error)
1541 			goto out_gunlock;
1542 	}
1543 
1544 	if (da.nr_blocks) {
1545 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1546 		error = gfs2_quota_lock_check(ndip, &ap);
1547 		if (error)
1548 			goto out_gunlock;
1549 
1550 		error = gfs2_inplace_reserve(ndip, &ap);
1551 		if (error)
1552 			goto out_gunlock_q;
1553 
1554 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(ndip, &da, 4) +
1555 					 4 * RES_LEAF + 4, 0);
1556 		if (error)
1557 			goto out_ipreserv;
1558 	} else {
1559 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1560 					 5 * RES_LEAF + 4, 0);
1561 		if (error)
1562 			goto out_gunlock;
1563 	}
1564 
1565 	/* Remove the target file, if it exists */
1566 
1567 	if (nip)
1568 		error = gfs2_unlink_inode(ndip, ndentry);
1569 
1570 	error = update_moved_ino(ip, ndip, dir_rename);
1571 	if (error)
1572 		goto out_end_trans;
1573 
1574 	error = gfs2_dir_del(odip, odentry);
1575 	if (error)
1576 		goto out_end_trans;
1577 
1578 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip, &da);
1579 	if (error)
1580 		goto out_end_trans;
1581 
1582 out_end_trans:
1583 	gfs2_trans_end(sdp);
1584 out_ipreserv:
1585 	if (da.nr_blocks)
1586 		gfs2_inplace_release(ndip);
1587 out_gunlock_q:
1588 	if (da.nr_blocks)
1589 		gfs2_quota_unlock(ndip);
1590 out_gunlock:
1591 	gfs2_dir_no_add(&da);
1592 	if (gfs2_holder_initialized(&rd_gh))
1593 		gfs2_glock_dq_uninit(&rd_gh);
1594 
1595 	while (x--) {
1596 		if (gfs2_holder_queued(ghs + x))
1597 			gfs2_glock_dq(ghs + x);
1598 		gfs2_holder_uninit(ghs + x);
1599 	}
1600 out_gunlock_r:
1601 	if (gfs2_holder_initialized(&r_gh))
1602 		gfs2_glock_dq_uninit(&r_gh);
1603 out:
1604 	gfs2_qa_put(ndip);
1605 	return error;
1606 }
1607 
1608 /**
1609  * gfs2_exchange - exchange two files
1610  * @odir: Parent directory of old file name
1611  * @odentry: The old dentry of the file
1612  * @ndir: Parent directory of new file name
1613  * @ndentry: The new dentry of the file
1614  * @flags: The rename flags
1615  *
1616  * Returns: errno
1617  */
1618 
1619 static int gfs2_exchange(struct inode *odir, struct dentry *odentry,
1620 			 struct inode *ndir, struct dentry *ndentry,
1621 			 unsigned int flags)
1622 {
1623 	struct gfs2_inode *odip = GFS2_I(odir);
1624 	struct gfs2_inode *ndip = GFS2_I(ndir);
1625 	struct gfs2_inode *oip = GFS2_I(odentry->d_inode);
1626 	struct gfs2_inode *nip = GFS2_I(ndentry->d_inode);
1627 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1628 	struct gfs2_holder ghs[4], r_gh;
1629 	unsigned int num_gh;
1630 	unsigned int x;
1631 	umode_t old_mode = oip->i_inode.i_mode;
1632 	umode_t new_mode = nip->i_inode.i_mode;
1633 	int error;
1634 
1635 	gfs2_holder_mark_uninitialized(&r_gh);
1636 	error = gfs2_rindex_update(sdp);
1637 	if (error)
1638 		return error;
1639 
1640 	if (odip != ndip) {
1641 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1642 					   0, &r_gh);
1643 		if (error)
1644 			goto out;
1645 
1646 		if (S_ISDIR(old_mode)) {
1647 			/* don't move a directory into its subdir */
1648 			error = gfs2_ok_to_move(oip, ndip);
1649 			if (error)
1650 				goto out_gunlock_r;
1651 		}
1652 
1653 		if (S_ISDIR(new_mode)) {
1654 			/* don't move a directory into its subdir */
1655 			error = gfs2_ok_to_move(nip, odip);
1656 			if (error)
1657 				goto out_gunlock_r;
1658 		}
1659 	}
1660 
1661 	num_gh = 1;
1662 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs);
1663 	if (odip != ndip) {
1664 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC,
1665 				 ghs + num_gh);
1666 		num_gh++;
1667 	}
1668 	gfs2_holder_init(oip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1669 	num_gh++;
1670 
1671 	gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1672 	num_gh++;
1673 
1674 	for (x = 0; x < num_gh; x++) {
1675 		error = gfs2_glock_nq(ghs + x);
1676 		if (error)
1677 			goto out_gunlock;
1678 	}
1679 
1680 	error = gfs2_glock_async_wait(num_gh, ghs);
1681 	if (error)
1682 		goto out_gunlock;
1683 
1684 	error = -ENOENT;
1685 	if (oip->i_inode.i_nlink == 0 || nip->i_inode.i_nlink == 0)
1686 		goto out_gunlock;
1687 
1688 	error = gfs2_unlink_ok(odip, &odentry->d_name, oip);
1689 	if (error)
1690 		goto out_gunlock;
1691 	error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1692 	if (error)
1693 		goto out_gunlock;
1694 
1695 	if (S_ISDIR(old_mode)) {
1696 		error = gfs2_permission(&init_user_ns, odentry->d_inode,
1697 					MAY_WRITE);
1698 		if (error)
1699 			goto out_gunlock;
1700 	}
1701 	if (S_ISDIR(new_mode)) {
1702 		error = gfs2_permission(&init_user_ns, ndentry->d_inode,
1703 					MAY_WRITE);
1704 		if (error)
1705 			goto out_gunlock;
1706 	}
1707 	error = gfs2_trans_begin(sdp, 4 * RES_DINODE + 4 * RES_LEAF, 0);
1708 	if (error)
1709 		goto out_gunlock;
1710 
1711 	error = update_moved_ino(oip, ndip, S_ISDIR(old_mode));
1712 	if (error)
1713 		goto out_end_trans;
1714 
1715 	error = update_moved_ino(nip, odip, S_ISDIR(new_mode));
1716 	if (error)
1717 		goto out_end_trans;
1718 
1719 	error = gfs2_dir_mvino(ndip, &ndentry->d_name, oip,
1720 			       IF2DT(old_mode));
1721 	if (error)
1722 		goto out_end_trans;
1723 
1724 	error = gfs2_dir_mvino(odip, &odentry->d_name, nip,
1725 			       IF2DT(new_mode));
1726 	if (error)
1727 		goto out_end_trans;
1728 
1729 	if (odip != ndip) {
1730 		if (S_ISDIR(new_mode) && !S_ISDIR(old_mode)) {
1731 			inc_nlink(&odip->i_inode);
1732 			drop_nlink(&ndip->i_inode);
1733 		} else if (S_ISDIR(old_mode) && !S_ISDIR(new_mode)) {
1734 			inc_nlink(&ndip->i_inode);
1735 			drop_nlink(&odip->i_inode);
1736 		}
1737 	}
1738 	mark_inode_dirty(&ndip->i_inode);
1739 	if (odip != ndip)
1740 		mark_inode_dirty(&odip->i_inode);
1741 
1742 out_end_trans:
1743 	gfs2_trans_end(sdp);
1744 out_gunlock:
1745 	while (x--) {
1746 		if (gfs2_holder_queued(ghs + x))
1747 			gfs2_glock_dq(ghs + x);
1748 		gfs2_holder_uninit(ghs + x);
1749 	}
1750 out_gunlock_r:
1751 	if (gfs2_holder_initialized(&r_gh))
1752 		gfs2_glock_dq_uninit(&r_gh);
1753 out:
1754 	return error;
1755 }
1756 
1757 static int gfs2_rename2(struct user_namespace *mnt_userns, struct inode *odir,
1758 			struct dentry *odentry, struct inode *ndir,
1759 			struct dentry *ndentry, unsigned int flags)
1760 {
1761 	flags &= ~RENAME_NOREPLACE;
1762 
1763 	if (flags & ~RENAME_EXCHANGE)
1764 		return -EINVAL;
1765 
1766 	if (flags & RENAME_EXCHANGE)
1767 		return gfs2_exchange(odir, odentry, ndir, ndentry, flags);
1768 
1769 	return gfs2_rename(odir, odentry, ndir, ndentry);
1770 }
1771 
1772 /**
1773  * gfs2_get_link - Follow a symbolic link
1774  * @dentry: The dentry of the link
1775  * @inode: The inode of the link
1776  * @done: destructor for return value
1777  *
1778  * This can handle symlinks of any size.
1779  *
1780  * Returns: 0 on success or error code
1781  */
1782 
1783 static const char *gfs2_get_link(struct dentry *dentry,
1784 				 struct inode *inode,
1785 				 struct delayed_call *done)
1786 {
1787 	struct gfs2_inode *ip = GFS2_I(inode);
1788 	struct gfs2_holder i_gh;
1789 	struct buffer_head *dibh;
1790 	unsigned int size;
1791 	char *buf;
1792 	int error;
1793 
1794 	if (!dentry)
1795 		return ERR_PTR(-ECHILD);
1796 
1797 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1798 	error = gfs2_glock_nq(&i_gh);
1799 	if (error) {
1800 		gfs2_holder_uninit(&i_gh);
1801 		return ERR_PTR(error);
1802 	}
1803 
1804 	size = (unsigned int)i_size_read(&ip->i_inode);
1805 	if (size == 0) {
1806 		gfs2_consist_inode(ip);
1807 		buf = ERR_PTR(-EIO);
1808 		goto out;
1809 	}
1810 
1811 	error = gfs2_meta_inode_buffer(ip, &dibh);
1812 	if (error) {
1813 		buf = ERR_PTR(error);
1814 		goto out;
1815 	}
1816 
1817 	buf = kzalloc(size + 1, GFP_NOFS);
1818 	if (!buf)
1819 		buf = ERR_PTR(-ENOMEM);
1820 	else
1821 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1822 	brelse(dibh);
1823 out:
1824 	gfs2_glock_dq_uninit(&i_gh);
1825 	if (!IS_ERR(buf))
1826 		set_delayed_call(done, kfree_link, buf);
1827 	return buf;
1828 }
1829 
1830 /**
1831  * gfs2_permission -
1832  * @inode: The inode
1833  * @mask: The mask to be tested
1834  * @flags: Indicates whether this is an RCU path walk or not
1835  *
1836  * This may be called from the VFS directly, or from within GFS2 with the
1837  * inode locked, so we look to see if the glock is already locked and only
1838  * lock the glock if its not already been done.
1839  *
1840  * Returns: errno
1841  */
1842 
1843 int gfs2_permission(struct user_namespace *mnt_userns, struct inode *inode,
1844 		    int mask)
1845 {
1846 	struct gfs2_inode *ip;
1847 	struct gfs2_holder i_gh;
1848 	int error;
1849 
1850 	gfs2_holder_mark_uninitialized(&i_gh);
1851 	ip = GFS2_I(inode);
1852 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1853 		if (mask & MAY_NOT_BLOCK)
1854 			return -ECHILD;
1855 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1856 		if (error)
1857 			return error;
1858 	}
1859 
1860 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1861 		error = -EPERM;
1862 	else
1863 		error = generic_permission(&init_user_ns, inode, mask);
1864 	if (gfs2_holder_initialized(&i_gh))
1865 		gfs2_glock_dq_uninit(&i_gh);
1866 
1867 	return error;
1868 }
1869 
1870 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1871 {
1872 	setattr_copy(&init_user_ns, inode, attr);
1873 	mark_inode_dirty(inode);
1874 	return 0;
1875 }
1876 
1877 /**
1878  * gfs2_setattr_simple -
1879  * @ip:
1880  * @attr:
1881  *
1882  * Returns: errno
1883  */
1884 
1885 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1886 {
1887 	int error;
1888 
1889 	if (current->journal_info)
1890 		return __gfs2_setattr_simple(inode, attr);
1891 
1892 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1893 	if (error)
1894 		return error;
1895 
1896 	error = __gfs2_setattr_simple(inode, attr);
1897 	gfs2_trans_end(GFS2_SB(inode));
1898 	return error;
1899 }
1900 
1901 static int setattr_chown(struct inode *inode, struct iattr *attr)
1902 {
1903 	struct gfs2_inode *ip = GFS2_I(inode);
1904 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1905 	kuid_t ouid, nuid;
1906 	kgid_t ogid, ngid;
1907 	int error;
1908 	struct gfs2_alloc_parms ap;
1909 
1910 	ouid = inode->i_uid;
1911 	ogid = inode->i_gid;
1912 	nuid = attr->ia_uid;
1913 	ngid = attr->ia_gid;
1914 
1915 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1916 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1917 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1918 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1919 	error = gfs2_qa_get(ip);
1920 	if (error)
1921 		return error;
1922 
1923 	error = gfs2_rindex_update(sdp);
1924 	if (error)
1925 		goto out;
1926 
1927 	error = gfs2_quota_lock(ip, nuid, ngid);
1928 	if (error)
1929 		goto out;
1930 
1931 	ap.target = gfs2_get_inode_blocks(&ip->i_inode);
1932 
1933 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1934 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1935 		error = gfs2_quota_check(ip, nuid, ngid, &ap);
1936 		if (error)
1937 			goto out_gunlock_q;
1938 	}
1939 
1940 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1941 	if (error)
1942 		goto out_gunlock_q;
1943 
1944 	error = gfs2_setattr_simple(inode, attr);
1945 	if (error)
1946 		goto out_end_trans;
1947 
1948 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1949 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1950 		gfs2_quota_change(ip, -(s64)ap.target, ouid, ogid);
1951 		gfs2_quota_change(ip, ap.target, nuid, ngid);
1952 	}
1953 
1954 out_end_trans:
1955 	gfs2_trans_end(sdp);
1956 out_gunlock_q:
1957 	gfs2_quota_unlock(ip);
1958 out:
1959 	gfs2_qa_put(ip);
1960 	return error;
1961 }
1962 
1963 /**
1964  * gfs2_setattr - Change attributes on an inode
1965  * @dentry: The dentry which is changing
1966  * @attr: The structure describing the change
1967  *
1968  * The VFS layer wants to change one or more of an inodes attributes.  Write
1969  * that change out to disk.
1970  *
1971  * Returns: errno
1972  */
1973 
1974 static int gfs2_setattr(struct user_namespace *mnt_userns,
1975 			struct dentry *dentry, struct iattr *attr)
1976 {
1977 	struct inode *inode = d_inode(dentry);
1978 	struct gfs2_inode *ip = GFS2_I(inode);
1979 	struct gfs2_holder i_gh;
1980 	int error;
1981 
1982 	error = gfs2_qa_get(ip);
1983 	if (error)
1984 		return error;
1985 
1986 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1987 	if (error)
1988 		goto out;
1989 
1990 	error = -EPERM;
1991 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1992 		goto error;
1993 
1994 	error = setattr_prepare(&init_user_ns, dentry, attr);
1995 	if (error)
1996 		goto error;
1997 
1998 	if (attr->ia_valid & ATTR_SIZE)
1999 		error = gfs2_setattr_size(inode, attr->ia_size);
2000 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
2001 		error = setattr_chown(inode, attr);
2002 	else {
2003 		error = gfs2_setattr_simple(inode, attr);
2004 		if (!error && attr->ia_valid & ATTR_MODE)
2005 			error = posix_acl_chmod(&init_user_ns, inode,
2006 						inode->i_mode);
2007 	}
2008 
2009 error:
2010 	if (!error)
2011 		mark_inode_dirty(inode);
2012 	gfs2_glock_dq_uninit(&i_gh);
2013 out:
2014 	gfs2_qa_put(ip);
2015 	return error;
2016 }
2017 
2018 /**
2019  * gfs2_getattr - Read out an inode's attributes
2020  * @mnt_userns:	user namespace of the mount the inode was found from
2021  * @path: Object to query
2022  * @stat: The inode's stats
2023  * @request_mask: Mask of STATX_xxx flags indicating the caller's interests
2024  * @flags: AT_STATX_xxx setting
2025  *
2026  * This may be called from the VFS directly, or from within GFS2 with the
2027  * inode locked, so we look to see if the glock is already locked and only
2028  * lock the glock if its not already been done. Note that its the NFS
2029  * readdirplus operation which causes this to be called (from filldir)
2030  * with the glock already held.
2031  *
2032  * Returns: errno
2033  */
2034 
2035 static int gfs2_getattr(struct user_namespace *mnt_userns,
2036 			const struct path *path, struct kstat *stat,
2037 			u32 request_mask, unsigned int flags)
2038 {
2039 	struct inode *inode = d_inode(path->dentry);
2040 	struct gfs2_inode *ip = GFS2_I(inode);
2041 	struct gfs2_holder gh;
2042 	u32 gfsflags;
2043 	int error;
2044 
2045 	gfs2_holder_mark_uninitialized(&gh);
2046 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
2047 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
2048 		if (error)
2049 			return error;
2050 	}
2051 
2052 	gfsflags = ip->i_diskflags;
2053 	if (gfsflags & GFS2_DIF_APPENDONLY)
2054 		stat->attributes |= STATX_ATTR_APPEND;
2055 	if (gfsflags & GFS2_DIF_IMMUTABLE)
2056 		stat->attributes |= STATX_ATTR_IMMUTABLE;
2057 
2058 	stat->attributes_mask |= (STATX_ATTR_APPEND |
2059 				  STATX_ATTR_COMPRESSED |
2060 				  STATX_ATTR_ENCRYPTED |
2061 				  STATX_ATTR_IMMUTABLE |
2062 				  STATX_ATTR_NODUMP);
2063 
2064 	generic_fillattr(&init_user_ns, inode, stat);
2065 
2066 	if (gfs2_holder_initialized(&gh))
2067 		gfs2_glock_dq_uninit(&gh);
2068 
2069 	return 0;
2070 }
2071 
2072 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2073 		       u64 start, u64 len)
2074 {
2075 	struct gfs2_inode *ip = GFS2_I(inode);
2076 	struct gfs2_holder gh;
2077 	int ret;
2078 
2079 	inode_lock_shared(inode);
2080 
2081 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
2082 	if (ret)
2083 		goto out;
2084 
2085 	ret = iomap_fiemap(inode, fieinfo, start, len, &gfs2_iomap_ops);
2086 
2087 	gfs2_glock_dq_uninit(&gh);
2088 
2089 out:
2090 	inode_unlock_shared(inode);
2091 	return ret;
2092 }
2093 
2094 loff_t gfs2_seek_data(struct file *file, loff_t offset)
2095 {
2096 	struct inode *inode = file->f_mapping->host;
2097 	struct gfs2_inode *ip = GFS2_I(inode);
2098 	struct gfs2_holder gh;
2099 	loff_t ret;
2100 
2101 	inode_lock_shared(inode);
2102 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
2103 	if (!ret)
2104 		ret = iomap_seek_data(inode, offset, &gfs2_iomap_ops);
2105 	gfs2_glock_dq_uninit(&gh);
2106 	inode_unlock_shared(inode);
2107 
2108 	if (ret < 0)
2109 		return ret;
2110 	return vfs_setpos(file, ret, inode->i_sb->s_maxbytes);
2111 }
2112 
2113 loff_t gfs2_seek_hole(struct file *file, loff_t offset)
2114 {
2115 	struct inode *inode = file->f_mapping->host;
2116 	struct gfs2_inode *ip = GFS2_I(inode);
2117 	struct gfs2_holder gh;
2118 	loff_t ret;
2119 
2120 	inode_lock_shared(inode);
2121 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
2122 	if (!ret)
2123 		ret = iomap_seek_hole(inode, offset, &gfs2_iomap_ops);
2124 	gfs2_glock_dq_uninit(&gh);
2125 	inode_unlock_shared(inode);
2126 
2127 	if (ret < 0)
2128 		return ret;
2129 	return vfs_setpos(file, ret, inode->i_sb->s_maxbytes);
2130 }
2131 
2132 static int gfs2_update_time(struct inode *inode, struct timespec64 *time,
2133 			    int flags)
2134 {
2135 	struct gfs2_inode *ip = GFS2_I(inode);
2136 	struct gfs2_glock *gl = ip->i_gl;
2137 	struct gfs2_holder *gh;
2138 	int error;
2139 
2140 	gh = gfs2_glock_is_locked_by_me(gl);
2141 	if (gh && !gfs2_glock_is_held_excl(gl)) {
2142 		gfs2_glock_dq(gh);
2143 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, gh);
2144 		error = gfs2_glock_nq(gh);
2145 		if (error)
2146 			return error;
2147 	}
2148 	return generic_update_time(inode, time, flags);
2149 }
2150 
2151 static const struct inode_operations gfs2_file_iops = {
2152 	.permission = gfs2_permission,
2153 	.setattr = gfs2_setattr,
2154 	.getattr = gfs2_getattr,
2155 	.listxattr = gfs2_listxattr,
2156 	.fiemap = gfs2_fiemap,
2157 	.get_acl = gfs2_get_acl,
2158 	.set_acl = gfs2_set_acl,
2159 	.update_time = gfs2_update_time,
2160 };
2161 
2162 static const struct inode_operations gfs2_dir_iops = {
2163 	.create = gfs2_create,
2164 	.lookup = gfs2_lookup,
2165 	.link = gfs2_link,
2166 	.unlink = gfs2_unlink,
2167 	.symlink = gfs2_symlink,
2168 	.mkdir = gfs2_mkdir,
2169 	.rmdir = gfs2_unlink,
2170 	.mknod = gfs2_mknod,
2171 	.rename = gfs2_rename2,
2172 	.permission = gfs2_permission,
2173 	.setattr = gfs2_setattr,
2174 	.getattr = gfs2_getattr,
2175 	.listxattr = gfs2_listxattr,
2176 	.fiemap = gfs2_fiemap,
2177 	.get_acl = gfs2_get_acl,
2178 	.set_acl = gfs2_set_acl,
2179 	.update_time = gfs2_update_time,
2180 	.atomic_open = gfs2_atomic_open,
2181 };
2182 
2183 static const struct inode_operations gfs2_symlink_iops = {
2184 	.get_link = gfs2_get_link,
2185 	.permission = gfs2_permission,
2186 	.setattr = gfs2_setattr,
2187 	.getattr = gfs2_getattr,
2188 	.listxattr = gfs2_listxattr,
2189 	.fiemap = gfs2_fiemap,
2190 };
2191 
2192