xref: /openbmc/linux/fs/gfs2/inode.c (revision 0a94608f)
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 /**
44  * gfs2_set_iop - Sets inode operations
45  * @inode: The inode with correct i_mode filled in
46  *
47  * GFS2 lookup code fills in vfs inode contents based on info obtained
48  * from directory entry inside gfs2_inode_lookup().
49  */
50 
51 static void gfs2_set_iop(struct inode *inode)
52 {
53 	struct gfs2_sbd *sdp = GFS2_SB(inode);
54 	umode_t mode = inode->i_mode;
55 
56 	if (S_ISREG(mode)) {
57 		inode->i_op = &gfs2_file_iops;
58 		if (gfs2_localflocks(sdp))
59 			inode->i_fop = &gfs2_file_fops_nolock;
60 		else
61 			inode->i_fop = &gfs2_file_fops;
62 	} else if (S_ISDIR(mode)) {
63 		inode->i_op = &gfs2_dir_iops;
64 		if (gfs2_localflocks(sdp))
65 			inode->i_fop = &gfs2_dir_fops_nolock;
66 		else
67 			inode->i_fop = &gfs2_dir_fops;
68 	} else if (S_ISLNK(mode)) {
69 		inode->i_op = &gfs2_symlink_iops;
70 	} else {
71 		inode->i_op = &gfs2_file_iops;
72 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
73 	}
74 }
75 
76 static int iget_test(struct inode *inode, void *opaque)
77 {
78 	u64 no_addr = *(u64 *)opaque;
79 
80 	return GFS2_I(inode)->i_no_addr == no_addr;
81 }
82 
83 static int iget_set(struct inode *inode, void *opaque)
84 {
85 	u64 no_addr = *(u64 *)opaque;
86 
87 	GFS2_I(inode)->i_no_addr = no_addr;
88 	inode->i_ino = no_addr;
89 	return 0;
90 }
91 
92 /**
93  * gfs2_inode_lookup - Lookup an inode
94  * @sb: The super block
95  * @type: The type of the inode
96  * @no_addr: The inode number
97  * @no_formal_ino: The inode generation number
98  * @blktype: Requested block type (GFS2_BLKST_DINODE or GFS2_BLKST_UNLINKED;
99  *           GFS2_BLKST_FREE to indicate not to verify)
100  *
101  * If @type is DT_UNKNOWN, the inode type is fetched from disk.
102  *
103  * If @blktype is anything other than GFS2_BLKST_FREE (which is used as a
104  * placeholder because it doesn't otherwise make sense), the on-disk block type
105  * is verified to be @blktype.
106  *
107  * When @no_formal_ino is non-zero, this function will return ERR_PTR(-ESTALE)
108  * if it detects that @no_formal_ino doesn't match the actual inode generation
109  * number.  However, it doesn't always know unless @type is DT_UNKNOWN.
110  *
111  * Returns: A VFS inode, or an error
112  */
113 
114 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
115 				u64 no_addr, u64 no_formal_ino,
116 				unsigned int blktype)
117 {
118 	struct inode *inode;
119 	struct gfs2_inode *ip;
120 	struct gfs2_holder i_gh;
121 	int error;
122 
123 	gfs2_holder_mark_uninitialized(&i_gh);
124 	inode = iget5_locked(sb, no_addr, iget_test, iget_set, &no_addr);
125 	if (!inode)
126 		return ERR_PTR(-ENOMEM);
127 
128 	ip = GFS2_I(inode);
129 
130 	if (inode->i_state & I_NEW) {
131 		struct gfs2_sbd *sdp = GFS2_SB(inode);
132 		struct gfs2_glock *io_gl;
133 
134 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE,
135 				       &ip->i_gl);
136 		if (unlikely(error))
137 			goto fail;
138 
139 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE,
140 				       &io_gl);
141 		if (unlikely(error))
142 			goto fail;
143 
144 		if (blktype != GFS2_BLKST_UNLINKED)
145 			gfs2_cancel_delete_work(io_gl);
146 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT,
147 					   &ip->i_iopen_gh);
148 		gfs2_glock_put(io_gl);
149 		if (unlikely(error))
150 			goto fail;
151 
152 		if (type == DT_UNKNOWN || blktype != GFS2_BLKST_FREE) {
153 			/*
154 			 * The GL_SKIP flag indicates to skip reading the inode
155 			 * block.  We read the inode when instantiating it
156 			 * after possibly checking the block type.
157 			 */
158 			error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE,
159 						   GL_SKIP, &i_gh);
160 			if (error)
161 				goto fail;
162 
163 			error = -ESTALE;
164 			if (no_formal_ino &&
165 			    gfs2_inode_already_deleted(ip->i_gl, no_formal_ino))
166 				goto fail;
167 
168 			if (blktype != GFS2_BLKST_FREE) {
169 				error = gfs2_check_blk_type(sdp, no_addr,
170 							    blktype);
171 				if (error)
172 					goto fail;
173 			}
174 		}
175 
176 		set_bit(GLF_INSTANTIATE_NEEDED, &ip->i_gl->gl_flags);
177 
178 		/* Lowest possible timestamp; will be overwritten in gfs2_dinode_in. */
179 		inode->i_atime.tv_sec = 1LL << (8 * sizeof(inode->i_atime.tv_sec) - 1);
180 		inode->i_atime.tv_nsec = 0;
181 
182 		glock_set_object(ip->i_gl, ip);
183 
184 		if (type == DT_UNKNOWN) {
185 			/* Inode glock must be locked already */
186 			error = gfs2_instantiate(&i_gh);
187 			if (error) {
188 				glock_clear_object(ip->i_gl, ip);
189 				goto fail;
190 			}
191 		} else {
192 			ip->i_no_formal_ino = no_formal_ino;
193 			inode->i_mode = DT2IF(type);
194 		}
195 
196 		if (gfs2_holder_initialized(&i_gh))
197 			gfs2_glock_dq_uninit(&i_gh);
198 		glock_set_object(ip->i_iopen_gh.gh_gl, ip);
199 
200 		gfs2_set_iop(inode);
201 		unlock_new_inode(inode);
202 	}
203 
204 	if (no_formal_ino && ip->i_no_formal_ino &&
205 	    no_formal_ino != ip->i_no_formal_ino) {
206 		iput(inode);
207 		return ERR_PTR(-ESTALE);
208 	}
209 
210 	return inode;
211 
212 fail:
213 	if (gfs2_holder_initialized(&ip->i_iopen_gh))
214 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
215 	if (gfs2_holder_initialized(&i_gh))
216 		gfs2_glock_dq_uninit(&i_gh);
217 	iget_failed(inode);
218 	return ERR_PTR(error);
219 }
220 
221 /**
222  * gfs2_lookup_by_inum - look up an inode by inode number
223  * @sdp: The super block
224  * @no_addr: The inode number
225  * @no_formal_ino: The inode generation number (0 for any)
226  * @blktype: Requested block type (see gfs2_inode_lookup)
227  */
228 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
229 				  u64 no_formal_ino, unsigned int blktype)
230 {
231 	struct super_block *sb = sdp->sd_vfs;
232 	struct inode *inode;
233 	int error;
234 
235 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, no_formal_ino,
236 				  blktype);
237 	if (IS_ERR(inode))
238 		return inode;
239 
240 	if (no_formal_ino) {
241 		error = -EIO;
242 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
243 			goto fail_iput;
244 	}
245 	return inode;
246 
247 fail_iput:
248 	iput(inode);
249 	return ERR_PTR(error);
250 }
251 
252 
253 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
254 {
255 	struct qstr qstr;
256 	struct inode *inode;
257 	gfs2_str2qstr(&qstr, name);
258 	inode = gfs2_lookupi(dip, &qstr, 1);
259 	/* gfs2_lookupi has inconsistent callers: vfs
260 	 * related routines expect NULL for no entry found,
261 	 * gfs2_lookup_simple callers expect ENOENT
262 	 * and do not check for NULL.
263 	 */
264 	if (inode == NULL)
265 		return ERR_PTR(-ENOENT);
266 	else
267 		return inode;
268 }
269 
270 
271 /**
272  * gfs2_lookupi - Look up a filename in a directory and return its inode
273  * @dir: The inode of the directory containing the inode to look-up
274  * @name: The name of the inode to look for
275  * @is_root: If 1, ignore the caller's permissions
276  *
277  * This can be called via the VFS filldir function when NFS is doing
278  * a readdirplus and the inode which its intending to stat isn't
279  * already in cache. In this case we must not take the directory glock
280  * again, since the readdir call will have already taken that lock.
281  *
282  * Returns: errno
283  */
284 
285 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
286 			   int is_root)
287 {
288 	struct super_block *sb = dir->i_sb;
289 	struct gfs2_inode *dip = GFS2_I(dir);
290 	struct gfs2_holder d_gh;
291 	int error = 0;
292 	struct inode *inode = NULL;
293 
294 	gfs2_holder_mark_uninitialized(&d_gh);
295 	if (!name->len || name->len > GFS2_FNAMESIZE)
296 		return ERR_PTR(-ENAMETOOLONG);
297 
298 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
299 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
300 	     dir == d_inode(sb->s_root))) {
301 		igrab(dir);
302 		return dir;
303 	}
304 
305 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
306 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
307 		if (error)
308 			return ERR_PTR(error);
309 	}
310 
311 	if (!is_root) {
312 		error = gfs2_permission(&init_user_ns, dir, MAY_EXEC);
313 		if (error)
314 			goto out;
315 	}
316 
317 	inode = gfs2_dir_search(dir, name, false);
318 	if (IS_ERR(inode))
319 		error = PTR_ERR(inode);
320 out:
321 	if (gfs2_holder_initialized(&d_gh))
322 		gfs2_glock_dq_uninit(&d_gh);
323 	if (error == -ENOENT)
324 		return NULL;
325 	return inode ? inode : ERR_PTR(error);
326 }
327 
328 /**
329  * create_ok - OK to create a new on-disk inode here?
330  * @dip:  Directory in which dinode is to be created
331  * @name:  Name of new dinode
332  * @mode:
333  *
334  * Returns: errno
335  */
336 
337 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
338 		     umode_t mode)
339 {
340 	int error;
341 
342 	error = gfs2_permission(&init_user_ns, &dip->i_inode,
343 				MAY_WRITE | MAY_EXEC);
344 	if (error)
345 		return error;
346 
347 	/*  Don't create entries in an unlinked directory  */
348 	if (!dip->i_inode.i_nlink)
349 		return -ENOENT;
350 
351 	if (dip->i_entries == (u32)-1)
352 		return -EFBIG;
353 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
354 		return -EMLINK;
355 
356 	return 0;
357 }
358 
359 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
360 			       struct inode *inode)
361 {
362 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
363 	    (dip->i_inode.i_mode & S_ISUID) &&
364 	    !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
365 		if (S_ISDIR(inode->i_mode))
366 			inode->i_mode |= S_ISUID;
367 		else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
368 			inode->i_mode &= ~07111;
369 		inode->i_uid = dip->i_inode.i_uid;
370 	} else
371 		inode->i_uid = current_fsuid();
372 
373 	if (dip->i_inode.i_mode & S_ISGID) {
374 		if (S_ISDIR(inode->i_mode))
375 			inode->i_mode |= S_ISGID;
376 		inode->i_gid = dip->i_inode.i_gid;
377 	} else
378 		inode->i_gid = current_fsgid();
379 }
380 
381 static int alloc_dinode(struct gfs2_inode *ip, u32 flags, unsigned *dblocks)
382 {
383 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
384 	struct gfs2_alloc_parms ap = { .target = *dblocks, .aflags = flags, };
385 	int error;
386 
387 	error = gfs2_quota_lock_check(ip, &ap);
388 	if (error)
389 		goto out;
390 
391 	error = gfs2_inplace_reserve(ip, &ap);
392 	if (error)
393 		goto out_quota;
394 
395 	error = gfs2_trans_begin(sdp, (*dblocks * RES_RG_BIT) + RES_STATFS + RES_QUOTA, 0);
396 	if (error)
397 		goto out_ipreserv;
398 
399 	error = gfs2_alloc_blocks(ip, &ip->i_no_addr, dblocks, 1, &ip->i_generation);
400 	ip->i_no_formal_ino = ip->i_generation;
401 	ip->i_inode.i_ino = ip->i_no_addr;
402 	ip->i_goal = ip->i_no_addr;
403 
404 	gfs2_trans_end(sdp);
405 
406 out_ipreserv:
407 	gfs2_inplace_release(ip);
408 out_quota:
409 	gfs2_quota_unlock(ip);
410 out:
411 	return error;
412 }
413 
414 static void gfs2_init_dir(struct buffer_head *dibh,
415 			  const struct gfs2_inode *parent)
416 {
417 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
418 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
419 
420 	gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
421 	dent->de_inum = di->di_num; /* already GFS2 endian */
422 	dent->de_type = cpu_to_be16(DT_DIR);
423 
424 	dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
425 	gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
426 	gfs2_inum_out(parent, dent);
427 	dent->de_type = cpu_to_be16(DT_DIR);
428 
429 }
430 
431 /**
432  * gfs2_init_xattr - Initialise an xattr block for a new inode
433  * @ip: The inode in question
434  *
435  * This sets up an empty xattr block for a new inode, ready to
436  * take any ACLs, LSM xattrs, etc.
437  */
438 
439 static void gfs2_init_xattr(struct gfs2_inode *ip)
440 {
441 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
442 	struct buffer_head *bh;
443 	struct gfs2_ea_header *ea;
444 
445 	bh = gfs2_meta_new(ip->i_gl, ip->i_eattr);
446 	gfs2_trans_add_meta(ip->i_gl, bh);
447 	gfs2_metatype_set(bh, GFS2_METATYPE_EA, GFS2_FORMAT_EA);
448 	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
449 
450 	ea = GFS2_EA_BH2FIRST(bh);
451 	ea->ea_rec_len = cpu_to_be32(sdp->sd_jbsize);
452 	ea->ea_type = GFS2_EATYPE_UNUSED;
453 	ea->ea_flags = GFS2_EAFLAG_LAST;
454 
455 	brelse(bh);
456 }
457 
458 /**
459  * init_dinode - Fill in a new dinode structure
460  * @dip: The directory this inode is being created in
461  * @ip: The inode
462  * @symname: The symlink destination (if a symlink)
463  *
464  */
465 
466 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
467 			const char *symname)
468 {
469 	struct gfs2_dinode *di;
470 	struct buffer_head *dibh;
471 
472 	dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
473 	gfs2_trans_add_meta(ip->i_gl, dibh);
474 	di = (struct gfs2_dinode *)dibh->b_data;
475 	gfs2_dinode_out(ip, di);
476 
477 	di->di_major = cpu_to_be32(imajor(&ip->i_inode));
478 	di->di_minor = cpu_to_be32(iminor(&ip->i_inode));
479 	di->__pad1 = 0;
480 	di->__pad2 = 0;
481 	di->__pad3 = 0;
482 	memset(&di->__pad4, 0, sizeof(di->__pad4));
483 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
484 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
485 
486 	switch(ip->i_inode.i_mode & S_IFMT) {
487 	case S_IFDIR:
488 		gfs2_init_dir(dibh, dip);
489 		break;
490 	case S_IFLNK:
491 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
492 		break;
493 	}
494 
495 	set_buffer_uptodate(dibh);
496 	brelse(dibh);
497 }
498 
499 /**
500  * gfs2_trans_da_blks - Calculate number of blocks to link inode
501  * @dip: The directory we are linking into
502  * @da: The dir add information
503  * @nr_inodes: The number of inodes involved
504  *
505  * This calculate the number of blocks we need to reserve in a
506  * transaction to link @nr_inodes into a directory. In most cases
507  * @nr_inodes will be 2 (the directory plus the inode being linked in)
508  * but in case of rename, 4 may be required.
509  *
510  * Returns: Number of blocks
511  */
512 
513 static unsigned gfs2_trans_da_blks(const struct gfs2_inode *dip,
514 				   const struct gfs2_diradd *da,
515 				   unsigned nr_inodes)
516 {
517 	return da->nr_blocks + gfs2_rg_blocks(dip, da->nr_blocks) +
518 	       (nr_inodes * RES_DINODE) + RES_QUOTA + RES_STATFS;
519 }
520 
521 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
522 		       struct gfs2_inode *ip, struct gfs2_diradd *da)
523 {
524 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
525 	struct gfs2_alloc_parms ap = { .target = da->nr_blocks, };
526 	int error;
527 
528 	if (da->nr_blocks) {
529 		error = gfs2_quota_lock_check(dip, &ap);
530 		if (error)
531 			goto fail_quota_locks;
532 
533 		error = gfs2_inplace_reserve(dip, &ap);
534 		if (error)
535 			goto fail_quota_locks;
536 
537 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, da, 2), 0);
538 		if (error)
539 			goto fail_ipreserv;
540 	} else {
541 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
542 		if (error)
543 			goto fail_quota_locks;
544 	}
545 
546 	error = gfs2_dir_add(&dip->i_inode, name, ip, da);
547 
548 	gfs2_trans_end(sdp);
549 fail_ipreserv:
550 	gfs2_inplace_release(dip);
551 fail_quota_locks:
552 	gfs2_quota_unlock(dip);
553 	return error;
554 }
555 
556 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
557 		    void *fs_info)
558 {
559 	const struct xattr *xattr;
560 	int err = 0;
561 
562 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
563 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
564 				       xattr->value_len, 0,
565 				       GFS2_EATYPE_SECURITY);
566 		if (err < 0)
567 			break;
568 	}
569 	return err;
570 }
571 
572 /**
573  * gfs2_create_inode - Create a new inode
574  * @dir: The parent directory
575  * @dentry: The new dentry
576  * @file: If non-NULL, the file which is being opened
577  * @mode: The permissions on the new inode
578  * @dev: For device nodes, this is the device number
579  * @symname: For symlinks, this is the link destination
580  * @size: The initial size of the inode (ignored for directories)
581  * @excl: Force fail if inode exists
582  *
583  * Returns: 0 on success, or error code
584  */
585 
586 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
587 			     struct file *file,
588 			     umode_t mode, dev_t dev, const char *symname,
589 			     unsigned int size, int excl)
590 {
591 	const struct qstr *name = &dentry->d_name;
592 	struct posix_acl *default_acl, *acl;
593 	struct gfs2_holder ghs[2];
594 	struct inode *inode = NULL;
595 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
596 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
597 	struct gfs2_glock *io_gl;
598 	int error, free_vfs_inode = 1;
599 	u32 aflags = 0;
600 	unsigned blocks = 1;
601 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
602 
603 	if (!name->len || name->len > GFS2_FNAMESIZE)
604 		return -ENAMETOOLONG;
605 
606 	error = gfs2_qa_get(dip);
607 	if (error)
608 		return error;
609 
610 	error = gfs2_rindex_update(sdp);
611 	if (error)
612 		goto fail;
613 
614 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
615 	if (error)
616 		goto fail;
617 	gfs2_holder_mark_uninitialized(ghs + 1);
618 
619 	error = create_ok(dip, name, mode);
620 	if (error)
621 		goto fail_gunlock;
622 
623 	inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
624 	error = PTR_ERR(inode);
625 	if (!IS_ERR(inode)) {
626 		if (S_ISDIR(inode->i_mode)) {
627 			iput(inode);
628 			inode = ERR_PTR(-EISDIR);
629 			goto fail_gunlock;
630 		}
631 		d_instantiate(dentry, inode);
632 		error = 0;
633 		if (file) {
634 			if (S_ISREG(inode->i_mode))
635 				error = finish_open(file, dentry, gfs2_open_common);
636 			else
637 				error = finish_no_open(file, NULL);
638 		}
639 		gfs2_glock_dq_uninit(ghs);
640 		goto fail;
641 	} else if (error != -ENOENT) {
642 		goto fail_gunlock;
643 	}
644 
645 	error = gfs2_diradd_alloc_required(dir, name, &da);
646 	if (error < 0)
647 		goto fail_gunlock;
648 
649 	inode = new_inode(sdp->sd_vfs);
650 	error = -ENOMEM;
651 	if (!inode)
652 		goto fail_gunlock;
653 
654 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
655 	if (error)
656 		goto fail_gunlock;
657 
658 	ip = GFS2_I(inode);
659 	error = gfs2_qa_get(ip);
660 	if (error)
661 		goto fail_free_acls;
662 
663 	inode->i_mode = mode;
664 	set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
665 	inode->i_rdev = dev;
666 	inode->i_size = size;
667 	inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
668 	munge_mode_uid_gid(dip, inode);
669 	check_and_update_goal(dip);
670 	ip->i_goal = dip->i_goal;
671 	ip->i_diskflags = 0;
672 	ip->i_eattr = 0;
673 	ip->i_height = 0;
674 	ip->i_depth = 0;
675 	ip->i_entries = 0;
676 	ip->i_no_addr = 0; /* Temporarily zero until real addr is assigned */
677 
678 	switch(mode & S_IFMT) {
679 	case S_IFREG:
680 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
681 		    gfs2_tune_get(sdp, gt_new_files_jdata))
682 			ip->i_diskflags |= GFS2_DIF_JDATA;
683 		gfs2_set_aops(inode);
684 		break;
685 	case S_IFDIR:
686 		ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
687 		ip->i_diskflags |= GFS2_DIF_JDATA;
688 		ip->i_entries = 2;
689 		break;
690 	}
691 
692 	/* Force SYSTEM flag on all files and subdirs of a SYSTEM directory */
693 	if (dip->i_diskflags & GFS2_DIF_SYSTEM)
694 		ip->i_diskflags |= GFS2_DIF_SYSTEM;
695 
696 	gfs2_set_inode_flags(inode);
697 
698 	if ((GFS2_I(d_inode(sdp->sd_root_dir)) == dip) ||
699 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
700 		aflags |= GFS2_AF_ORLOV;
701 
702 	if (default_acl || acl)
703 		blocks++;
704 
705 	error = alloc_dinode(ip, aflags, &blocks);
706 	if (error)
707 		goto fail_free_inode;
708 
709 	gfs2_set_inode_blocks(inode, blocks);
710 
711 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
712 	if (error)
713 		goto fail_free_inode;
714 
715 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
716 	if (error)
717 		goto fail_free_inode;
718 	gfs2_cancel_delete_work(io_gl);
719 
720 	error = insert_inode_locked4(inode, ip->i_no_addr, iget_test, &ip->i_no_addr);
721 	BUG_ON(error);
722 
723 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
724 	if (error)
725 		goto fail_gunlock2;
726 
727 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
728 	if (error)
729 		goto fail_gunlock3;
730 
731 	error = gfs2_trans_begin(sdp, blocks, 0);
732 	if (error)
733 		goto fail_gunlock3;
734 
735 	if (blocks > 1) {
736 		ip->i_eattr = ip->i_no_addr + 1;
737 		gfs2_init_xattr(ip);
738 	}
739 	init_dinode(dip, ip, symname);
740 	gfs2_trans_end(sdp);
741 
742 	glock_set_object(ip->i_gl, ip);
743 	glock_set_object(io_gl, ip);
744 	gfs2_set_iop(inode);
745 
746 	free_vfs_inode = 0; /* After this point, the inode is no longer
747 			       considered free. Any failures need to undo
748 			       the gfs2 structures. */
749 	if (default_acl) {
750 		error = __gfs2_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
751 		if (error)
752 			goto fail_gunlock4;
753 		posix_acl_release(default_acl);
754 		default_acl = NULL;
755 	}
756 	if (acl) {
757 		error = __gfs2_set_acl(inode, acl, ACL_TYPE_ACCESS);
758 		if (error)
759 			goto fail_gunlock4;
760 		posix_acl_release(acl);
761 		acl = NULL;
762 	}
763 
764 	error = security_inode_init_security(&ip->i_inode, &dip->i_inode, name,
765 					     &gfs2_initxattrs, NULL);
766 	if (error)
767 		goto fail_gunlock4;
768 
769 	error = link_dinode(dip, name, ip, &da);
770 	if (error)
771 		goto fail_gunlock4;
772 
773 	mark_inode_dirty(inode);
774 	d_instantiate(dentry, inode);
775 	/* After instantiate, errors should result in evict which will destroy
776 	 * both inode and iopen glocks properly. */
777 	if (file) {
778 		file->f_mode |= FMODE_CREATED;
779 		error = finish_open(file, dentry, gfs2_open_common);
780 	}
781 	gfs2_glock_dq_uninit(ghs);
782 	gfs2_qa_put(ip);
783 	gfs2_glock_dq_uninit(ghs + 1);
784 	gfs2_glock_put(io_gl);
785 	gfs2_qa_put(dip);
786 	unlock_new_inode(inode);
787 	return error;
788 
789 fail_gunlock4:
790 	glock_clear_object(ip->i_gl, ip);
791 	glock_clear_object(io_gl, ip);
792 fail_gunlock3:
793 	gfs2_glock_dq_uninit(&ip->i_iopen_gh);
794 fail_gunlock2:
795 	gfs2_glock_put(io_gl);
796 fail_free_inode:
797 	if (ip->i_gl) {
798 		if (free_vfs_inode) /* else evict will do the put for us */
799 			gfs2_glock_put(ip->i_gl);
800 	}
801 	gfs2_rs_deltree(&ip->i_res);
802 	gfs2_qa_put(ip);
803 fail_free_acls:
804 	posix_acl_release(default_acl);
805 	posix_acl_release(acl);
806 fail_gunlock:
807 	gfs2_dir_no_add(&da);
808 	gfs2_glock_dq_uninit(ghs);
809 	if (!IS_ERR_OR_NULL(inode)) {
810 		clear_nlink(inode);
811 		if (!free_vfs_inode)
812 			mark_inode_dirty(inode);
813 		set_bit(free_vfs_inode ? GIF_FREE_VFS_INODE : GIF_ALLOC_FAILED,
814 			&GFS2_I(inode)->i_flags);
815 		if (inode->i_state & I_NEW)
816 			iget_failed(inode);
817 		else
818 			iput(inode);
819 	}
820 	if (gfs2_holder_initialized(ghs + 1))
821 		gfs2_glock_dq_uninit(ghs + 1);
822 fail:
823 	gfs2_qa_put(dip);
824 	return error;
825 }
826 
827 /**
828  * gfs2_create - Create a file
829  * @mnt_userns: User namespace of the mount the inode was found from
830  * @dir: The directory in which to create the file
831  * @dentry: The dentry of the new file
832  * @mode: The mode of the new file
833  * @excl: Force fail if inode exists
834  *
835  * Returns: errno
836  */
837 
838 static int gfs2_create(struct user_namespace *mnt_userns, struct inode *dir,
839 		       struct dentry *dentry, umode_t mode, bool excl)
840 {
841 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl);
842 }
843 
844 /**
845  * __gfs2_lookup - Look up a filename in a directory and return its inode
846  * @dir: The directory inode
847  * @dentry: The dentry of the new inode
848  * @file: File to be opened
849  *
850  *
851  * Returns: errno
852  */
853 
854 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
855 				    struct file *file)
856 {
857 	struct inode *inode;
858 	struct dentry *d;
859 	struct gfs2_holder gh;
860 	struct gfs2_glock *gl;
861 	int error;
862 
863 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
864 	if (inode == NULL) {
865 		d_add(dentry, NULL);
866 		return NULL;
867 	}
868 	if (IS_ERR(inode))
869 		return ERR_CAST(inode);
870 
871 	gl = GFS2_I(inode)->i_gl;
872 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
873 	if (error) {
874 		iput(inode);
875 		return ERR_PTR(error);
876 	}
877 
878 	d = d_splice_alias(inode, dentry);
879 	if (IS_ERR(d)) {
880 		gfs2_glock_dq_uninit(&gh);
881 		return d;
882 	}
883 	if (file && S_ISREG(inode->i_mode))
884 		error = finish_open(file, dentry, gfs2_open_common);
885 
886 	gfs2_glock_dq_uninit(&gh);
887 	if (error) {
888 		dput(d);
889 		return ERR_PTR(error);
890 	}
891 	return d;
892 }
893 
894 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
895 				  unsigned flags)
896 {
897 	return __gfs2_lookup(dir, dentry, NULL);
898 }
899 
900 /**
901  * gfs2_link - Link to a file
902  * @old_dentry: The inode to link
903  * @dir: Add link to this directory
904  * @dentry: The name of the link
905  *
906  * Link the inode in "old_dentry" into the directory "dir" with the
907  * name in "dentry".
908  *
909  * Returns: errno
910  */
911 
912 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
913 		     struct dentry *dentry)
914 {
915 	struct gfs2_inode *dip = GFS2_I(dir);
916 	struct gfs2_sbd *sdp = GFS2_SB(dir);
917 	struct inode *inode = d_inode(old_dentry);
918 	struct gfs2_inode *ip = GFS2_I(inode);
919 	struct gfs2_holder ghs[2];
920 	struct buffer_head *dibh;
921 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
922 	int error;
923 
924 	if (S_ISDIR(inode->i_mode))
925 		return -EPERM;
926 
927 	error = gfs2_qa_get(dip);
928 	if (error)
929 		return error;
930 
931 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
932 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
933 
934 	error = gfs2_glock_nq(ghs); /* parent */
935 	if (error)
936 		goto out_parent;
937 
938 	error = gfs2_glock_nq(ghs + 1); /* child */
939 	if (error)
940 		goto out_child;
941 
942 	error = -ENOENT;
943 	if (inode->i_nlink == 0)
944 		goto out_gunlock;
945 
946 	error = gfs2_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
947 	if (error)
948 		goto out_gunlock;
949 
950 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
951 	switch (error) {
952 	case -ENOENT:
953 		break;
954 	case 0:
955 		error = -EEXIST;
956 		goto out_gunlock;
957 	default:
958 		goto out_gunlock;
959 	}
960 
961 	error = -EINVAL;
962 	if (!dip->i_inode.i_nlink)
963 		goto out_gunlock;
964 	error = -EFBIG;
965 	if (dip->i_entries == (u32)-1)
966 		goto out_gunlock;
967 	error = -EPERM;
968 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
969 		goto out_gunlock;
970 	error = -EINVAL;
971 	if (!ip->i_inode.i_nlink)
972 		goto out_gunlock;
973 	error = -EMLINK;
974 	if (ip->i_inode.i_nlink == (u32)-1)
975 		goto out_gunlock;
976 
977 	error = gfs2_diradd_alloc_required(dir, &dentry->d_name, &da);
978 	if (error < 0)
979 		goto out_gunlock;
980 
981 	if (da.nr_blocks) {
982 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
983 		error = gfs2_quota_lock_check(dip, &ap);
984 		if (error)
985 			goto out_gunlock;
986 
987 		error = gfs2_inplace_reserve(dip, &ap);
988 		if (error)
989 			goto out_gunlock_q;
990 
991 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, &da, 2), 0);
992 		if (error)
993 			goto out_ipres;
994 	} else {
995 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
996 		if (error)
997 			goto out_ipres;
998 	}
999 
1000 	error = gfs2_meta_inode_buffer(ip, &dibh);
1001 	if (error)
1002 		goto out_end_trans;
1003 
1004 	error = gfs2_dir_add(dir, &dentry->d_name, ip, &da);
1005 	if (error)
1006 		goto out_brelse;
1007 
1008 	gfs2_trans_add_meta(ip->i_gl, dibh);
1009 	inc_nlink(&ip->i_inode);
1010 	ip->i_inode.i_ctime = current_time(&ip->i_inode);
1011 	ihold(inode);
1012 	d_instantiate(dentry, inode);
1013 	mark_inode_dirty(inode);
1014 
1015 out_brelse:
1016 	brelse(dibh);
1017 out_end_trans:
1018 	gfs2_trans_end(sdp);
1019 out_ipres:
1020 	if (da.nr_blocks)
1021 		gfs2_inplace_release(dip);
1022 out_gunlock_q:
1023 	if (da.nr_blocks)
1024 		gfs2_quota_unlock(dip);
1025 out_gunlock:
1026 	gfs2_dir_no_add(&da);
1027 	gfs2_glock_dq(ghs + 1);
1028 out_child:
1029 	gfs2_glock_dq(ghs);
1030 out_parent:
1031 	gfs2_qa_put(dip);
1032 	gfs2_holder_uninit(ghs);
1033 	gfs2_holder_uninit(ghs + 1);
1034 	return error;
1035 }
1036 
1037 /*
1038  * gfs2_unlink_ok - check to see that a inode is still in a directory
1039  * @dip: the directory
1040  * @name: the name of the file
1041  * @ip: the inode
1042  *
1043  * Assumes that the lock on (at least) @dip is held.
1044  *
1045  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1046  */
1047 
1048 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
1049 			  const struct gfs2_inode *ip)
1050 {
1051 	int error;
1052 
1053 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
1054 		return -EPERM;
1055 
1056 	if ((dip->i_inode.i_mode & S_ISVTX) &&
1057 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
1058 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
1059 		return -EPERM;
1060 
1061 	if (IS_APPEND(&dip->i_inode))
1062 		return -EPERM;
1063 
1064 	error = gfs2_permission(&init_user_ns, &dip->i_inode,
1065 				MAY_WRITE | MAY_EXEC);
1066 	if (error)
1067 		return error;
1068 
1069 	return gfs2_dir_check(&dip->i_inode, name, ip);
1070 }
1071 
1072 /**
1073  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1074  * @dip: The parent directory
1075  * @dentry: The dentry to unlink
1076  *
1077  * Called with all the locks and in a transaction. This will only be
1078  * called for a directory after it has been checked to ensure it is empty.
1079  *
1080  * Returns: 0 on success, or an error
1081  */
1082 
1083 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1084 			     const struct dentry *dentry)
1085 {
1086 	struct inode *inode = d_inode(dentry);
1087 	struct gfs2_inode *ip = GFS2_I(inode);
1088 	int error;
1089 
1090 	error = gfs2_dir_del(dip, dentry);
1091 	if (error)
1092 		return error;
1093 
1094 	ip->i_entries = 0;
1095 	inode->i_ctime = current_time(inode);
1096 	if (S_ISDIR(inode->i_mode))
1097 		clear_nlink(inode);
1098 	else
1099 		drop_nlink(inode);
1100 	mark_inode_dirty(inode);
1101 	if (inode->i_nlink == 0)
1102 		gfs2_unlink_di(inode);
1103 	return 0;
1104 }
1105 
1106 
1107 /**
1108  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1109  * @dir: The inode of the directory containing the inode to unlink
1110  * @dentry: The file itself
1111  *
1112  * This routine uses the type of the inode as a flag to figure out
1113  * whether this is an unlink or an rmdir.
1114  *
1115  * Returns: errno
1116  */
1117 
1118 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1119 {
1120 	struct gfs2_inode *dip = GFS2_I(dir);
1121 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1122 	struct inode *inode = d_inode(dentry);
1123 	struct gfs2_inode *ip = GFS2_I(inode);
1124 	struct gfs2_holder ghs[3];
1125 	struct gfs2_rgrpd *rgd;
1126 	int error;
1127 
1128 	error = gfs2_rindex_update(sdp);
1129 	if (error)
1130 		return error;
1131 
1132 	error = -EROFS;
1133 
1134 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1135 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1136 
1137 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1138 	if (!rgd)
1139 		goto out_inodes;
1140 
1141 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, LM_FLAG_NODE_SCOPE, ghs + 2);
1142 
1143 
1144 	error = gfs2_glock_nq(ghs); /* parent */
1145 	if (error)
1146 		goto out_parent;
1147 
1148 	error = gfs2_glock_nq(ghs + 1); /* child */
1149 	if (error)
1150 		goto out_child;
1151 
1152 	error = -ENOENT;
1153 	if (inode->i_nlink == 0)
1154 		goto out_rgrp;
1155 
1156 	if (S_ISDIR(inode->i_mode)) {
1157 		error = -ENOTEMPTY;
1158 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1159 			goto out_rgrp;
1160 	}
1161 
1162 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1163 	if (error)
1164 		goto out_rgrp;
1165 
1166 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1167 	if (error)
1168 		goto out_gunlock;
1169 
1170 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1171 	if (error)
1172 		goto out_gunlock;
1173 
1174 	error = gfs2_unlink_inode(dip, dentry);
1175 	gfs2_trans_end(sdp);
1176 
1177 out_gunlock:
1178 	gfs2_glock_dq(ghs + 2);
1179 out_rgrp:
1180 	gfs2_glock_dq(ghs + 1);
1181 out_child:
1182 	gfs2_glock_dq(ghs);
1183 out_parent:
1184 	gfs2_holder_uninit(ghs + 2);
1185 out_inodes:
1186 	gfs2_holder_uninit(ghs + 1);
1187 	gfs2_holder_uninit(ghs);
1188 	return error;
1189 }
1190 
1191 /**
1192  * gfs2_symlink - Create a symlink
1193  * @mnt_userns: User namespace of the mount the inode was found from
1194  * @dir: The directory to create the symlink in
1195  * @dentry: The dentry to put the symlink in
1196  * @symname: The thing which the link points to
1197  *
1198  * Returns: errno
1199  */
1200 
1201 static int gfs2_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1202 			struct dentry *dentry, const char *symname)
1203 {
1204 	unsigned int size;
1205 
1206 	size = strlen(symname);
1207 	if (size >= gfs2_max_stuffed_size(GFS2_I(dir)))
1208 		return -ENAMETOOLONG;
1209 
1210 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1211 }
1212 
1213 /**
1214  * gfs2_mkdir - Make a directory
1215  * @mnt_userns: User namespace of the mount the inode was found from
1216  * @dir: The parent directory of the new one
1217  * @dentry: The dentry of the new directory
1218  * @mode: The mode of the new directory
1219  *
1220  * Returns: errno
1221  */
1222 
1223 static int gfs2_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1224 		      struct dentry *dentry, umode_t mode)
1225 {
1226 	unsigned dsize = gfs2_max_stuffed_size(GFS2_I(dir));
1227 	return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0);
1228 }
1229 
1230 /**
1231  * gfs2_mknod - Make a special file
1232  * @mnt_userns: User namespace of the mount the inode was found from
1233  * @dir: The directory in which the special file will reside
1234  * @dentry: The dentry of the special file
1235  * @mode: The mode of the special file
1236  * @dev: The device specification of the special file
1237  *
1238  */
1239 
1240 static int gfs2_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1241 		      struct dentry *dentry, umode_t mode, dev_t dev)
1242 {
1243 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0);
1244 }
1245 
1246 /**
1247  * gfs2_atomic_open - Atomically open a file
1248  * @dir: The directory
1249  * @dentry: The proposed new entry
1250  * @file: The proposed new struct file
1251  * @flags: open flags
1252  * @mode: File mode
1253  *
1254  * Returns: error code or 0 for success
1255  */
1256 
1257 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1258 			    struct file *file, unsigned flags,
1259 			    umode_t mode)
1260 {
1261 	struct dentry *d;
1262 	bool excl = !!(flags & O_EXCL);
1263 
1264 	if (!d_in_lookup(dentry))
1265 		goto skip_lookup;
1266 
1267 	d = __gfs2_lookup(dir, dentry, file);
1268 	if (IS_ERR(d))
1269 		return PTR_ERR(d);
1270 	if (d != NULL)
1271 		dentry = d;
1272 	if (d_really_is_positive(dentry)) {
1273 		if (!(file->f_mode & FMODE_OPENED))
1274 			return finish_no_open(file, d);
1275 		dput(d);
1276 		return excl && (flags & O_CREAT) ? -EEXIST : 0;
1277 	}
1278 
1279 	BUG_ON(d != NULL);
1280 
1281 skip_lookup:
1282 	if (!(flags & O_CREAT))
1283 		return -ENOENT;
1284 
1285 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl);
1286 }
1287 
1288 /*
1289  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1290  * @this: move this
1291  * @to: to here
1292  *
1293  * Follow @to back to the root and make sure we don't encounter @this
1294  * Assumes we already hold the rename lock.
1295  *
1296  * Returns: errno
1297  */
1298 
1299 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1300 {
1301 	struct inode *dir = &to->i_inode;
1302 	struct super_block *sb = dir->i_sb;
1303 	struct inode *tmp;
1304 	int error = 0;
1305 
1306 	igrab(dir);
1307 
1308 	for (;;) {
1309 		if (dir == &this->i_inode) {
1310 			error = -EINVAL;
1311 			break;
1312 		}
1313 		if (dir == d_inode(sb->s_root)) {
1314 			error = 0;
1315 			break;
1316 		}
1317 
1318 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1319 		if (!tmp) {
1320 			error = -ENOENT;
1321 			break;
1322 		}
1323 		if (IS_ERR(tmp)) {
1324 			error = PTR_ERR(tmp);
1325 			break;
1326 		}
1327 
1328 		iput(dir);
1329 		dir = tmp;
1330 	}
1331 
1332 	iput(dir);
1333 
1334 	return error;
1335 }
1336 
1337 /**
1338  * update_moved_ino - Update an inode that's being moved
1339  * @ip: The inode being moved
1340  * @ndip: The parent directory of the new filename
1341  * @dir_rename: True of ip is a directory
1342  *
1343  * Returns: errno
1344  */
1345 
1346 static int update_moved_ino(struct gfs2_inode *ip, struct gfs2_inode *ndip,
1347 			    int dir_rename)
1348 {
1349 	if (dir_rename)
1350 		return gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1351 
1352 	ip->i_inode.i_ctime = current_time(&ip->i_inode);
1353 	mark_inode_dirty_sync(&ip->i_inode);
1354 	return 0;
1355 }
1356 
1357 
1358 /**
1359  * gfs2_rename - Rename a file
1360  * @odir: Parent directory of old file name
1361  * @odentry: The old dentry of the file
1362  * @ndir: Parent directory of new file name
1363  * @ndentry: The new dentry of the file
1364  *
1365  * Returns: errno
1366  */
1367 
1368 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1369 		       struct inode *ndir, struct dentry *ndentry)
1370 {
1371 	struct gfs2_inode *odip = GFS2_I(odir);
1372 	struct gfs2_inode *ndip = GFS2_I(ndir);
1373 	struct gfs2_inode *ip = GFS2_I(d_inode(odentry));
1374 	struct gfs2_inode *nip = NULL;
1375 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1376 	struct gfs2_holder ghs[4], r_gh, rd_gh;
1377 	struct gfs2_rgrpd *nrgd;
1378 	unsigned int num_gh;
1379 	int dir_rename = 0;
1380 	struct gfs2_diradd da = { .nr_blocks = 0, .save_loc = 0, };
1381 	unsigned int x;
1382 	int error;
1383 
1384 	gfs2_holder_mark_uninitialized(&r_gh);
1385 	gfs2_holder_mark_uninitialized(&rd_gh);
1386 	if (d_really_is_positive(ndentry)) {
1387 		nip = GFS2_I(d_inode(ndentry));
1388 		if (ip == nip)
1389 			return 0;
1390 	}
1391 
1392 	error = gfs2_rindex_update(sdp);
1393 	if (error)
1394 		return error;
1395 
1396 	error = gfs2_qa_get(ndip);
1397 	if (error)
1398 		return error;
1399 
1400 	if (odip != ndip) {
1401 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1402 					   0, &r_gh);
1403 		if (error)
1404 			goto out;
1405 
1406 		if (S_ISDIR(ip->i_inode.i_mode)) {
1407 			dir_rename = 1;
1408 			/* don't move a directory into its subdir */
1409 			error = gfs2_ok_to_move(ip, ndip);
1410 			if (error)
1411 				goto out_gunlock_r;
1412 		}
1413 	}
1414 
1415 	num_gh = 1;
1416 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs);
1417 	if (odip != ndip) {
1418 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE,GL_ASYNC,
1419 				 ghs + num_gh);
1420 		num_gh++;
1421 	}
1422 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1423 	num_gh++;
1424 
1425 	if (nip) {
1426 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC,
1427 				 ghs + num_gh);
1428 		num_gh++;
1429 	}
1430 
1431 	for (x = 0; x < num_gh; x++) {
1432 		error = gfs2_glock_nq(ghs + x);
1433 		if (error)
1434 			goto out_gunlock;
1435 	}
1436 	error = gfs2_glock_async_wait(num_gh, ghs);
1437 	if (error)
1438 		goto out_gunlock;
1439 
1440 	if (nip) {
1441 		/* Grab the resource group glock for unlink flag twiddling.
1442 		 * This is the case where the target dinode already exists
1443 		 * so we unlink before doing the rename.
1444 		 */
1445 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1446 		if (!nrgd) {
1447 			error = -ENOENT;
1448 			goto out_gunlock;
1449 		}
1450 		error = gfs2_glock_nq_init(nrgd->rd_gl, LM_ST_EXCLUSIVE,
1451 					   LM_FLAG_NODE_SCOPE, &rd_gh);
1452 		if (error)
1453 			goto out_gunlock;
1454 	}
1455 
1456 	error = -ENOENT;
1457 	if (ip->i_inode.i_nlink == 0)
1458 		goto out_gunlock;
1459 
1460 	/* Check out the old directory */
1461 
1462 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1463 	if (error)
1464 		goto out_gunlock;
1465 
1466 	/* Check out the new directory */
1467 
1468 	if (nip) {
1469 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1470 		if (error)
1471 			goto out_gunlock;
1472 
1473 		if (nip->i_inode.i_nlink == 0) {
1474 			error = -EAGAIN;
1475 			goto out_gunlock;
1476 		}
1477 
1478 		if (S_ISDIR(nip->i_inode.i_mode)) {
1479 			if (nip->i_entries < 2) {
1480 				gfs2_consist_inode(nip);
1481 				error = -EIO;
1482 				goto out_gunlock;
1483 			}
1484 			if (nip->i_entries > 2) {
1485 				error = -ENOTEMPTY;
1486 				goto out_gunlock;
1487 			}
1488 		}
1489 	} else {
1490 		error = gfs2_permission(&init_user_ns, ndir,
1491 					MAY_WRITE | MAY_EXEC);
1492 		if (error)
1493 			goto out_gunlock;
1494 
1495 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1496 		switch (error) {
1497 		case -ENOENT:
1498 			error = 0;
1499 			break;
1500 		case 0:
1501 			error = -EEXIST;
1502 			goto out_gunlock;
1503 		default:
1504 			goto out_gunlock;
1505 		}
1506 
1507 		if (odip != ndip) {
1508 			if (!ndip->i_inode.i_nlink) {
1509 				error = -ENOENT;
1510 				goto out_gunlock;
1511 			}
1512 			if (ndip->i_entries == (u32)-1) {
1513 				error = -EFBIG;
1514 				goto out_gunlock;
1515 			}
1516 			if (S_ISDIR(ip->i_inode.i_mode) &&
1517 			    ndip->i_inode.i_nlink == (u32)-1) {
1518 				error = -EMLINK;
1519 				goto out_gunlock;
1520 			}
1521 		}
1522 	}
1523 
1524 	/* Check out the dir to be renamed */
1525 
1526 	if (dir_rename) {
1527 		error = gfs2_permission(&init_user_ns, d_inode(odentry),
1528 					MAY_WRITE);
1529 		if (error)
1530 			goto out_gunlock;
1531 	}
1532 
1533 	if (nip == NULL) {
1534 		error = gfs2_diradd_alloc_required(ndir, &ndentry->d_name, &da);
1535 		if (error)
1536 			goto out_gunlock;
1537 	}
1538 
1539 	if (da.nr_blocks) {
1540 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1541 		error = gfs2_quota_lock_check(ndip, &ap);
1542 		if (error)
1543 			goto out_gunlock;
1544 
1545 		error = gfs2_inplace_reserve(ndip, &ap);
1546 		if (error)
1547 			goto out_gunlock_q;
1548 
1549 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(ndip, &da, 4) +
1550 					 4 * RES_LEAF + 4, 0);
1551 		if (error)
1552 			goto out_ipreserv;
1553 	} else {
1554 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1555 					 5 * RES_LEAF + 4, 0);
1556 		if (error)
1557 			goto out_gunlock;
1558 	}
1559 
1560 	/* Remove the target file, if it exists */
1561 
1562 	if (nip)
1563 		error = gfs2_unlink_inode(ndip, ndentry);
1564 
1565 	error = update_moved_ino(ip, ndip, dir_rename);
1566 	if (error)
1567 		goto out_end_trans;
1568 
1569 	error = gfs2_dir_del(odip, odentry);
1570 	if (error)
1571 		goto out_end_trans;
1572 
1573 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip, &da);
1574 	if (error)
1575 		goto out_end_trans;
1576 
1577 out_end_trans:
1578 	gfs2_trans_end(sdp);
1579 out_ipreserv:
1580 	if (da.nr_blocks)
1581 		gfs2_inplace_release(ndip);
1582 out_gunlock_q:
1583 	if (da.nr_blocks)
1584 		gfs2_quota_unlock(ndip);
1585 out_gunlock:
1586 	gfs2_dir_no_add(&da);
1587 	if (gfs2_holder_initialized(&rd_gh))
1588 		gfs2_glock_dq_uninit(&rd_gh);
1589 
1590 	while (x--) {
1591 		if (gfs2_holder_queued(ghs + x))
1592 			gfs2_glock_dq(ghs + x);
1593 		gfs2_holder_uninit(ghs + x);
1594 	}
1595 out_gunlock_r:
1596 	if (gfs2_holder_initialized(&r_gh))
1597 		gfs2_glock_dq_uninit(&r_gh);
1598 out:
1599 	gfs2_qa_put(ndip);
1600 	return error;
1601 }
1602 
1603 /**
1604  * gfs2_exchange - exchange two files
1605  * @odir: Parent directory of old file name
1606  * @odentry: The old dentry of the file
1607  * @ndir: Parent directory of new file name
1608  * @ndentry: The new dentry of the file
1609  * @flags: The rename flags
1610  *
1611  * Returns: errno
1612  */
1613 
1614 static int gfs2_exchange(struct inode *odir, struct dentry *odentry,
1615 			 struct inode *ndir, struct dentry *ndentry,
1616 			 unsigned int flags)
1617 {
1618 	struct gfs2_inode *odip = GFS2_I(odir);
1619 	struct gfs2_inode *ndip = GFS2_I(ndir);
1620 	struct gfs2_inode *oip = GFS2_I(odentry->d_inode);
1621 	struct gfs2_inode *nip = GFS2_I(ndentry->d_inode);
1622 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1623 	struct gfs2_holder ghs[4], r_gh;
1624 	unsigned int num_gh;
1625 	unsigned int x;
1626 	umode_t old_mode = oip->i_inode.i_mode;
1627 	umode_t new_mode = nip->i_inode.i_mode;
1628 	int error;
1629 
1630 	gfs2_holder_mark_uninitialized(&r_gh);
1631 	error = gfs2_rindex_update(sdp);
1632 	if (error)
1633 		return error;
1634 
1635 	if (odip != ndip) {
1636 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1637 					   0, &r_gh);
1638 		if (error)
1639 			goto out;
1640 
1641 		if (S_ISDIR(old_mode)) {
1642 			/* don't move a directory into its subdir */
1643 			error = gfs2_ok_to_move(oip, ndip);
1644 			if (error)
1645 				goto out_gunlock_r;
1646 		}
1647 
1648 		if (S_ISDIR(new_mode)) {
1649 			/* don't move a directory into its subdir */
1650 			error = gfs2_ok_to_move(nip, odip);
1651 			if (error)
1652 				goto out_gunlock_r;
1653 		}
1654 	}
1655 
1656 	num_gh = 1;
1657 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs);
1658 	if (odip != ndip) {
1659 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC,
1660 				 ghs + num_gh);
1661 		num_gh++;
1662 	}
1663 	gfs2_holder_init(oip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1664 	num_gh++;
1665 
1666 	gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1667 	num_gh++;
1668 
1669 	for (x = 0; x < num_gh; x++) {
1670 		error = gfs2_glock_nq(ghs + x);
1671 		if (error)
1672 			goto out_gunlock;
1673 	}
1674 
1675 	error = gfs2_glock_async_wait(num_gh, ghs);
1676 	if (error)
1677 		goto out_gunlock;
1678 
1679 	error = -ENOENT;
1680 	if (oip->i_inode.i_nlink == 0 || nip->i_inode.i_nlink == 0)
1681 		goto out_gunlock;
1682 
1683 	error = gfs2_unlink_ok(odip, &odentry->d_name, oip);
1684 	if (error)
1685 		goto out_gunlock;
1686 	error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1687 	if (error)
1688 		goto out_gunlock;
1689 
1690 	if (S_ISDIR(old_mode)) {
1691 		error = gfs2_permission(&init_user_ns, odentry->d_inode,
1692 					MAY_WRITE);
1693 		if (error)
1694 			goto out_gunlock;
1695 	}
1696 	if (S_ISDIR(new_mode)) {
1697 		error = gfs2_permission(&init_user_ns, ndentry->d_inode,
1698 					MAY_WRITE);
1699 		if (error)
1700 			goto out_gunlock;
1701 	}
1702 	error = gfs2_trans_begin(sdp, 4 * RES_DINODE + 4 * RES_LEAF, 0);
1703 	if (error)
1704 		goto out_gunlock;
1705 
1706 	error = update_moved_ino(oip, ndip, S_ISDIR(old_mode));
1707 	if (error)
1708 		goto out_end_trans;
1709 
1710 	error = update_moved_ino(nip, odip, S_ISDIR(new_mode));
1711 	if (error)
1712 		goto out_end_trans;
1713 
1714 	error = gfs2_dir_mvino(ndip, &ndentry->d_name, oip,
1715 			       IF2DT(old_mode));
1716 	if (error)
1717 		goto out_end_trans;
1718 
1719 	error = gfs2_dir_mvino(odip, &odentry->d_name, nip,
1720 			       IF2DT(new_mode));
1721 	if (error)
1722 		goto out_end_trans;
1723 
1724 	if (odip != ndip) {
1725 		if (S_ISDIR(new_mode) && !S_ISDIR(old_mode)) {
1726 			inc_nlink(&odip->i_inode);
1727 			drop_nlink(&ndip->i_inode);
1728 		} else if (S_ISDIR(old_mode) && !S_ISDIR(new_mode)) {
1729 			inc_nlink(&ndip->i_inode);
1730 			drop_nlink(&odip->i_inode);
1731 		}
1732 	}
1733 	mark_inode_dirty(&ndip->i_inode);
1734 	if (odip != ndip)
1735 		mark_inode_dirty(&odip->i_inode);
1736 
1737 out_end_trans:
1738 	gfs2_trans_end(sdp);
1739 out_gunlock:
1740 	while (x--) {
1741 		if (gfs2_holder_queued(ghs + x))
1742 			gfs2_glock_dq(ghs + x);
1743 		gfs2_holder_uninit(ghs + x);
1744 	}
1745 out_gunlock_r:
1746 	if (gfs2_holder_initialized(&r_gh))
1747 		gfs2_glock_dq_uninit(&r_gh);
1748 out:
1749 	return error;
1750 }
1751 
1752 static int gfs2_rename2(struct user_namespace *mnt_userns, struct inode *odir,
1753 			struct dentry *odentry, struct inode *ndir,
1754 			struct dentry *ndentry, unsigned int flags)
1755 {
1756 	flags &= ~RENAME_NOREPLACE;
1757 
1758 	if (flags & ~RENAME_EXCHANGE)
1759 		return -EINVAL;
1760 
1761 	if (flags & RENAME_EXCHANGE)
1762 		return gfs2_exchange(odir, odentry, ndir, ndentry, flags);
1763 
1764 	return gfs2_rename(odir, odentry, ndir, ndentry);
1765 }
1766 
1767 /**
1768  * gfs2_get_link - Follow a symbolic link
1769  * @dentry: The dentry of the link
1770  * @inode: The inode of the link
1771  * @done: destructor for return value
1772  *
1773  * This can handle symlinks of any size.
1774  *
1775  * Returns: 0 on success or error code
1776  */
1777 
1778 static const char *gfs2_get_link(struct dentry *dentry,
1779 				 struct inode *inode,
1780 				 struct delayed_call *done)
1781 {
1782 	struct gfs2_inode *ip = GFS2_I(inode);
1783 	struct gfs2_holder i_gh;
1784 	struct buffer_head *dibh;
1785 	unsigned int size;
1786 	char *buf;
1787 	int error;
1788 
1789 	if (!dentry)
1790 		return ERR_PTR(-ECHILD);
1791 
1792 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1793 	error = gfs2_glock_nq(&i_gh);
1794 	if (error) {
1795 		gfs2_holder_uninit(&i_gh);
1796 		return ERR_PTR(error);
1797 	}
1798 
1799 	size = (unsigned int)i_size_read(&ip->i_inode);
1800 	if (size == 0) {
1801 		gfs2_consist_inode(ip);
1802 		buf = ERR_PTR(-EIO);
1803 		goto out;
1804 	}
1805 
1806 	error = gfs2_meta_inode_buffer(ip, &dibh);
1807 	if (error) {
1808 		buf = ERR_PTR(error);
1809 		goto out;
1810 	}
1811 
1812 	buf = kzalloc(size + 1, GFP_NOFS);
1813 	if (!buf)
1814 		buf = ERR_PTR(-ENOMEM);
1815 	else
1816 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1817 	brelse(dibh);
1818 out:
1819 	gfs2_glock_dq_uninit(&i_gh);
1820 	if (!IS_ERR(buf))
1821 		set_delayed_call(done, kfree_link, buf);
1822 	return buf;
1823 }
1824 
1825 /**
1826  * gfs2_permission
1827  * @mnt_userns: User namespace of the mount the inode was found from
1828  * @inode: The inode
1829  * @mask: The mask to be tested
1830  *
1831  * This may be called from the VFS directly, or from within GFS2 with the
1832  * inode locked, so we look to see if the glock is already locked and only
1833  * lock the glock if its not already been done.
1834  *
1835  * Returns: errno
1836  */
1837 
1838 int gfs2_permission(struct user_namespace *mnt_userns, struct inode *inode,
1839 		    int mask)
1840 {
1841 	struct gfs2_inode *ip;
1842 	struct gfs2_holder i_gh;
1843 	int error;
1844 
1845 	gfs2_holder_mark_uninitialized(&i_gh);
1846 	ip = GFS2_I(inode);
1847 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1848 		if (mask & MAY_NOT_BLOCK)
1849 			return -ECHILD;
1850 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1851 		if (error)
1852 			return error;
1853 	}
1854 
1855 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1856 		error = -EPERM;
1857 	else
1858 		error = generic_permission(&init_user_ns, inode, mask);
1859 	if (gfs2_holder_initialized(&i_gh))
1860 		gfs2_glock_dq_uninit(&i_gh);
1861 
1862 	return error;
1863 }
1864 
1865 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1866 {
1867 	setattr_copy(&init_user_ns, inode, attr);
1868 	mark_inode_dirty(inode);
1869 	return 0;
1870 }
1871 
1872 static int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1873 {
1874 	int error;
1875 
1876 	if (current->journal_info)
1877 		return __gfs2_setattr_simple(inode, attr);
1878 
1879 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1880 	if (error)
1881 		return error;
1882 
1883 	error = __gfs2_setattr_simple(inode, attr);
1884 	gfs2_trans_end(GFS2_SB(inode));
1885 	return error;
1886 }
1887 
1888 static int setattr_chown(struct inode *inode, struct iattr *attr)
1889 {
1890 	struct gfs2_inode *ip = GFS2_I(inode);
1891 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1892 	kuid_t ouid, nuid;
1893 	kgid_t ogid, ngid;
1894 	int error;
1895 	struct gfs2_alloc_parms ap;
1896 
1897 	ouid = inode->i_uid;
1898 	ogid = inode->i_gid;
1899 	nuid = attr->ia_uid;
1900 	ngid = attr->ia_gid;
1901 
1902 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1903 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1904 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1905 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1906 	error = gfs2_qa_get(ip);
1907 	if (error)
1908 		return error;
1909 
1910 	error = gfs2_rindex_update(sdp);
1911 	if (error)
1912 		goto out;
1913 
1914 	error = gfs2_quota_lock(ip, nuid, ngid);
1915 	if (error)
1916 		goto out;
1917 
1918 	ap.target = gfs2_get_inode_blocks(&ip->i_inode);
1919 
1920 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1921 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1922 		error = gfs2_quota_check(ip, nuid, ngid, &ap);
1923 		if (error)
1924 			goto out_gunlock_q;
1925 	}
1926 
1927 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1928 	if (error)
1929 		goto out_gunlock_q;
1930 
1931 	error = gfs2_setattr_simple(inode, attr);
1932 	if (error)
1933 		goto out_end_trans;
1934 
1935 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1936 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1937 		gfs2_quota_change(ip, -(s64)ap.target, ouid, ogid);
1938 		gfs2_quota_change(ip, ap.target, nuid, ngid);
1939 	}
1940 
1941 out_end_trans:
1942 	gfs2_trans_end(sdp);
1943 out_gunlock_q:
1944 	gfs2_quota_unlock(ip);
1945 out:
1946 	gfs2_qa_put(ip);
1947 	return error;
1948 }
1949 
1950 /**
1951  * gfs2_setattr - Change attributes on an inode
1952  * @mnt_userns: User namespace of the mount the inode was found from
1953  * @dentry: The dentry which is changing
1954  * @attr: The structure describing the change
1955  *
1956  * The VFS layer wants to change one or more of an inodes attributes.  Write
1957  * that change out to disk.
1958  *
1959  * Returns: errno
1960  */
1961 
1962 static int gfs2_setattr(struct user_namespace *mnt_userns,
1963 			struct dentry *dentry, struct iattr *attr)
1964 {
1965 	struct inode *inode = d_inode(dentry);
1966 	struct gfs2_inode *ip = GFS2_I(inode);
1967 	struct gfs2_holder i_gh;
1968 	int error;
1969 
1970 	error = gfs2_qa_get(ip);
1971 	if (error)
1972 		return error;
1973 
1974 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1975 	if (error)
1976 		goto out;
1977 
1978 	error = may_setattr(&init_user_ns, inode, attr->ia_valid);
1979 	if (error)
1980 		goto error;
1981 
1982 	error = setattr_prepare(&init_user_ns, dentry, attr);
1983 	if (error)
1984 		goto error;
1985 
1986 	if (attr->ia_valid & ATTR_SIZE)
1987 		error = gfs2_setattr_size(inode, attr->ia_size);
1988 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1989 		error = setattr_chown(inode, attr);
1990 	else {
1991 		error = gfs2_setattr_simple(inode, attr);
1992 		if (!error && attr->ia_valid & ATTR_MODE)
1993 			error = posix_acl_chmod(&init_user_ns, inode,
1994 						inode->i_mode);
1995 	}
1996 
1997 error:
1998 	if (!error)
1999 		mark_inode_dirty(inode);
2000 	gfs2_glock_dq_uninit(&i_gh);
2001 out:
2002 	gfs2_qa_put(ip);
2003 	return error;
2004 }
2005 
2006 /**
2007  * gfs2_getattr - Read out an inode's attributes
2008  * @mnt_userns:	user namespace of the mount the inode was found from
2009  * @path: Object to query
2010  * @stat: The inode's stats
2011  * @request_mask: Mask of STATX_xxx flags indicating the caller's interests
2012  * @flags: AT_STATX_xxx setting
2013  *
2014  * This may be called from the VFS directly, or from within GFS2 with the
2015  * inode locked, so we look to see if the glock is already locked and only
2016  * lock the glock if its not already been done. Note that its the NFS
2017  * readdirplus operation which causes this to be called (from filldir)
2018  * with the glock already held.
2019  *
2020  * Returns: errno
2021  */
2022 
2023 static int gfs2_getattr(struct user_namespace *mnt_userns,
2024 			const struct path *path, struct kstat *stat,
2025 			u32 request_mask, unsigned int flags)
2026 {
2027 	struct inode *inode = d_inode(path->dentry);
2028 	struct gfs2_inode *ip = GFS2_I(inode);
2029 	struct gfs2_holder gh;
2030 	u32 gfsflags;
2031 	int error;
2032 
2033 	gfs2_holder_mark_uninitialized(&gh);
2034 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
2035 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
2036 		if (error)
2037 			return error;
2038 	}
2039 
2040 	gfsflags = ip->i_diskflags;
2041 	if (gfsflags & GFS2_DIF_APPENDONLY)
2042 		stat->attributes |= STATX_ATTR_APPEND;
2043 	if (gfsflags & GFS2_DIF_IMMUTABLE)
2044 		stat->attributes |= STATX_ATTR_IMMUTABLE;
2045 
2046 	stat->attributes_mask |= (STATX_ATTR_APPEND |
2047 				  STATX_ATTR_COMPRESSED |
2048 				  STATX_ATTR_ENCRYPTED |
2049 				  STATX_ATTR_IMMUTABLE |
2050 				  STATX_ATTR_NODUMP);
2051 
2052 	generic_fillattr(&init_user_ns, inode, stat);
2053 
2054 	if (gfs2_holder_initialized(&gh))
2055 		gfs2_glock_dq_uninit(&gh);
2056 
2057 	return 0;
2058 }
2059 
2060 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2061 		       u64 start, u64 len)
2062 {
2063 	struct gfs2_inode *ip = GFS2_I(inode);
2064 	struct gfs2_holder gh;
2065 	int ret;
2066 
2067 	inode_lock_shared(inode);
2068 
2069 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
2070 	if (ret)
2071 		goto out;
2072 
2073 	ret = iomap_fiemap(inode, fieinfo, start, len, &gfs2_iomap_ops);
2074 
2075 	gfs2_glock_dq_uninit(&gh);
2076 
2077 out:
2078 	inode_unlock_shared(inode);
2079 	return ret;
2080 }
2081 
2082 loff_t gfs2_seek_data(struct file *file, loff_t offset)
2083 {
2084 	struct inode *inode = file->f_mapping->host;
2085 	struct gfs2_inode *ip = GFS2_I(inode);
2086 	struct gfs2_holder gh;
2087 	loff_t ret;
2088 
2089 	inode_lock_shared(inode);
2090 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
2091 	if (!ret)
2092 		ret = iomap_seek_data(inode, offset, &gfs2_iomap_ops);
2093 	gfs2_glock_dq_uninit(&gh);
2094 	inode_unlock_shared(inode);
2095 
2096 	if (ret < 0)
2097 		return ret;
2098 	return vfs_setpos(file, ret, inode->i_sb->s_maxbytes);
2099 }
2100 
2101 loff_t gfs2_seek_hole(struct file *file, loff_t offset)
2102 {
2103 	struct inode *inode = file->f_mapping->host;
2104 	struct gfs2_inode *ip = GFS2_I(inode);
2105 	struct gfs2_holder gh;
2106 	loff_t ret;
2107 
2108 	inode_lock_shared(inode);
2109 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
2110 	if (!ret)
2111 		ret = iomap_seek_hole(inode, offset, &gfs2_iomap_ops);
2112 	gfs2_glock_dq_uninit(&gh);
2113 	inode_unlock_shared(inode);
2114 
2115 	if (ret < 0)
2116 		return ret;
2117 	return vfs_setpos(file, ret, inode->i_sb->s_maxbytes);
2118 }
2119 
2120 static int gfs2_update_time(struct inode *inode, struct timespec64 *time,
2121 			    int flags)
2122 {
2123 	struct gfs2_inode *ip = GFS2_I(inode);
2124 	struct gfs2_glock *gl = ip->i_gl;
2125 	struct gfs2_holder *gh;
2126 	int error;
2127 
2128 	gh = gfs2_glock_is_locked_by_me(gl);
2129 	if (gh && !gfs2_glock_is_held_excl(gl)) {
2130 		gfs2_glock_dq(gh);
2131 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, gh);
2132 		error = gfs2_glock_nq(gh);
2133 		if (error)
2134 			return error;
2135 	}
2136 	return generic_update_time(inode, time, flags);
2137 }
2138 
2139 static const struct inode_operations gfs2_file_iops = {
2140 	.permission = gfs2_permission,
2141 	.setattr = gfs2_setattr,
2142 	.getattr = gfs2_getattr,
2143 	.listxattr = gfs2_listxattr,
2144 	.fiemap = gfs2_fiemap,
2145 	.get_acl = gfs2_get_acl,
2146 	.set_acl = gfs2_set_acl,
2147 	.update_time = gfs2_update_time,
2148 	.fileattr_get = gfs2_fileattr_get,
2149 	.fileattr_set = gfs2_fileattr_set,
2150 };
2151 
2152 static const struct inode_operations gfs2_dir_iops = {
2153 	.create = gfs2_create,
2154 	.lookup = gfs2_lookup,
2155 	.link = gfs2_link,
2156 	.unlink = gfs2_unlink,
2157 	.symlink = gfs2_symlink,
2158 	.mkdir = gfs2_mkdir,
2159 	.rmdir = gfs2_unlink,
2160 	.mknod = gfs2_mknod,
2161 	.rename = gfs2_rename2,
2162 	.permission = gfs2_permission,
2163 	.setattr = gfs2_setattr,
2164 	.getattr = gfs2_getattr,
2165 	.listxattr = gfs2_listxattr,
2166 	.fiemap = gfs2_fiemap,
2167 	.get_acl = gfs2_get_acl,
2168 	.set_acl = gfs2_set_acl,
2169 	.update_time = gfs2_update_time,
2170 	.atomic_open = gfs2_atomic_open,
2171 	.fileattr_get = gfs2_fileattr_get,
2172 	.fileattr_set = gfs2_fileattr_set,
2173 };
2174 
2175 static const struct inode_operations gfs2_symlink_iops = {
2176 	.get_link = gfs2_get_link,
2177 	.permission = gfs2_permission,
2178 	.setattr = gfs2_setattr,
2179 	.getattr = gfs2_getattr,
2180 	.listxattr = gfs2_listxattr,
2181 	.fiemap = gfs2_fiemap,
2182 };
2183 
2184