xref: /openbmc/linux/fs/gfs2/inode.c (revision a531b0c2)
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, &ip->i_gl);
135 		if (unlikely(error))
136 			goto fail;
137 
138 		if (type == DT_UNKNOWN || blktype != GFS2_BLKST_FREE) {
139 			/*
140 			 * The GL_SKIP flag indicates to skip reading the inode
141 			 * block.  We read the inode when instantiating it
142 			 * after possibly checking the block type.
143 			 */
144 			error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE,
145 						   GL_SKIP, &i_gh);
146 			if (error)
147 				goto fail;
148 
149 			error = -ESTALE;
150 			if (no_formal_ino &&
151 			    gfs2_inode_already_deleted(ip->i_gl, no_formal_ino))
152 				goto fail;
153 
154 			if (blktype != GFS2_BLKST_FREE) {
155 				error = gfs2_check_blk_type(sdp, no_addr,
156 							    blktype);
157 				if (error)
158 					goto fail;
159 			}
160 		}
161 
162 		set_bit(GLF_INSTANTIATE_NEEDED, &ip->i_gl->gl_flags);
163 
164 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
165 		if (unlikely(error))
166 			goto fail;
167 		if (blktype != GFS2_BLKST_UNLINKED)
168 			gfs2_cancel_delete_work(io_gl);
169 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
170 		gfs2_glock_put(io_gl);
171 		if (unlikely(error))
172 			goto fail;
173 
174 		/* Lowest possible timestamp; will be overwritten in gfs2_dinode_in. */
175 		inode->i_atime.tv_sec = 1LL << (8 * sizeof(inode->i_atime.tv_sec) - 1);
176 		inode->i_atime.tv_nsec = 0;
177 
178 		glock_set_object(ip->i_gl, ip);
179 
180 		if (type == DT_UNKNOWN) {
181 			/* Inode glock must be locked already */
182 			error = gfs2_instantiate(&i_gh);
183 			if (error) {
184 				glock_clear_object(ip->i_gl, ip);
185 				goto fail;
186 			}
187 		} else {
188 			ip->i_no_formal_ino = no_formal_ino;
189 			inode->i_mode = DT2IF(type);
190 		}
191 
192 		if (gfs2_holder_initialized(&i_gh))
193 			gfs2_glock_dq_uninit(&i_gh);
194 		glock_set_object(ip->i_iopen_gh.gh_gl, ip);
195 
196 		gfs2_set_iop(inode);
197 		unlock_new_inode(inode);
198 	}
199 
200 	if (no_formal_ino && ip->i_no_formal_ino &&
201 	    no_formal_ino != ip->i_no_formal_ino) {
202 		iput(inode);
203 		return ERR_PTR(-ESTALE);
204 	}
205 
206 	return inode;
207 
208 fail:
209 	if (gfs2_holder_initialized(&ip->i_iopen_gh))
210 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
211 	if (gfs2_holder_initialized(&i_gh))
212 		gfs2_glock_dq_uninit(&i_gh);
213 	iget_failed(inode);
214 	return ERR_PTR(error);
215 }
216 
217 /**
218  * gfs2_lookup_by_inum - look up an inode by inode number
219  * @sdp: The super block
220  * @no_addr: The inode number
221  * @no_formal_ino: The inode generation number (0 for any)
222  * @blktype: Requested block type (see gfs2_inode_lookup)
223  */
224 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
225 				  u64 no_formal_ino, unsigned int blktype)
226 {
227 	struct super_block *sb = sdp->sd_vfs;
228 	struct inode *inode;
229 	int error;
230 
231 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, no_formal_ino,
232 				  blktype);
233 	if (IS_ERR(inode))
234 		return inode;
235 
236 	if (no_formal_ino) {
237 		error = -EIO;
238 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
239 			goto fail_iput;
240 	}
241 	return inode;
242 
243 fail_iput:
244 	iput(inode);
245 	return ERR_PTR(error);
246 }
247 
248 
249 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
250 {
251 	struct qstr qstr;
252 	struct inode *inode;
253 	gfs2_str2qstr(&qstr, name);
254 	inode = gfs2_lookupi(dip, &qstr, 1);
255 	/* gfs2_lookupi has inconsistent callers: vfs
256 	 * related routines expect NULL for no entry found,
257 	 * gfs2_lookup_simple callers expect ENOENT
258 	 * and do not check for NULL.
259 	 */
260 	if (inode == NULL)
261 		return ERR_PTR(-ENOENT);
262 	else
263 		return inode;
264 }
265 
266 
267 /**
268  * gfs2_lookupi - Look up a filename in a directory and return its inode
269  * @dir: The inode of the directory containing the inode to look-up
270  * @name: The name of the inode to look for
271  * @is_root: If 1, ignore the caller's permissions
272  *
273  * This can be called via the VFS filldir function when NFS is doing
274  * a readdirplus and the inode which its intending to stat isn't
275  * already in cache. In this case we must not take the directory glock
276  * again, since the readdir call will have already taken that lock.
277  *
278  * Returns: errno
279  */
280 
281 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
282 			   int is_root)
283 {
284 	struct super_block *sb = dir->i_sb;
285 	struct gfs2_inode *dip = GFS2_I(dir);
286 	struct gfs2_holder d_gh;
287 	int error = 0;
288 	struct inode *inode = NULL;
289 
290 	gfs2_holder_mark_uninitialized(&d_gh);
291 	if (!name->len || name->len > GFS2_FNAMESIZE)
292 		return ERR_PTR(-ENAMETOOLONG);
293 
294 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
295 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
296 	     dir == d_inode(sb->s_root))) {
297 		igrab(dir);
298 		return dir;
299 	}
300 
301 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
302 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
303 		if (error)
304 			return ERR_PTR(error);
305 	}
306 
307 	if (!is_root) {
308 		error = gfs2_permission(&init_user_ns, dir, MAY_EXEC);
309 		if (error)
310 			goto out;
311 	}
312 
313 	inode = gfs2_dir_search(dir, name, false);
314 	if (IS_ERR(inode))
315 		error = PTR_ERR(inode);
316 out:
317 	if (gfs2_holder_initialized(&d_gh))
318 		gfs2_glock_dq_uninit(&d_gh);
319 	if (error == -ENOENT)
320 		return NULL;
321 	return inode ? inode : ERR_PTR(error);
322 }
323 
324 /**
325  * create_ok - OK to create a new on-disk inode here?
326  * @dip:  Directory in which dinode is to be created
327  * @name:  Name of new dinode
328  * @mode:
329  *
330  * Returns: errno
331  */
332 
333 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
334 		     umode_t mode)
335 {
336 	int error;
337 
338 	error = gfs2_permission(&init_user_ns, &dip->i_inode,
339 				MAY_WRITE | MAY_EXEC);
340 	if (error)
341 		return error;
342 
343 	/*  Don't create entries in an unlinked directory  */
344 	if (!dip->i_inode.i_nlink)
345 		return -ENOENT;
346 
347 	if (dip->i_entries == (u32)-1)
348 		return -EFBIG;
349 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
350 		return -EMLINK;
351 
352 	return 0;
353 }
354 
355 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
356 			       struct inode *inode)
357 {
358 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
359 	    (dip->i_inode.i_mode & S_ISUID) &&
360 	    !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
361 		if (S_ISDIR(inode->i_mode))
362 			inode->i_mode |= S_ISUID;
363 		else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
364 			inode->i_mode &= ~07111;
365 		inode->i_uid = dip->i_inode.i_uid;
366 	} else
367 		inode->i_uid = current_fsuid();
368 
369 	if (dip->i_inode.i_mode & S_ISGID) {
370 		if (S_ISDIR(inode->i_mode))
371 			inode->i_mode |= S_ISGID;
372 		inode->i_gid = dip->i_inode.i_gid;
373 	} else
374 		inode->i_gid = current_fsgid();
375 }
376 
377 static int alloc_dinode(struct gfs2_inode *ip, u32 flags, unsigned *dblocks)
378 {
379 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
380 	struct gfs2_alloc_parms ap = { .target = *dblocks, .aflags = flags, };
381 	int error;
382 
383 	error = gfs2_quota_lock_check(ip, &ap);
384 	if (error)
385 		goto out;
386 
387 	error = gfs2_inplace_reserve(ip, &ap);
388 	if (error)
389 		goto out_quota;
390 
391 	error = gfs2_trans_begin(sdp, (*dblocks * RES_RG_BIT) + RES_STATFS + RES_QUOTA, 0);
392 	if (error)
393 		goto out_ipreserv;
394 
395 	error = gfs2_alloc_blocks(ip, &ip->i_no_addr, dblocks, 1, &ip->i_generation);
396 	ip->i_no_formal_ino = ip->i_generation;
397 	ip->i_inode.i_ino = ip->i_no_addr;
398 	ip->i_goal = ip->i_no_addr;
399 
400 	gfs2_trans_end(sdp);
401 
402 out_ipreserv:
403 	gfs2_inplace_release(ip);
404 out_quota:
405 	gfs2_quota_unlock(ip);
406 out:
407 	return error;
408 }
409 
410 static void gfs2_init_dir(struct buffer_head *dibh,
411 			  const struct gfs2_inode *parent)
412 {
413 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
414 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
415 
416 	gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
417 	dent->de_inum = di->di_num; /* already GFS2 endian */
418 	dent->de_type = cpu_to_be16(DT_DIR);
419 
420 	dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
421 	gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
422 	gfs2_inum_out(parent, dent);
423 	dent->de_type = cpu_to_be16(DT_DIR);
424 
425 }
426 
427 /**
428  * gfs2_init_xattr - Initialise an xattr block for a new inode
429  * @ip: The inode in question
430  *
431  * This sets up an empty xattr block for a new inode, ready to
432  * take any ACLs, LSM xattrs, etc.
433  */
434 
435 static void gfs2_init_xattr(struct gfs2_inode *ip)
436 {
437 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
438 	struct buffer_head *bh;
439 	struct gfs2_ea_header *ea;
440 
441 	bh = gfs2_meta_new(ip->i_gl, ip->i_eattr);
442 	gfs2_trans_add_meta(ip->i_gl, bh);
443 	gfs2_metatype_set(bh, GFS2_METATYPE_EA, GFS2_FORMAT_EA);
444 	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
445 
446 	ea = GFS2_EA_BH2FIRST(bh);
447 	ea->ea_rec_len = cpu_to_be32(sdp->sd_jbsize);
448 	ea->ea_type = GFS2_EATYPE_UNUSED;
449 	ea->ea_flags = GFS2_EAFLAG_LAST;
450 
451 	brelse(bh);
452 }
453 
454 /**
455  * init_dinode - Fill in a new dinode structure
456  * @dip: The directory this inode is being created in
457  * @ip: The inode
458  * @symname: The symlink destination (if a symlink)
459  *
460  */
461 
462 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
463 			const char *symname)
464 {
465 	struct gfs2_dinode *di;
466 	struct buffer_head *dibh;
467 
468 	dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
469 	gfs2_trans_add_meta(ip->i_gl, dibh);
470 	di = (struct gfs2_dinode *)dibh->b_data;
471 	gfs2_dinode_out(ip, di);
472 
473 	di->di_major = cpu_to_be32(imajor(&ip->i_inode));
474 	di->di_minor = cpu_to_be32(iminor(&ip->i_inode));
475 	di->__pad1 = 0;
476 	di->__pad2 = 0;
477 	di->__pad3 = 0;
478 	memset(&di->__pad4, 0, sizeof(di->__pad4));
479 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
480 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
481 
482 	switch(ip->i_inode.i_mode & S_IFMT) {
483 	case S_IFDIR:
484 		gfs2_init_dir(dibh, dip);
485 		break;
486 	case S_IFLNK:
487 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
488 		break;
489 	}
490 
491 	set_buffer_uptodate(dibh);
492 	brelse(dibh);
493 }
494 
495 /**
496  * gfs2_trans_da_blks - Calculate number of blocks to link inode
497  * @dip: The directory we are linking into
498  * @da: The dir add information
499  * @nr_inodes: The number of inodes involved
500  *
501  * This calculate the number of blocks we need to reserve in a
502  * transaction to link @nr_inodes into a directory. In most cases
503  * @nr_inodes will be 2 (the directory plus the inode being linked in)
504  * but in case of rename, 4 may be required.
505  *
506  * Returns: Number of blocks
507  */
508 
509 static unsigned gfs2_trans_da_blks(const struct gfs2_inode *dip,
510 				   const struct gfs2_diradd *da,
511 				   unsigned nr_inodes)
512 {
513 	return da->nr_blocks + gfs2_rg_blocks(dip, da->nr_blocks) +
514 	       (nr_inodes * RES_DINODE) + RES_QUOTA + RES_STATFS;
515 }
516 
517 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
518 		       struct gfs2_inode *ip, struct gfs2_diradd *da)
519 {
520 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
521 	struct gfs2_alloc_parms ap = { .target = da->nr_blocks, };
522 	int error;
523 
524 	if (da->nr_blocks) {
525 		error = gfs2_quota_lock_check(dip, &ap);
526 		if (error)
527 			goto fail_quota_locks;
528 
529 		error = gfs2_inplace_reserve(dip, &ap);
530 		if (error)
531 			goto fail_quota_locks;
532 
533 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, da, 2), 0);
534 		if (error)
535 			goto fail_ipreserv;
536 	} else {
537 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
538 		if (error)
539 			goto fail_quota_locks;
540 	}
541 
542 	error = gfs2_dir_add(&dip->i_inode, name, ip, da);
543 
544 	gfs2_trans_end(sdp);
545 fail_ipreserv:
546 	gfs2_inplace_release(dip);
547 fail_quota_locks:
548 	gfs2_quota_unlock(dip);
549 	return error;
550 }
551 
552 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
553 		    void *fs_info)
554 {
555 	const struct xattr *xattr;
556 	int err = 0;
557 
558 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
559 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
560 				       xattr->value_len, 0,
561 				       GFS2_EATYPE_SECURITY);
562 		if (err < 0)
563 			break;
564 	}
565 	return err;
566 }
567 
568 /**
569  * gfs2_create_inode - Create a new inode
570  * @dir: The parent directory
571  * @dentry: The new dentry
572  * @file: If non-NULL, the file which is being opened
573  * @mode: The permissions on the new inode
574  * @dev: For device nodes, this is the device number
575  * @symname: For symlinks, this is the link destination
576  * @size: The initial size of the inode (ignored for directories)
577  * @excl: Force fail if inode exists
578  *
579  * Returns: 0 on success, or error code
580  */
581 
582 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
583 			     struct file *file,
584 			     umode_t mode, dev_t dev, const char *symname,
585 			     unsigned int size, int excl)
586 {
587 	const struct qstr *name = &dentry->d_name;
588 	struct posix_acl *default_acl, *acl;
589 	struct gfs2_holder ghs[2];
590 	struct inode *inode = NULL;
591 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
592 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
593 	struct gfs2_glock *io_gl;
594 	int error, free_vfs_inode = 1;
595 	u32 aflags = 0;
596 	unsigned blocks = 1;
597 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
598 
599 	if (!name->len || name->len > GFS2_FNAMESIZE)
600 		return -ENAMETOOLONG;
601 
602 	error = gfs2_qa_get(dip);
603 	if (error)
604 		return error;
605 
606 	error = gfs2_rindex_update(sdp);
607 	if (error)
608 		goto fail;
609 
610 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
611 	if (error)
612 		goto fail;
613 	gfs2_holder_mark_uninitialized(ghs + 1);
614 
615 	error = create_ok(dip, name, mode);
616 	if (error)
617 		goto fail_gunlock;
618 
619 	inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
620 	error = PTR_ERR(inode);
621 	if (!IS_ERR(inode)) {
622 		if (S_ISDIR(inode->i_mode)) {
623 			iput(inode);
624 			inode = ERR_PTR(-EISDIR);
625 			goto fail_gunlock;
626 		}
627 		d_instantiate(dentry, inode);
628 		error = 0;
629 		if (file) {
630 			if (S_ISREG(inode->i_mode))
631 				error = finish_open(file, dentry, gfs2_open_common);
632 			else
633 				error = finish_no_open(file, NULL);
634 		}
635 		gfs2_glock_dq_uninit(ghs);
636 		goto fail;
637 	} else if (error != -ENOENT) {
638 		goto fail_gunlock;
639 	}
640 
641 	error = gfs2_diradd_alloc_required(dir, name, &da);
642 	if (error < 0)
643 		goto fail_gunlock;
644 
645 	inode = new_inode(sdp->sd_vfs);
646 	error = -ENOMEM;
647 	if (!inode)
648 		goto fail_gunlock;
649 
650 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
651 	if (error)
652 		goto fail_gunlock;
653 
654 	ip = GFS2_I(inode);
655 	error = gfs2_qa_get(ip);
656 	if (error)
657 		goto fail_free_acls;
658 
659 	inode->i_mode = mode;
660 	set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
661 	inode->i_rdev = dev;
662 	inode->i_size = size;
663 	inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
664 	munge_mode_uid_gid(dip, inode);
665 	check_and_update_goal(dip);
666 	ip->i_goal = dip->i_goal;
667 	ip->i_diskflags = 0;
668 	ip->i_eattr = 0;
669 	ip->i_height = 0;
670 	ip->i_depth = 0;
671 	ip->i_entries = 0;
672 	ip->i_no_addr = 0; /* Temporarily zero until real addr is assigned */
673 
674 	switch(mode & S_IFMT) {
675 	case S_IFREG:
676 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
677 		    gfs2_tune_get(sdp, gt_new_files_jdata))
678 			ip->i_diskflags |= GFS2_DIF_JDATA;
679 		gfs2_set_aops(inode);
680 		break;
681 	case S_IFDIR:
682 		ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
683 		ip->i_diskflags |= GFS2_DIF_JDATA;
684 		ip->i_entries = 2;
685 		break;
686 	}
687 
688 	/* Force SYSTEM flag on all files and subdirs of a SYSTEM directory */
689 	if (dip->i_diskflags & GFS2_DIF_SYSTEM)
690 		ip->i_diskflags |= GFS2_DIF_SYSTEM;
691 
692 	gfs2_set_inode_flags(inode);
693 
694 	if ((GFS2_I(d_inode(sdp->sd_root_dir)) == dip) ||
695 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
696 		aflags |= GFS2_AF_ORLOV;
697 
698 	if (default_acl || acl)
699 		blocks++;
700 
701 	error = alloc_dinode(ip, aflags, &blocks);
702 	if (error)
703 		goto fail_free_inode;
704 
705 	gfs2_set_inode_blocks(inode, blocks);
706 
707 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
708 	if (error)
709 		goto fail_free_inode;
710 
711 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
712 	if (error)
713 		goto fail_free_inode;
714 	gfs2_cancel_delete_work(io_gl);
715 
716 	error = insert_inode_locked4(inode, ip->i_no_addr, iget_test, &ip->i_no_addr);
717 	BUG_ON(error);
718 
719 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
720 	if (error)
721 		goto fail_gunlock2;
722 
723 	error = gfs2_trans_begin(sdp, blocks, 0);
724 	if (error)
725 		goto fail_gunlock2;
726 
727 	if (blocks > 1) {
728 		ip->i_eattr = ip->i_no_addr + 1;
729 		gfs2_init_xattr(ip);
730 	}
731 	init_dinode(dip, ip, symname);
732 	gfs2_trans_end(sdp);
733 
734 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
735 	if (error)
736 		goto fail_gunlock2;
737 
738 	glock_set_object(ip->i_gl, ip);
739 	glock_set_object(io_gl, ip);
740 	gfs2_set_iop(inode);
741 
742 	free_vfs_inode = 0; /* After this point, the inode is no longer
743 			       considered free. Any failures need to undo
744 			       the gfs2 structures. */
745 	if (default_acl) {
746 		error = __gfs2_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
747 		if (error)
748 			goto fail_gunlock3;
749 		posix_acl_release(default_acl);
750 		default_acl = NULL;
751 	}
752 	if (acl) {
753 		error = __gfs2_set_acl(inode, acl, ACL_TYPE_ACCESS);
754 		if (error)
755 			goto fail_gunlock3;
756 		posix_acl_release(acl);
757 		acl = NULL;
758 	}
759 
760 	error = security_inode_init_security(&ip->i_inode, &dip->i_inode, name,
761 					     &gfs2_initxattrs, NULL);
762 	if (error)
763 		goto fail_gunlock3;
764 
765 	error = link_dinode(dip, name, ip, &da);
766 	if (error)
767 		goto fail_gunlock3;
768 
769 	mark_inode_dirty(inode);
770 	d_instantiate(dentry, inode);
771 	/* After instantiate, errors should result in evict which will destroy
772 	 * both inode and iopen glocks properly. */
773 	if (file) {
774 		file->f_mode |= FMODE_CREATED;
775 		error = finish_open(file, dentry, gfs2_open_common);
776 	}
777 	gfs2_glock_dq_uninit(ghs);
778 	gfs2_qa_put(ip);
779 	gfs2_glock_dq_uninit(ghs + 1);
780 	gfs2_glock_put(io_gl);
781 	gfs2_qa_put(dip);
782 	unlock_new_inode(inode);
783 	return error;
784 
785 fail_gunlock3:
786 	glock_clear_object(ip->i_gl, ip);
787 	glock_clear_object(io_gl, ip);
788 	gfs2_glock_dq_uninit(&ip->i_iopen_gh);
789 fail_gunlock2:
790 	gfs2_glock_put(io_gl);
791 fail_free_inode:
792 	if (ip->i_gl) {
793 		if (free_vfs_inode) /* else evict will do the put for us */
794 			gfs2_glock_put(ip->i_gl);
795 	}
796 	gfs2_rs_delete(ip, NULL);
797 	gfs2_qa_put(ip);
798 fail_free_acls:
799 	posix_acl_release(default_acl);
800 	posix_acl_release(acl);
801 fail_gunlock:
802 	gfs2_dir_no_add(&da);
803 	gfs2_glock_dq_uninit(ghs);
804 	if (!IS_ERR_OR_NULL(inode)) {
805 		clear_nlink(inode);
806 		if (!free_vfs_inode)
807 			mark_inode_dirty(inode);
808 		set_bit(free_vfs_inode ? GIF_FREE_VFS_INODE : GIF_ALLOC_FAILED,
809 			&GFS2_I(inode)->i_flags);
810 		if (inode->i_state & I_NEW)
811 			iget_failed(inode);
812 		else
813 			iput(inode);
814 	}
815 	if (gfs2_holder_initialized(ghs + 1))
816 		gfs2_glock_dq_uninit(ghs + 1);
817 fail:
818 	gfs2_qa_put(dip);
819 	return error;
820 }
821 
822 /**
823  * gfs2_create - Create a file
824  * @mnt_userns: User namespace of the mount the inode was found from
825  * @dir: The directory in which to create the file
826  * @dentry: The dentry of the new file
827  * @mode: The mode of the new file
828  * @excl: Force fail if inode exists
829  *
830  * Returns: errno
831  */
832 
833 static int gfs2_create(struct user_namespace *mnt_userns, struct inode *dir,
834 		       struct dentry *dentry, umode_t mode, bool excl)
835 {
836 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl);
837 }
838 
839 /**
840  * __gfs2_lookup - Look up a filename in a directory and return its inode
841  * @dir: The directory inode
842  * @dentry: The dentry of the new inode
843  * @file: File to be opened
844  *
845  *
846  * Returns: errno
847  */
848 
849 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
850 				    struct file *file)
851 {
852 	struct inode *inode;
853 	struct dentry *d;
854 	struct gfs2_holder gh;
855 	struct gfs2_glock *gl;
856 	int error;
857 
858 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
859 	if (inode == NULL) {
860 		d_add(dentry, NULL);
861 		return NULL;
862 	}
863 	if (IS_ERR(inode))
864 		return ERR_CAST(inode);
865 
866 	gl = GFS2_I(inode)->i_gl;
867 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
868 	if (error) {
869 		iput(inode);
870 		return ERR_PTR(error);
871 	}
872 
873 	d = d_splice_alias(inode, dentry);
874 	if (IS_ERR(d)) {
875 		gfs2_glock_dq_uninit(&gh);
876 		return d;
877 	}
878 	if (file && S_ISREG(inode->i_mode))
879 		error = finish_open(file, dentry, gfs2_open_common);
880 
881 	gfs2_glock_dq_uninit(&gh);
882 	if (error) {
883 		dput(d);
884 		return ERR_PTR(error);
885 	}
886 	return d;
887 }
888 
889 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
890 				  unsigned flags)
891 {
892 	return __gfs2_lookup(dir, dentry, NULL);
893 }
894 
895 /**
896  * gfs2_link - Link to a file
897  * @old_dentry: The inode to link
898  * @dir: Add link to this directory
899  * @dentry: The name of the link
900  *
901  * Link the inode in "old_dentry" into the directory "dir" with the
902  * name in "dentry".
903  *
904  * Returns: errno
905  */
906 
907 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
908 		     struct dentry *dentry)
909 {
910 	struct gfs2_inode *dip = GFS2_I(dir);
911 	struct gfs2_sbd *sdp = GFS2_SB(dir);
912 	struct inode *inode = d_inode(old_dentry);
913 	struct gfs2_inode *ip = GFS2_I(inode);
914 	struct gfs2_holder ghs[2];
915 	struct buffer_head *dibh;
916 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
917 	int error;
918 
919 	if (S_ISDIR(inode->i_mode))
920 		return -EPERM;
921 
922 	error = gfs2_qa_get(dip);
923 	if (error)
924 		return error;
925 
926 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
927 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
928 
929 	error = gfs2_glock_nq(ghs); /* parent */
930 	if (error)
931 		goto out_parent;
932 
933 	error = gfs2_glock_nq(ghs + 1); /* child */
934 	if (error)
935 		goto out_child;
936 
937 	error = -ENOENT;
938 	if (inode->i_nlink == 0)
939 		goto out_gunlock;
940 
941 	error = gfs2_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
942 	if (error)
943 		goto out_gunlock;
944 
945 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
946 	switch (error) {
947 	case -ENOENT:
948 		break;
949 	case 0:
950 		error = -EEXIST;
951 		goto out_gunlock;
952 	default:
953 		goto out_gunlock;
954 	}
955 
956 	error = -EINVAL;
957 	if (!dip->i_inode.i_nlink)
958 		goto out_gunlock;
959 	error = -EFBIG;
960 	if (dip->i_entries == (u32)-1)
961 		goto out_gunlock;
962 	error = -EPERM;
963 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
964 		goto out_gunlock;
965 	error = -EINVAL;
966 	if (!ip->i_inode.i_nlink)
967 		goto out_gunlock;
968 	error = -EMLINK;
969 	if (ip->i_inode.i_nlink == (u32)-1)
970 		goto out_gunlock;
971 
972 	error = gfs2_diradd_alloc_required(dir, &dentry->d_name, &da);
973 	if (error < 0)
974 		goto out_gunlock;
975 
976 	if (da.nr_blocks) {
977 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
978 		error = gfs2_quota_lock_check(dip, &ap);
979 		if (error)
980 			goto out_gunlock;
981 
982 		error = gfs2_inplace_reserve(dip, &ap);
983 		if (error)
984 			goto out_gunlock_q;
985 
986 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, &da, 2), 0);
987 		if (error)
988 			goto out_ipres;
989 	} else {
990 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
991 		if (error)
992 			goto out_ipres;
993 	}
994 
995 	error = gfs2_meta_inode_buffer(ip, &dibh);
996 	if (error)
997 		goto out_end_trans;
998 
999 	error = gfs2_dir_add(dir, &dentry->d_name, ip, &da);
1000 	if (error)
1001 		goto out_brelse;
1002 
1003 	gfs2_trans_add_meta(ip->i_gl, dibh);
1004 	inc_nlink(&ip->i_inode);
1005 	ip->i_inode.i_ctime = current_time(&ip->i_inode);
1006 	ihold(inode);
1007 	d_instantiate(dentry, inode);
1008 	mark_inode_dirty(inode);
1009 
1010 out_brelse:
1011 	brelse(dibh);
1012 out_end_trans:
1013 	gfs2_trans_end(sdp);
1014 out_ipres:
1015 	if (da.nr_blocks)
1016 		gfs2_inplace_release(dip);
1017 out_gunlock_q:
1018 	if (da.nr_blocks)
1019 		gfs2_quota_unlock(dip);
1020 out_gunlock:
1021 	gfs2_dir_no_add(&da);
1022 	gfs2_glock_dq(ghs + 1);
1023 out_child:
1024 	gfs2_glock_dq(ghs);
1025 out_parent:
1026 	gfs2_qa_put(dip);
1027 	gfs2_holder_uninit(ghs);
1028 	gfs2_holder_uninit(ghs + 1);
1029 	return error;
1030 }
1031 
1032 /*
1033  * gfs2_unlink_ok - check to see that a inode is still in a directory
1034  * @dip: the directory
1035  * @name: the name of the file
1036  * @ip: the inode
1037  *
1038  * Assumes that the lock on (at least) @dip is held.
1039  *
1040  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1041  */
1042 
1043 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
1044 			  const struct gfs2_inode *ip)
1045 {
1046 	int error;
1047 
1048 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
1049 		return -EPERM;
1050 
1051 	if ((dip->i_inode.i_mode & S_ISVTX) &&
1052 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
1053 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
1054 		return -EPERM;
1055 
1056 	if (IS_APPEND(&dip->i_inode))
1057 		return -EPERM;
1058 
1059 	error = gfs2_permission(&init_user_ns, &dip->i_inode,
1060 				MAY_WRITE | MAY_EXEC);
1061 	if (error)
1062 		return error;
1063 
1064 	return gfs2_dir_check(&dip->i_inode, name, ip);
1065 }
1066 
1067 /**
1068  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1069  * @dip: The parent directory
1070  * @dentry: The dentry to unlink
1071  *
1072  * Called with all the locks and in a transaction. This will only be
1073  * called for a directory after it has been checked to ensure it is empty.
1074  *
1075  * Returns: 0 on success, or an error
1076  */
1077 
1078 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1079 			     const struct dentry *dentry)
1080 {
1081 	struct inode *inode = d_inode(dentry);
1082 	struct gfs2_inode *ip = GFS2_I(inode);
1083 	int error;
1084 
1085 	error = gfs2_dir_del(dip, dentry);
1086 	if (error)
1087 		return error;
1088 
1089 	ip->i_entries = 0;
1090 	inode->i_ctime = current_time(inode);
1091 	if (S_ISDIR(inode->i_mode))
1092 		clear_nlink(inode);
1093 	else
1094 		drop_nlink(inode);
1095 	mark_inode_dirty(inode);
1096 	if (inode->i_nlink == 0)
1097 		gfs2_unlink_di(inode);
1098 	return 0;
1099 }
1100 
1101 
1102 /**
1103  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1104  * @dir: The inode of the directory containing the inode to unlink
1105  * @dentry: The file itself
1106  *
1107  * This routine uses the type of the inode as a flag to figure out
1108  * whether this is an unlink or an rmdir.
1109  *
1110  * Returns: errno
1111  */
1112 
1113 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1114 {
1115 	struct gfs2_inode *dip = GFS2_I(dir);
1116 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1117 	struct inode *inode = d_inode(dentry);
1118 	struct gfs2_inode *ip = GFS2_I(inode);
1119 	struct gfs2_holder ghs[3];
1120 	struct gfs2_rgrpd *rgd;
1121 	int error;
1122 
1123 	error = gfs2_rindex_update(sdp);
1124 	if (error)
1125 		return error;
1126 
1127 	error = -EROFS;
1128 
1129 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1130 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1131 
1132 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1133 	if (!rgd)
1134 		goto out_inodes;
1135 
1136 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, LM_FLAG_NODE_SCOPE, ghs + 2);
1137 
1138 
1139 	error = gfs2_glock_nq(ghs); /* parent */
1140 	if (error)
1141 		goto out_parent;
1142 
1143 	error = gfs2_glock_nq(ghs + 1); /* child */
1144 	if (error)
1145 		goto out_child;
1146 
1147 	error = -ENOENT;
1148 	if (inode->i_nlink == 0)
1149 		goto out_rgrp;
1150 
1151 	if (S_ISDIR(inode->i_mode)) {
1152 		error = -ENOTEMPTY;
1153 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1154 			goto out_rgrp;
1155 	}
1156 
1157 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1158 	if (error)
1159 		goto out_rgrp;
1160 
1161 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1162 	if (error)
1163 		goto out_gunlock;
1164 
1165 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1166 	if (error)
1167 		goto out_gunlock;
1168 
1169 	error = gfs2_unlink_inode(dip, dentry);
1170 	gfs2_trans_end(sdp);
1171 
1172 out_gunlock:
1173 	gfs2_glock_dq(ghs + 2);
1174 out_rgrp:
1175 	gfs2_glock_dq(ghs + 1);
1176 out_child:
1177 	gfs2_glock_dq(ghs);
1178 out_parent:
1179 	gfs2_holder_uninit(ghs + 2);
1180 out_inodes:
1181 	gfs2_holder_uninit(ghs + 1);
1182 	gfs2_holder_uninit(ghs);
1183 	return error;
1184 }
1185 
1186 /**
1187  * gfs2_symlink - Create a symlink
1188  * @mnt_userns: User namespace of the mount the inode was found from
1189  * @dir: The directory to create the symlink in
1190  * @dentry: The dentry to put the symlink in
1191  * @symname: The thing which the link points to
1192  *
1193  * Returns: errno
1194  */
1195 
1196 static int gfs2_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1197 			struct dentry *dentry, const char *symname)
1198 {
1199 	unsigned int size;
1200 
1201 	size = strlen(symname);
1202 	if (size >= gfs2_max_stuffed_size(GFS2_I(dir)))
1203 		return -ENAMETOOLONG;
1204 
1205 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1206 }
1207 
1208 /**
1209  * gfs2_mkdir - Make a directory
1210  * @mnt_userns: User namespace of the mount the inode was found from
1211  * @dir: The parent directory of the new one
1212  * @dentry: The dentry of the new directory
1213  * @mode: The mode of the new directory
1214  *
1215  * Returns: errno
1216  */
1217 
1218 static int gfs2_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1219 		      struct dentry *dentry, umode_t mode)
1220 {
1221 	unsigned dsize = gfs2_max_stuffed_size(GFS2_I(dir));
1222 	return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0);
1223 }
1224 
1225 /**
1226  * gfs2_mknod - Make a special file
1227  * @mnt_userns: User namespace of the mount the inode was found from
1228  * @dir: The directory in which the special file will reside
1229  * @dentry: The dentry of the special file
1230  * @mode: The mode of the special file
1231  * @dev: The device specification of the special file
1232  *
1233  */
1234 
1235 static int gfs2_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1236 		      struct dentry *dentry, umode_t mode, dev_t dev)
1237 {
1238 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0);
1239 }
1240 
1241 /**
1242  * gfs2_atomic_open - Atomically open a file
1243  * @dir: The directory
1244  * @dentry: The proposed new entry
1245  * @file: The proposed new struct file
1246  * @flags: open flags
1247  * @mode: File mode
1248  *
1249  * Returns: error code or 0 for success
1250  */
1251 
1252 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1253 			    struct file *file, unsigned flags,
1254 			    umode_t mode)
1255 {
1256 	struct dentry *d;
1257 	bool excl = !!(flags & O_EXCL);
1258 
1259 	if (!d_in_lookup(dentry))
1260 		goto skip_lookup;
1261 
1262 	d = __gfs2_lookup(dir, dentry, file);
1263 	if (IS_ERR(d))
1264 		return PTR_ERR(d);
1265 	if (d != NULL)
1266 		dentry = d;
1267 	if (d_really_is_positive(dentry)) {
1268 		if (!(file->f_mode & FMODE_OPENED))
1269 			return finish_no_open(file, d);
1270 		dput(d);
1271 		return excl && (flags & O_CREAT) ? -EEXIST : 0;
1272 	}
1273 
1274 	BUG_ON(d != NULL);
1275 
1276 skip_lookup:
1277 	if (!(flags & O_CREAT))
1278 		return -ENOENT;
1279 
1280 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl);
1281 }
1282 
1283 /*
1284  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1285  * @this: move this
1286  * @to: to here
1287  *
1288  * Follow @to back to the root and make sure we don't encounter @this
1289  * Assumes we already hold the rename lock.
1290  *
1291  * Returns: errno
1292  */
1293 
1294 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1295 {
1296 	struct inode *dir = &to->i_inode;
1297 	struct super_block *sb = dir->i_sb;
1298 	struct inode *tmp;
1299 	int error = 0;
1300 
1301 	igrab(dir);
1302 
1303 	for (;;) {
1304 		if (dir == &this->i_inode) {
1305 			error = -EINVAL;
1306 			break;
1307 		}
1308 		if (dir == d_inode(sb->s_root)) {
1309 			error = 0;
1310 			break;
1311 		}
1312 
1313 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1314 		if (!tmp) {
1315 			error = -ENOENT;
1316 			break;
1317 		}
1318 		if (IS_ERR(tmp)) {
1319 			error = PTR_ERR(tmp);
1320 			break;
1321 		}
1322 
1323 		iput(dir);
1324 		dir = tmp;
1325 	}
1326 
1327 	iput(dir);
1328 
1329 	return error;
1330 }
1331 
1332 /**
1333  * update_moved_ino - Update an inode that's being moved
1334  * @ip: The inode being moved
1335  * @ndip: The parent directory of the new filename
1336  * @dir_rename: True of ip is a directory
1337  *
1338  * Returns: errno
1339  */
1340 
1341 static int update_moved_ino(struct gfs2_inode *ip, struct gfs2_inode *ndip,
1342 			    int dir_rename)
1343 {
1344 	if (dir_rename)
1345 		return gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1346 
1347 	ip->i_inode.i_ctime = current_time(&ip->i_inode);
1348 	mark_inode_dirty_sync(&ip->i_inode);
1349 	return 0;
1350 }
1351 
1352 
1353 /**
1354  * gfs2_rename - Rename a file
1355  * @odir: Parent directory of old file name
1356  * @odentry: The old dentry of the file
1357  * @ndir: Parent directory of new file name
1358  * @ndentry: The new dentry of the file
1359  *
1360  * Returns: errno
1361  */
1362 
1363 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1364 		       struct inode *ndir, struct dentry *ndentry)
1365 {
1366 	struct gfs2_inode *odip = GFS2_I(odir);
1367 	struct gfs2_inode *ndip = GFS2_I(ndir);
1368 	struct gfs2_inode *ip = GFS2_I(d_inode(odentry));
1369 	struct gfs2_inode *nip = NULL;
1370 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1371 	struct gfs2_holder ghs[4], r_gh, rd_gh;
1372 	struct gfs2_rgrpd *nrgd;
1373 	unsigned int num_gh;
1374 	int dir_rename = 0;
1375 	struct gfs2_diradd da = { .nr_blocks = 0, .save_loc = 0, };
1376 	unsigned int x;
1377 	int error;
1378 
1379 	gfs2_holder_mark_uninitialized(&r_gh);
1380 	gfs2_holder_mark_uninitialized(&rd_gh);
1381 	if (d_really_is_positive(ndentry)) {
1382 		nip = GFS2_I(d_inode(ndentry));
1383 		if (ip == nip)
1384 			return 0;
1385 	}
1386 
1387 	error = gfs2_rindex_update(sdp);
1388 	if (error)
1389 		return error;
1390 
1391 	error = gfs2_qa_get(ndip);
1392 	if (error)
1393 		return error;
1394 
1395 	if (odip != ndip) {
1396 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1397 					   0, &r_gh);
1398 		if (error)
1399 			goto out;
1400 
1401 		if (S_ISDIR(ip->i_inode.i_mode)) {
1402 			dir_rename = 1;
1403 			/* don't move a directory into its subdir */
1404 			error = gfs2_ok_to_move(ip, ndip);
1405 			if (error)
1406 				goto out_gunlock_r;
1407 		}
1408 	}
1409 
1410 	num_gh = 1;
1411 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs);
1412 	if (odip != ndip) {
1413 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE,GL_ASYNC,
1414 				 ghs + num_gh);
1415 		num_gh++;
1416 	}
1417 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1418 	num_gh++;
1419 
1420 	if (nip) {
1421 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC,
1422 				 ghs + num_gh);
1423 		num_gh++;
1424 	}
1425 
1426 	for (x = 0; x < num_gh; x++) {
1427 		error = gfs2_glock_nq(ghs + x);
1428 		if (error)
1429 			goto out_gunlock;
1430 	}
1431 	error = gfs2_glock_async_wait(num_gh, ghs);
1432 	if (error)
1433 		goto out_gunlock;
1434 
1435 	if (nip) {
1436 		/* Grab the resource group glock for unlink flag twiddling.
1437 		 * This is the case where the target dinode already exists
1438 		 * so we unlink before doing the rename.
1439 		 */
1440 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1441 		if (!nrgd) {
1442 			error = -ENOENT;
1443 			goto out_gunlock;
1444 		}
1445 		error = gfs2_glock_nq_init(nrgd->rd_gl, LM_ST_EXCLUSIVE,
1446 					   LM_FLAG_NODE_SCOPE, &rd_gh);
1447 		if (error)
1448 			goto out_gunlock;
1449 	}
1450 
1451 	error = -ENOENT;
1452 	if (ip->i_inode.i_nlink == 0)
1453 		goto out_gunlock;
1454 
1455 	/* Check out the old directory */
1456 
1457 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1458 	if (error)
1459 		goto out_gunlock;
1460 
1461 	/* Check out the new directory */
1462 
1463 	if (nip) {
1464 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1465 		if (error)
1466 			goto out_gunlock;
1467 
1468 		if (nip->i_inode.i_nlink == 0) {
1469 			error = -EAGAIN;
1470 			goto out_gunlock;
1471 		}
1472 
1473 		if (S_ISDIR(nip->i_inode.i_mode)) {
1474 			if (nip->i_entries < 2) {
1475 				gfs2_consist_inode(nip);
1476 				error = -EIO;
1477 				goto out_gunlock;
1478 			}
1479 			if (nip->i_entries > 2) {
1480 				error = -ENOTEMPTY;
1481 				goto out_gunlock;
1482 			}
1483 		}
1484 	} else {
1485 		error = gfs2_permission(&init_user_ns, ndir,
1486 					MAY_WRITE | MAY_EXEC);
1487 		if (error)
1488 			goto out_gunlock;
1489 
1490 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1491 		switch (error) {
1492 		case -ENOENT:
1493 			error = 0;
1494 			break;
1495 		case 0:
1496 			error = -EEXIST;
1497 			goto out_gunlock;
1498 		default:
1499 			goto out_gunlock;
1500 		}
1501 
1502 		if (odip != ndip) {
1503 			if (!ndip->i_inode.i_nlink) {
1504 				error = -ENOENT;
1505 				goto out_gunlock;
1506 			}
1507 			if (ndip->i_entries == (u32)-1) {
1508 				error = -EFBIG;
1509 				goto out_gunlock;
1510 			}
1511 			if (S_ISDIR(ip->i_inode.i_mode) &&
1512 			    ndip->i_inode.i_nlink == (u32)-1) {
1513 				error = -EMLINK;
1514 				goto out_gunlock;
1515 			}
1516 		}
1517 	}
1518 
1519 	/* Check out the dir to be renamed */
1520 
1521 	if (dir_rename) {
1522 		error = gfs2_permission(&init_user_ns, d_inode(odentry),
1523 					MAY_WRITE);
1524 		if (error)
1525 			goto out_gunlock;
1526 	}
1527 
1528 	if (nip == NULL) {
1529 		error = gfs2_diradd_alloc_required(ndir, &ndentry->d_name, &da);
1530 		if (error)
1531 			goto out_gunlock;
1532 	}
1533 
1534 	if (da.nr_blocks) {
1535 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1536 		error = gfs2_quota_lock_check(ndip, &ap);
1537 		if (error)
1538 			goto out_gunlock;
1539 
1540 		error = gfs2_inplace_reserve(ndip, &ap);
1541 		if (error)
1542 			goto out_gunlock_q;
1543 
1544 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(ndip, &da, 4) +
1545 					 4 * RES_LEAF + 4, 0);
1546 		if (error)
1547 			goto out_ipreserv;
1548 	} else {
1549 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1550 					 5 * RES_LEAF + 4, 0);
1551 		if (error)
1552 			goto out_gunlock;
1553 	}
1554 
1555 	/* Remove the target file, if it exists */
1556 
1557 	if (nip)
1558 		error = gfs2_unlink_inode(ndip, ndentry);
1559 
1560 	error = update_moved_ino(ip, ndip, dir_rename);
1561 	if (error)
1562 		goto out_end_trans;
1563 
1564 	error = gfs2_dir_del(odip, odentry);
1565 	if (error)
1566 		goto out_end_trans;
1567 
1568 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip, &da);
1569 	if (error)
1570 		goto out_end_trans;
1571 
1572 out_end_trans:
1573 	gfs2_trans_end(sdp);
1574 out_ipreserv:
1575 	if (da.nr_blocks)
1576 		gfs2_inplace_release(ndip);
1577 out_gunlock_q:
1578 	if (da.nr_blocks)
1579 		gfs2_quota_unlock(ndip);
1580 out_gunlock:
1581 	gfs2_dir_no_add(&da);
1582 	if (gfs2_holder_initialized(&rd_gh))
1583 		gfs2_glock_dq_uninit(&rd_gh);
1584 
1585 	while (x--) {
1586 		if (gfs2_holder_queued(ghs + x))
1587 			gfs2_glock_dq(ghs + x);
1588 		gfs2_holder_uninit(ghs + x);
1589 	}
1590 out_gunlock_r:
1591 	if (gfs2_holder_initialized(&r_gh))
1592 		gfs2_glock_dq_uninit(&r_gh);
1593 out:
1594 	gfs2_qa_put(ndip);
1595 	return error;
1596 }
1597 
1598 /**
1599  * gfs2_exchange - exchange two files
1600  * @odir: Parent directory of old file name
1601  * @odentry: The old dentry of the file
1602  * @ndir: Parent directory of new file name
1603  * @ndentry: The new dentry of the file
1604  * @flags: The rename flags
1605  *
1606  * Returns: errno
1607  */
1608 
1609 static int gfs2_exchange(struct inode *odir, struct dentry *odentry,
1610 			 struct inode *ndir, struct dentry *ndentry,
1611 			 unsigned int flags)
1612 {
1613 	struct gfs2_inode *odip = GFS2_I(odir);
1614 	struct gfs2_inode *ndip = GFS2_I(ndir);
1615 	struct gfs2_inode *oip = GFS2_I(odentry->d_inode);
1616 	struct gfs2_inode *nip = GFS2_I(ndentry->d_inode);
1617 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1618 	struct gfs2_holder ghs[4], r_gh;
1619 	unsigned int num_gh;
1620 	unsigned int x;
1621 	umode_t old_mode = oip->i_inode.i_mode;
1622 	umode_t new_mode = nip->i_inode.i_mode;
1623 	int error;
1624 
1625 	gfs2_holder_mark_uninitialized(&r_gh);
1626 	error = gfs2_rindex_update(sdp);
1627 	if (error)
1628 		return error;
1629 
1630 	if (odip != ndip) {
1631 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1632 					   0, &r_gh);
1633 		if (error)
1634 			goto out;
1635 
1636 		if (S_ISDIR(old_mode)) {
1637 			/* don't move a directory into its subdir */
1638 			error = gfs2_ok_to_move(oip, ndip);
1639 			if (error)
1640 				goto out_gunlock_r;
1641 		}
1642 
1643 		if (S_ISDIR(new_mode)) {
1644 			/* don't move a directory into its subdir */
1645 			error = gfs2_ok_to_move(nip, odip);
1646 			if (error)
1647 				goto out_gunlock_r;
1648 		}
1649 	}
1650 
1651 	num_gh = 1;
1652 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs);
1653 	if (odip != ndip) {
1654 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC,
1655 				 ghs + num_gh);
1656 		num_gh++;
1657 	}
1658 	gfs2_holder_init(oip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1659 	num_gh++;
1660 
1661 	gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1662 	num_gh++;
1663 
1664 	for (x = 0; x < num_gh; x++) {
1665 		error = gfs2_glock_nq(ghs + x);
1666 		if (error)
1667 			goto out_gunlock;
1668 	}
1669 
1670 	error = gfs2_glock_async_wait(num_gh, ghs);
1671 	if (error)
1672 		goto out_gunlock;
1673 
1674 	error = -ENOENT;
1675 	if (oip->i_inode.i_nlink == 0 || nip->i_inode.i_nlink == 0)
1676 		goto out_gunlock;
1677 
1678 	error = gfs2_unlink_ok(odip, &odentry->d_name, oip);
1679 	if (error)
1680 		goto out_gunlock;
1681 	error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1682 	if (error)
1683 		goto out_gunlock;
1684 
1685 	if (S_ISDIR(old_mode)) {
1686 		error = gfs2_permission(&init_user_ns, odentry->d_inode,
1687 					MAY_WRITE);
1688 		if (error)
1689 			goto out_gunlock;
1690 	}
1691 	if (S_ISDIR(new_mode)) {
1692 		error = gfs2_permission(&init_user_ns, ndentry->d_inode,
1693 					MAY_WRITE);
1694 		if (error)
1695 			goto out_gunlock;
1696 	}
1697 	error = gfs2_trans_begin(sdp, 4 * RES_DINODE + 4 * RES_LEAF, 0);
1698 	if (error)
1699 		goto out_gunlock;
1700 
1701 	error = update_moved_ino(oip, ndip, S_ISDIR(old_mode));
1702 	if (error)
1703 		goto out_end_trans;
1704 
1705 	error = update_moved_ino(nip, odip, S_ISDIR(new_mode));
1706 	if (error)
1707 		goto out_end_trans;
1708 
1709 	error = gfs2_dir_mvino(ndip, &ndentry->d_name, oip,
1710 			       IF2DT(old_mode));
1711 	if (error)
1712 		goto out_end_trans;
1713 
1714 	error = gfs2_dir_mvino(odip, &odentry->d_name, nip,
1715 			       IF2DT(new_mode));
1716 	if (error)
1717 		goto out_end_trans;
1718 
1719 	if (odip != ndip) {
1720 		if (S_ISDIR(new_mode) && !S_ISDIR(old_mode)) {
1721 			inc_nlink(&odip->i_inode);
1722 			drop_nlink(&ndip->i_inode);
1723 		} else if (S_ISDIR(old_mode) && !S_ISDIR(new_mode)) {
1724 			inc_nlink(&ndip->i_inode);
1725 			drop_nlink(&odip->i_inode);
1726 		}
1727 	}
1728 	mark_inode_dirty(&ndip->i_inode);
1729 	if (odip != ndip)
1730 		mark_inode_dirty(&odip->i_inode);
1731 
1732 out_end_trans:
1733 	gfs2_trans_end(sdp);
1734 out_gunlock:
1735 	while (x--) {
1736 		if (gfs2_holder_queued(ghs + x))
1737 			gfs2_glock_dq(ghs + x);
1738 		gfs2_holder_uninit(ghs + x);
1739 	}
1740 out_gunlock_r:
1741 	if (gfs2_holder_initialized(&r_gh))
1742 		gfs2_glock_dq_uninit(&r_gh);
1743 out:
1744 	return error;
1745 }
1746 
1747 static int gfs2_rename2(struct user_namespace *mnt_userns, struct inode *odir,
1748 			struct dentry *odentry, struct inode *ndir,
1749 			struct dentry *ndentry, unsigned int flags)
1750 {
1751 	flags &= ~RENAME_NOREPLACE;
1752 
1753 	if (flags & ~RENAME_EXCHANGE)
1754 		return -EINVAL;
1755 
1756 	if (flags & RENAME_EXCHANGE)
1757 		return gfs2_exchange(odir, odentry, ndir, ndentry, flags);
1758 
1759 	return gfs2_rename(odir, odentry, ndir, ndentry);
1760 }
1761 
1762 /**
1763  * gfs2_get_link - Follow a symbolic link
1764  * @dentry: The dentry of the link
1765  * @inode: The inode of the link
1766  * @done: destructor for return value
1767  *
1768  * This can handle symlinks of any size.
1769  *
1770  * Returns: 0 on success or error code
1771  */
1772 
1773 static const char *gfs2_get_link(struct dentry *dentry,
1774 				 struct inode *inode,
1775 				 struct delayed_call *done)
1776 {
1777 	struct gfs2_inode *ip = GFS2_I(inode);
1778 	struct gfs2_holder i_gh;
1779 	struct buffer_head *dibh;
1780 	unsigned int size;
1781 	char *buf;
1782 	int error;
1783 
1784 	if (!dentry)
1785 		return ERR_PTR(-ECHILD);
1786 
1787 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1788 	error = gfs2_glock_nq(&i_gh);
1789 	if (error) {
1790 		gfs2_holder_uninit(&i_gh);
1791 		return ERR_PTR(error);
1792 	}
1793 
1794 	size = (unsigned int)i_size_read(&ip->i_inode);
1795 	if (size == 0) {
1796 		gfs2_consist_inode(ip);
1797 		buf = ERR_PTR(-EIO);
1798 		goto out;
1799 	}
1800 
1801 	error = gfs2_meta_inode_buffer(ip, &dibh);
1802 	if (error) {
1803 		buf = ERR_PTR(error);
1804 		goto out;
1805 	}
1806 
1807 	buf = kzalloc(size + 1, GFP_NOFS);
1808 	if (!buf)
1809 		buf = ERR_PTR(-ENOMEM);
1810 	else
1811 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1812 	brelse(dibh);
1813 out:
1814 	gfs2_glock_dq_uninit(&i_gh);
1815 	if (!IS_ERR(buf))
1816 		set_delayed_call(done, kfree_link, buf);
1817 	return buf;
1818 }
1819 
1820 /**
1821  * gfs2_permission
1822  * @mnt_userns: User namespace of the mount the inode was found from
1823  * @inode: The inode
1824  * @mask: The mask to be tested
1825  *
1826  * This may be called from the VFS directly, or from within GFS2 with the
1827  * inode locked, so we look to see if the glock is already locked and only
1828  * lock the glock if its not already been done.
1829  *
1830  * Returns: errno
1831  */
1832 
1833 int gfs2_permission(struct user_namespace *mnt_userns, struct inode *inode,
1834 		    int mask)
1835 {
1836 	struct gfs2_inode *ip;
1837 	struct gfs2_holder i_gh;
1838 	int error;
1839 
1840 	gfs2_holder_mark_uninitialized(&i_gh);
1841 	ip = GFS2_I(inode);
1842 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1843 		if (mask & MAY_NOT_BLOCK)
1844 			return -ECHILD;
1845 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1846 		if (error)
1847 			return error;
1848 	}
1849 
1850 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1851 		error = -EPERM;
1852 	else
1853 		error = generic_permission(&init_user_ns, inode, mask);
1854 	if (gfs2_holder_initialized(&i_gh))
1855 		gfs2_glock_dq_uninit(&i_gh);
1856 
1857 	return error;
1858 }
1859 
1860 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1861 {
1862 	setattr_copy(&init_user_ns, inode, attr);
1863 	mark_inode_dirty(inode);
1864 	return 0;
1865 }
1866 
1867 static int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1868 {
1869 	int error;
1870 
1871 	if (current->journal_info)
1872 		return __gfs2_setattr_simple(inode, attr);
1873 
1874 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1875 	if (error)
1876 		return error;
1877 
1878 	error = __gfs2_setattr_simple(inode, attr);
1879 	gfs2_trans_end(GFS2_SB(inode));
1880 	return error;
1881 }
1882 
1883 static int setattr_chown(struct inode *inode, struct iattr *attr)
1884 {
1885 	struct gfs2_inode *ip = GFS2_I(inode);
1886 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1887 	kuid_t ouid, nuid;
1888 	kgid_t ogid, ngid;
1889 	int error;
1890 	struct gfs2_alloc_parms ap;
1891 
1892 	ouid = inode->i_uid;
1893 	ogid = inode->i_gid;
1894 	nuid = attr->ia_uid;
1895 	ngid = attr->ia_gid;
1896 
1897 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1898 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1899 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1900 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1901 	error = gfs2_qa_get(ip);
1902 	if (error)
1903 		return error;
1904 
1905 	error = gfs2_rindex_update(sdp);
1906 	if (error)
1907 		goto out;
1908 
1909 	error = gfs2_quota_lock(ip, nuid, ngid);
1910 	if (error)
1911 		goto out;
1912 
1913 	ap.target = gfs2_get_inode_blocks(&ip->i_inode);
1914 
1915 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1916 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1917 		error = gfs2_quota_check(ip, nuid, ngid, &ap);
1918 		if (error)
1919 			goto out_gunlock_q;
1920 	}
1921 
1922 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1923 	if (error)
1924 		goto out_gunlock_q;
1925 
1926 	error = gfs2_setattr_simple(inode, attr);
1927 	if (error)
1928 		goto out_end_trans;
1929 
1930 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1931 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1932 		gfs2_quota_change(ip, -(s64)ap.target, ouid, ogid);
1933 		gfs2_quota_change(ip, ap.target, nuid, ngid);
1934 	}
1935 
1936 out_end_trans:
1937 	gfs2_trans_end(sdp);
1938 out_gunlock_q:
1939 	gfs2_quota_unlock(ip);
1940 out:
1941 	gfs2_qa_put(ip);
1942 	return error;
1943 }
1944 
1945 /**
1946  * gfs2_setattr - Change attributes on an inode
1947  * @mnt_userns: User namespace of the mount the inode was found from
1948  * @dentry: The dentry which is changing
1949  * @attr: The structure describing the change
1950  *
1951  * The VFS layer wants to change one or more of an inodes attributes.  Write
1952  * that change out to disk.
1953  *
1954  * Returns: errno
1955  */
1956 
1957 static int gfs2_setattr(struct user_namespace *mnt_userns,
1958 			struct dentry *dentry, struct iattr *attr)
1959 {
1960 	struct inode *inode = d_inode(dentry);
1961 	struct gfs2_inode *ip = GFS2_I(inode);
1962 	struct gfs2_holder i_gh;
1963 	int error;
1964 
1965 	error = gfs2_qa_get(ip);
1966 	if (error)
1967 		return error;
1968 
1969 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1970 	if (error)
1971 		goto out;
1972 
1973 	error = may_setattr(&init_user_ns, inode, attr->ia_valid);
1974 	if (error)
1975 		goto error;
1976 
1977 	error = setattr_prepare(&init_user_ns, dentry, attr);
1978 	if (error)
1979 		goto error;
1980 
1981 	if (attr->ia_valid & ATTR_SIZE)
1982 		error = gfs2_setattr_size(inode, attr->ia_size);
1983 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1984 		error = setattr_chown(inode, attr);
1985 	else {
1986 		error = gfs2_setattr_simple(inode, attr);
1987 		if (!error && attr->ia_valid & ATTR_MODE)
1988 			error = posix_acl_chmod(&init_user_ns, inode,
1989 						inode->i_mode);
1990 	}
1991 
1992 error:
1993 	if (!error)
1994 		mark_inode_dirty(inode);
1995 	gfs2_glock_dq_uninit(&i_gh);
1996 out:
1997 	gfs2_qa_put(ip);
1998 	return error;
1999 }
2000 
2001 /**
2002  * gfs2_getattr - Read out an inode's attributes
2003  * @mnt_userns:	user namespace of the mount the inode was found from
2004  * @path: Object to query
2005  * @stat: The inode's stats
2006  * @request_mask: Mask of STATX_xxx flags indicating the caller's interests
2007  * @flags: AT_STATX_xxx setting
2008  *
2009  * This may be called from the VFS directly, or from within GFS2 with the
2010  * inode locked, so we look to see if the glock is already locked and only
2011  * lock the glock if its not already been done. Note that its the NFS
2012  * readdirplus operation which causes this to be called (from filldir)
2013  * with the glock already held.
2014  *
2015  * Returns: errno
2016  */
2017 
2018 static int gfs2_getattr(struct user_namespace *mnt_userns,
2019 			const struct path *path, struct kstat *stat,
2020 			u32 request_mask, unsigned int flags)
2021 {
2022 	struct inode *inode = d_inode(path->dentry);
2023 	struct gfs2_inode *ip = GFS2_I(inode);
2024 	struct gfs2_holder gh;
2025 	u32 gfsflags;
2026 	int error;
2027 
2028 	gfs2_holder_mark_uninitialized(&gh);
2029 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
2030 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
2031 		if (error)
2032 			return error;
2033 	}
2034 
2035 	gfsflags = ip->i_diskflags;
2036 	if (gfsflags & GFS2_DIF_APPENDONLY)
2037 		stat->attributes |= STATX_ATTR_APPEND;
2038 	if (gfsflags & GFS2_DIF_IMMUTABLE)
2039 		stat->attributes |= STATX_ATTR_IMMUTABLE;
2040 
2041 	stat->attributes_mask |= (STATX_ATTR_APPEND |
2042 				  STATX_ATTR_COMPRESSED |
2043 				  STATX_ATTR_ENCRYPTED |
2044 				  STATX_ATTR_IMMUTABLE |
2045 				  STATX_ATTR_NODUMP);
2046 
2047 	generic_fillattr(&init_user_ns, inode, stat);
2048 
2049 	if (gfs2_holder_initialized(&gh))
2050 		gfs2_glock_dq_uninit(&gh);
2051 
2052 	return 0;
2053 }
2054 
2055 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2056 		       u64 start, u64 len)
2057 {
2058 	struct gfs2_inode *ip = GFS2_I(inode);
2059 	struct gfs2_holder gh;
2060 	int ret;
2061 
2062 	inode_lock_shared(inode);
2063 
2064 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
2065 	if (ret)
2066 		goto out;
2067 
2068 	ret = iomap_fiemap(inode, fieinfo, start, len, &gfs2_iomap_ops);
2069 
2070 	gfs2_glock_dq_uninit(&gh);
2071 
2072 out:
2073 	inode_unlock_shared(inode);
2074 	return ret;
2075 }
2076 
2077 loff_t gfs2_seek_data(struct file *file, loff_t offset)
2078 {
2079 	struct inode *inode = file->f_mapping->host;
2080 	struct gfs2_inode *ip = GFS2_I(inode);
2081 	struct gfs2_holder gh;
2082 	loff_t ret;
2083 
2084 	inode_lock_shared(inode);
2085 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
2086 	if (!ret)
2087 		ret = iomap_seek_data(inode, offset, &gfs2_iomap_ops);
2088 	gfs2_glock_dq_uninit(&gh);
2089 	inode_unlock_shared(inode);
2090 
2091 	if (ret < 0)
2092 		return ret;
2093 	return vfs_setpos(file, ret, inode->i_sb->s_maxbytes);
2094 }
2095 
2096 loff_t gfs2_seek_hole(struct file *file, loff_t offset)
2097 {
2098 	struct inode *inode = file->f_mapping->host;
2099 	struct gfs2_inode *ip = GFS2_I(inode);
2100 	struct gfs2_holder gh;
2101 	loff_t ret;
2102 
2103 	inode_lock_shared(inode);
2104 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
2105 	if (!ret)
2106 		ret = iomap_seek_hole(inode, offset, &gfs2_iomap_ops);
2107 	gfs2_glock_dq_uninit(&gh);
2108 	inode_unlock_shared(inode);
2109 
2110 	if (ret < 0)
2111 		return ret;
2112 	return vfs_setpos(file, ret, inode->i_sb->s_maxbytes);
2113 }
2114 
2115 static int gfs2_update_time(struct inode *inode, struct timespec64 *time,
2116 			    int flags)
2117 {
2118 	struct gfs2_inode *ip = GFS2_I(inode);
2119 	struct gfs2_glock *gl = ip->i_gl;
2120 	struct gfs2_holder *gh;
2121 	int error;
2122 
2123 	gh = gfs2_glock_is_locked_by_me(gl);
2124 	if (gh && !gfs2_glock_is_held_excl(gl)) {
2125 		gfs2_glock_dq(gh);
2126 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, gh);
2127 		error = gfs2_glock_nq(gh);
2128 		if (error)
2129 			return error;
2130 	}
2131 	return generic_update_time(inode, time, flags);
2132 }
2133 
2134 static const struct inode_operations gfs2_file_iops = {
2135 	.permission = gfs2_permission,
2136 	.setattr = gfs2_setattr,
2137 	.getattr = gfs2_getattr,
2138 	.listxattr = gfs2_listxattr,
2139 	.fiemap = gfs2_fiemap,
2140 	.get_acl = gfs2_get_acl,
2141 	.set_acl = gfs2_set_acl,
2142 	.update_time = gfs2_update_time,
2143 	.fileattr_get = gfs2_fileattr_get,
2144 	.fileattr_set = gfs2_fileattr_set,
2145 };
2146 
2147 static const struct inode_operations gfs2_dir_iops = {
2148 	.create = gfs2_create,
2149 	.lookup = gfs2_lookup,
2150 	.link = gfs2_link,
2151 	.unlink = gfs2_unlink,
2152 	.symlink = gfs2_symlink,
2153 	.mkdir = gfs2_mkdir,
2154 	.rmdir = gfs2_unlink,
2155 	.mknod = gfs2_mknod,
2156 	.rename = gfs2_rename2,
2157 	.permission = gfs2_permission,
2158 	.setattr = gfs2_setattr,
2159 	.getattr = gfs2_getattr,
2160 	.listxattr = gfs2_listxattr,
2161 	.fiemap = gfs2_fiemap,
2162 	.get_acl = gfs2_get_acl,
2163 	.set_acl = gfs2_set_acl,
2164 	.update_time = gfs2_update_time,
2165 	.atomic_open = gfs2_atomic_open,
2166 	.fileattr_get = gfs2_fileattr_get,
2167 	.fileattr_set = gfs2_fileattr_set,
2168 };
2169 
2170 static const struct inode_operations gfs2_symlink_iops = {
2171 	.get_link = gfs2_get_link,
2172 	.permission = gfs2_permission,
2173 	.setattr = gfs2_setattr,
2174 	.getattr = gfs2_getattr,
2175 	.listxattr = gfs2_listxattr,
2176 	.fiemap = gfs2_fiemap,
2177 };
2178 
2179