xref: /linux/fs/gfs2/inode.c (revision 910a00830bc151d6fd8e515c834149080c4c4940)
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 void gfs2_setup_inode(struct inode *inode)
93 {
94 	gfp_t gfp_mask;
95 
96 	/*
97 	 * Ensure all page cache allocations are done from GFP_NOFS context to
98 	 * prevent direct reclaim recursion back into the filesystem and blowing
99 	 * stacks or deadlocking.
100 	 */
101 	gfp_mask = mapping_gfp_mask(inode->i_mapping);
102 	mapping_set_gfp_mask(inode->i_mapping, gfp_mask & ~__GFP_FS);
103 }
104 
105 /**
106  * gfs2_inode_lookup - Lookup an inode
107  * @sb: The super block
108  * @type: The type of the inode
109  * @no_addr: The inode number
110  * @no_formal_ino: The inode generation number
111  * @blktype: Requested block type (GFS2_BLKST_DINODE or GFS2_BLKST_UNLINKED;
112  *           GFS2_BLKST_FREE to indicate not to verify)
113  *
114  * If @type is DT_UNKNOWN, the inode type is fetched from disk.
115  *
116  * If @blktype is anything other than GFS2_BLKST_FREE (which is used as a
117  * placeholder because it doesn't otherwise make sense), the on-disk block type
118  * is verified to be @blktype.
119  *
120  * When @no_formal_ino is non-zero, this function will return ERR_PTR(-ESTALE)
121  * if it detects that @no_formal_ino doesn't match the actual inode generation
122  * number.  However, it doesn't always know unless @type is DT_UNKNOWN.
123  *
124  * Returns: A VFS inode, or an error
125  */
126 
127 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
128 				u64 no_addr, u64 no_formal_ino,
129 				unsigned int blktype)
130 {
131 	struct inode *inode;
132 	struct gfs2_inode *ip;
133 	struct gfs2_glock *gl = NULL;
134 	struct gfs2_holder i_gh;
135 	int error;
136 
137 	gfs2_holder_mark_uninitialized(&i_gh);
138 	inode = iget5_locked(sb, no_addr, iget_test, iget_set, &no_addr);
139 	if (!inode)
140 		return ERR_PTR(-ENOMEM);
141 
142 	ip = GFS2_I(inode);
143 
144 	if (inode_state_read_once(inode) & I_NEW) {
145 		struct gfs2_sbd *sdp = GFS2_SB(inode);
146 		struct gfs2_glock *io_gl;
147 		int extra_flags = 0;
148 
149 		gfs2_setup_inode(inode);
150 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE,
151 				       &gl);
152 		if (unlikely(error))
153 			goto fail;
154 		rcu_assign_pointer(ip->i_gl, gl);
155 
156 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE,
157 				       &io_gl);
158 		if (unlikely(error))
159 			goto fail;
160 
161 		/*
162 		 * The only caller that sets @blktype to GFS2_BLKST_UNLINKED is
163 		 * delete_work_func().  Make sure not to cancel the delete work
164 		 * from within itself here.
165 		 */
166 		if (blktype == GFS2_BLKST_UNLINKED)
167 			extra_flags |= LM_FLAG_TRY;
168 		else
169 			gfs2_cancel_delete_work(io_gl);
170 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED,
171 					   GL_EXACT | GL_NOPID | extra_flags,
172 					   &ip->i_iopen_gh);
173 		gfs2_glock_put(io_gl);
174 		if (unlikely(error))
175 			goto fail;
176 
177 		if (type == DT_UNKNOWN || blktype != GFS2_BLKST_FREE) {
178 			/*
179 			 * The GL_SKIP flag indicates to skip reading the inode
180 			 * block.  We read the inode when instantiating it
181 			 * after possibly checking the block type.
182 			 */
183 			error = gfs2_glock_nq_init(gl, LM_ST_EXCLUSIVE,
184 						   GL_SKIP, &i_gh);
185 			if (error)
186 				goto fail;
187 
188 			error = -ESTALE;
189 			if (no_formal_ino &&
190 			    gfs2_inode_already_deleted(gl, no_formal_ino))
191 				goto fail;
192 
193 			if (blktype != GFS2_BLKST_FREE) {
194 				error = gfs2_check_blk_type(sdp, no_addr,
195 							    blktype);
196 				if (error)
197 					goto fail;
198 			}
199 		}
200 
201 		set_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
202 
203 		/* Lowest possible timestamp; will be overwritten in gfs2_dinode_in. */
204 		inode_set_atime(inode,
205 				1LL << (8 * sizeof(inode_get_atime_sec(inode)) - 1),
206 				0);
207 
208 		glock_set_object(gl, ip);
209 
210 		if (type == DT_UNKNOWN) {
211 			/* Inode glock must be locked already */
212 			error = gfs2_instantiate(&i_gh);
213 			if (error) {
214 				glock_clear_object(gl, ip);
215 				goto fail;
216 			}
217 		} else {
218 			ip->i_no_formal_ino = no_formal_ino;
219 			inode->i_mode = DT2IF(type);
220 		}
221 
222 		if (gfs2_holder_initialized(&i_gh))
223 			gfs2_glock_dq_uninit(&i_gh);
224 		glock_set_object(ip->i_iopen_gh.gh_gl, ip);
225 
226 		gfs2_set_iop(inode);
227 		unlock_new_inode(inode);
228 	}
229 
230 	if (no_formal_ino && ip->i_no_formal_ino &&
231 	    no_formal_ino != ip->i_no_formal_ino) {
232 		iput(inode);
233 		return ERR_PTR(-ESTALE);
234 	}
235 
236 	return inode;
237 
238 fail:
239 	if (error == GLR_TRYFAILED)
240 		error = -EAGAIN;
241 	if (gfs2_holder_initialized(&ip->i_iopen_gh))
242 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
243 	if (gfs2_holder_initialized(&i_gh))
244 		gfs2_glock_dq_uninit(&i_gh);
245 	if (gl) {
246 		gfs2_glock_put(gl);
247 		rcu_assign_pointer(ip->i_gl, NULL);
248 	}
249 	iget_failed(inode);
250 	return ERR_PTR(error);
251 }
252 
253 /**
254  * gfs2_lookup_by_inum - look up an inode by inode number
255  * @sdp: The super block
256  * @no_addr: The inode number
257  * @no_formal_ino: The inode generation number (0 for any)
258  * @blktype: Requested block type (see gfs2_inode_lookup)
259  */
260 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
261 				  u64 no_formal_ino, unsigned int blktype)
262 {
263 	struct super_block *sb = sdp->sd_vfs;
264 	struct inode *inode;
265 	int error;
266 
267 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, no_formal_ino,
268 				  blktype);
269 	if (IS_ERR(inode))
270 		return inode;
271 
272 	if (no_formal_ino) {
273 		error = -EIO;
274 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
275 			goto fail_iput;
276 	}
277 	return inode;
278 
279 fail_iput:
280 	iput(inode);
281 	return ERR_PTR(error);
282 }
283 
284 
285 /**
286  * gfs2_lookup_meta - Look up an inode in a metadata directory
287  * @dip: The directory
288  * @name: The name of the inode
289  */
290 struct inode *gfs2_lookup_meta(struct inode *dip, const char *name)
291 {
292 	struct qstr qstr;
293 	struct inode *inode;
294 
295 	gfs2_str2qstr(&qstr, name);
296 	inode = gfs2_lookupi(dip, &qstr, 1);
297 	if (IS_ERR_OR_NULL(inode))
298 		return inode ? inode : ERR_PTR(-ENOENT);
299 
300 	/*
301 	 * Must not call back into the filesystem when allocating
302 	 * pages in the metadata inode's address space.
303 	 */
304 	mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
305 
306 	return inode;
307 }
308 
309 
310 /**
311  * gfs2_lookupi - Look up a filename in a directory and return its inode
312  * @dir: The inode of the directory containing the inode to look-up
313  * @name: The name of the inode to look for
314  * @is_root: If 1, ignore the caller's permissions
315  *
316  * This can be called via the VFS filldir function when NFS is doing
317  * a readdirplus and the inode which its intending to stat isn't
318  * already in cache. In this case we must not take the directory glock
319  * again, since the readdir call will have already taken that lock.
320  *
321  * Returns: errno
322  */
323 
324 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
325 			   int is_root)
326 {
327 	struct gfs2_glock *gl = gfs2_inode_glock(dir);
328 	struct super_block *sb = dir->i_sb;
329 	struct gfs2_holder d_gh;
330 	int error = 0;
331 	struct inode *inode = NULL;
332 
333 	gfs2_holder_mark_uninitialized(&d_gh);
334 	if (!name->len || name->len > GFS2_FNAMESIZE)
335 		return ERR_PTR(-ENAMETOOLONG);
336 
337 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
338 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
339 	     dir == d_inode(sb->s_root))) {
340 		igrab(dir);
341 		return dir;
342 	}
343 
344 	if (gfs2_glock_is_locked_by_me(gl) == NULL) {
345 		error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &d_gh);
346 		if (error)
347 			return ERR_PTR(error);
348 	}
349 
350 	if (!is_root) {
351 		error = gfs2_permission(&nop_mnt_idmap, dir, MAY_EXEC);
352 		if (error)
353 			goto out;
354 	}
355 
356 	inode = gfs2_dir_search(dir, name, false);
357 	if (IS_ERR(inode))
358 		error = PTR_ERR(inode);
359 out:
360 	if (gfs2_holder_initialized(&d_gh))
361 		gfs2_glock_dq_uninit(&d_gh);
362 	if (error == -ENOENT)
363 		return NULL;
364 	return inode ? inode : ERR_PTR(error);
365 }
366 
367 /**
368  * create_ok - OK to create a new on-disk inode here?
369  * @dip:  Directory in which dinode is to be created
370  * @name:  Name of new dinode
371  * @mode:
372  *
373  * Returns: errno
374  */
375 
376 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
377 		     umode_t mode)
378 {
379 	int error;
380 
381 	error = gfs2_permission(&nop_mnt_idmap, &dip->i_inode,
382 				MAY_WRITE | MAY_EXEC);
383 	if (error)
384 		return error;
385 
386 	/*  Don't create entries in an unlinked directory  */
387 	if (!dip->i_inode.i_nlink)
388 		return -ENOENT;
389 
390 	if (dip->i_entries == (u32)-1)
391 		return -EFBIG;
392 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
393 		return -EMLINK;
394 
395 	return 0;
396 }
397 
398 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
399 			       struct inode *inode)
400 {
401 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
402 	    (dip->i_inode.i_mode & S_ISUID) &&
403 	    !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
404 		if (S_ISDIR(inode->i_mode))
405 			inode->i_mode |= S_ISUID;
406 		else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
407 			inode->i_mode &= ~07111;
408 		inode->i_uid = dip->i_inode.i_uid;
409 	} else
410 		inode->i_uid = current_fsuid();
411 
412 	if (dip->i_inode.i_mode & S_ISGID) {
413 		if (S_ISDIR(inode->i_mode))
414 			inode->i_mode |= S_ISGID;
415 		inode->i_gid = dip->i_inode.i_gid;
416 	} else
417 		inode->i_gid = current_fsgid();
418 }
419 
420 static int alloc_dinode(struct gfs2_inode *ip, u32 flags, unsigned *dblocks)
421 {
422 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
423 	struct gfs2_alloc_parms ap = { .target = *dblocks, .aflags = flags, };
424 	int error;
425 
426 	error = gfs2_quota_lock_check(ip, &ap);
427 	if (error)
428 		goto out;
429 
430 	error = gfs2_inplace_reserve(ip, &ap);
431 	if (error)
432 		goto out_quota;
433 
434 	error = gfs2_trans_begin(sdp, (*dblocks * RES_RG_BIT) + RES_STATFS + RES_QUOTA, 0);
435 	if (error)
436 		goto out_ipreserv;
437 
438 	error = gfs2_alloc_blocks(ip, &ip->i_no_addr, dblocks, 1);
439 	if (error)
440 		goto out_trans_end;
441 
442 	ip->i_no_formal_ino = ip->i_generation;
443 	ip->i_inode.i_ino = ip->i_no_addr;
444 	ip->i_goal = ip->i_no_addr;
445 	if (*dblocks > 1)
446 		ip->i_eattr = ip->i_no_addr + 1;
447 
448 out_trans_end:
449 	gfs2_trans_end(sdp);
450 out_ipreserv:
451 	gfs2_inplace_release(ip);
452 out_quota:
453 	gfs2_quota_unlock(ip);
454 out:
455 	return error;
456 }
457 
458 static void gfs2_final_release_pages(struct gfs2_inode *ip)
459 {
460 	struct inode *inode = &ip->i_inode;
461 	struct gfs2_glock *gl = gfs2_inode_glock(inode);
462 
463 	/* This can only happen during incomplete inode creation. */
464 	if (unlikely(!gl))
465 		return;
466 
467 	truncate_inode_pages(gfs2_glock2aspace(gl), 0);
468 	truncate_inode_pages(&inode->i_data, 0);
469 
470 	if (atomic_read(&gl->gl_revokes) == 0) {
471 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
472 		clear_bit(GLF_DIRTY, &gl->gl_flags);
473 	}
474 }
475 
476 int gfs2_dinode_dealloc(struct gfs2_inode *ip)
477 {
478 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
479 	struct gfs2_rgrpd *rgd;
480 	struct gfs2_holder gh;
481 	int error;
482 
483 	if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
484 		gfs2_consist_inode(ip);
485 		return -EIO;
486 	}
487 
488 	gfs2_rindex_update(sdp);
489 
490 	error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
491 	if (error)
492 		return error;
493 
494 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
495 	if (!rgd) {
496 		gfs2_consist_inode(ip);
497 		error = -EIO;
498 		goto out_qs;
499 	}
500 
501 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
502 				   LM_FLAG_NODE_SCOPE, &gh);
503 	if (error)
504 		goto out_qs;
505 
506 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
507 				 sdp->sd_jdesc->jd_blocks);
508 	if (error)
509 		goto out_rg_gunlock;
510 
511 	gfs2_free_di(rgd, ip);
512 
513 	gfs2_final_release_pages(ip);
514 
515 	gfs2_trans_end(sdp);
516 
517 out_rg_gunlock:
518 	gfs2_glock_dq_uninit(&gh);
519 out_qs:
520 	gfs2_quota_unhold(ip);
521 	return error;
522 }
523 
524 static void gfs2_init_dir(struct buffer_head *dibh,
525 			  const struct gfs2_inode *parent)
526 {
527 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
528 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
529 
530 	gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
531 	dent->de_inum = di->di_num; /* already GFS2 endian */
532 	dent->de_type = cpu_to_be16(DT_DIR);
533 
534 	dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
535 	gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
536 	gfs2_inum_out(parent, dent);
537 	dent->de_type = cpu_to_be16(DT_DIR);
538 
539 }
540 
541 /**
542  * gfs2_init_xattr - Initialise an xattr block for a new inode
543  * @ip: The inode in question
544  *
545  * This sets up an empty xattr block for a new inode, ready to
546  * take any ACLs, LSM xattrs, etc.
547  */
548 
549 static void gfs2_init_xattr(struct gfs2_inode *ip)
550 {
551 	struct gfs2_glock *gl = gfs2_inode_glock(&ip->i_inode);
552 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
553 	struct buffer_head *bh;
554 	struct gfs2_ea_header *ea;
555 
556 	bh = gfs2_meta_new(gl, ip->i_eattr);
557 	gfs2_trans_add_meta(gl, bh);
558 	gfs2_metatype_set(bh, GFS2_METATYPE_EA, GFS2_FORMAT_EA);
559 	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
560 
561 	ea = GFS2_EA_BH2FIRST(bh);
562 	ea->ea_rec_len = cpu_to_be32(sdp->sd_jbsize);
563 	ea->ea_type = GFS2_EATYPE_UNUSED;
564 	ea->ea_flags = GFS2_EAFLAG_LAST;
565 
566 	brelse(bh);
567 }
568 
569 /**
570  * init_dinode - Fill in a new dinode structure
571  * @dip: The directory this inode is being created in
572  * @ip: The inode
573  * @symname: The symlink destination (if a symlink)
574  *
575  */
576 
577 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
578 			const char *symname)
579 {
580 	struct gfs2_glock *gl = gfs2_inode_glock(&ip->i_inode);
581 	struct gfs2_dinode *di;
582 	struct buffer_head *dibh;
583 
584 	dibh = gfs2_meta_new(gl, ip->i_no_addr);
585 	gfs2_trans_add_meta(gl, dibh);
586 	di = (struct gfs2_dinode *)dibh->b_data;
587 	gfs2_dinode_out(ip, di);
588 
589 	di->di_major = cpu_to_be32(imajor(&ip->i_inode));
590 	di->di_minor = cpu_to_be32(iminor(&ip->i_inode));
591 	di->__pad1 = 0;
592 	di->__pad2 = 0;
593 	di->__pad3 = 0;
594 	memset(&di->__pad4, 0, sizeof(di->__pad4));
595 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
596 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
597 
598 	switch(ip->i_inode.i_mode & S_IFMT) {
599 	case S_IFDIR:
600 		gfs2_init_dir(dibh, dip);
601 		break;
602 	case S_IFLNK:
603 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
604 		break;
605 	}
606 
607 	set_buffer_uptodate(dibh);
608 	brelse(dibh);
609 }
610 
611 /**
612  * gfs2_trans_da_blks - Calculate number of blocks to link inode
613  * @dip: The directory we are linking into
614  * @da: The dir add information
615  * @nr_inodes: The number of inodes involved
616  *
617  * This calculate the number of blocks we need to reserve in a
618  * transaction to link @nr_inodes into a directory. In most cases
619  * @nr_inodes will be 2 (the directory plus the inode being linked in)
620  * but in case of rename, 4 may be required.
621  *
622  * Returns: Number of blocks
623  */
624 
625 static unsigned gfs2_trans_da_blks(const struct gfs2_inode *dip,
626 				   const struct gfs2_diradd *da,
627 				   unsigned nr_inodes)
628 {
629 	return da->nr_blocks + gfs2_rg_blocks(dip, da->nr_blocks) +
630 	       (nr_inodes * RES_DINODE) + RES_QUOTA + RES_STATFS;
631 }
632 
633 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
634 		       struct gfs2_inode *ip, struct gfs2_diradd *da)
635 {
636 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
637 	struct gfs2_alloc_parms ap = { .target = da->nr_blocks, };
638 	int error;
639 
640 	if (da->nr_blocks) {
641 		error = gfs2_quota_lock_check(dip, &ap);
642 		if (error)
643 			goto fail_quota_locks;
644 
645 		error = gfs2_inplace_reserve(dip, &ap);
646 		if (error)
647 			goto fail_quota_locks;
648 
649 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, da, 2), 0);
650 		if (error)
651 			goto fail_ipreserv;
652 	} else {
653 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
654 		if (error)
655 			goto fail_quota_locks;
656 	}
657 
658 	error = gfs2_dir_add(&dip->i_inode, name, ip, da);
659 
660 	gfs2_trans_end(sdp);
661 fail_ipreserv:
662 	gfs2_inplace_release(dip);
663 fail_quota_locks:
664 	gfs2_quota_unlock(dip);
665 	return error;
666 }
667 
668 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
669 		    void *fs_info)
670 {
671 	const struct xattr *xattr;
672 	int err = 0;
673 
674 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
675 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
676 				       xattr->value_len, 0,
677 				       GFS2_EATYPE_SECURITY);
678 		if (err < 0)
679 			break;
680 	}
681 	return err;
682 }
683 
684 /**
685  * gfs2_create_inode - Create a new inode
686  * @dir: The parent directory
687  * @dentry: The new dentry
688  * @file: If non-NULL, the file which is being opened
689  * @mode: The permissions on the new inode
690  * @dev: For device nodes, this is the device number
691  * @symname: For symlinks, this is the link destination
692  * @size: The initial size of the inode (ignored for directories)
693  * @excl: Force fail if inode exists
694  *
695  * FIXME: Change to allocate the disk blocks and write them out in the same
696  * transaction.  That way, we can no longer end up in a situation in which an
697  * inode is allocated, the node crashes, and the block looks like a valid
698  * inode.  (With atomic creates in place, we will also no longer need to zero
699  * the link count and dirty the inode here on failure.)
700  *
701  * Returns: 0 on success, or error code
702  */
703 
704 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
705 			     struct file *file,
706 			     umode_t mode, dev_t dev, const char *symname,
707 			     unsigned int size, int excl)
708 {
709 	const struct qstr *name = &dentry->d_name;
710 	struct posix_acl *default_acl, *acl;
711 	struct gfs2_holder d_gh, gh;
712 	struct inode *inode = NULL;
713 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
714 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
715 	struct gfs2_glock *gl = NULL, *io_gl;
716 	int error, dealloc_error;
717 	u32 aflags = 0;
718 	unsigned blocks = 1;
719 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
720 	bool xattr_initialized = false;
721 
722 	if (!name->len || name->len > GFS2_FNAMESIZE)
723 		return -ENAMETOOLONG;
724 
725 	error = gfs2_qa_get(dip);
726 	if (error)
727 		return error;
728 
729 	error = gfs2_rindex_update(sdp);
730 	if (error)
731 		goto fail;
732 
733 	error = gfs2_glock_nq_init(gfs2_inode_glock(dir), LM_ST_EXCLUSIVE, 0,
734 				   &d_gh);
735 	if (error)
736 		goto fail;
737 	gfs2_holder_mark_uninitialized(&gh);
738 
739 	error = create_ok(dip, name, mode);
740 	if (error)
741 		goto fail_gunlock;
742 
743 	inode = gfs2_dir_search(dir, &dentry->d_name, excl);
744 	error = PTR_ERR(inode);
745 	if (!IS_ERR(inode)) {
746 		if (file && (file->f_flags & __O_REGULAR) &&
747 		    !S_ISREG(inode->i_mode)) {
748 			iput(inode);
749 			inode = NULL;
750 			error = -EFTYPE;
751 			goto fail_gunlock;
752 		}
753 		if (S_ISDIR(inode->i_mode)) {
754 			iput(inode);
755 			inode = NULL;
756 			error = -EISDIR;
757 			goto fail_gunlock;
758 		}
759 		d_instantiate(dentry, inode);
760 		error = 0;
761 		if (file) {
762 			if (S_ISREG(inode->i_mode))
763 				error = finish_open(file, dentry, gfs2_open_common);
764 			else
765 				error = finish_no_open(file, NULL);
766 		}
767 		gfs2_glock_dq_uninit(&d_gh);
768 		goto fail;
769 	} else if (error != -ENOENT) {
770 		goto fail_gunlock;
771 	}
772 
773 	error = gfs2_diradd_alloc_required(dir, name, &da);
774 	if (error < 0)
775 		goto fail_gunlock;
776 
777 	inode = new_inode(sdp->sd_vfs);
778 	error = -ENOMEM;
779 	if (!inode)
780 		goto fail_gunlock;
781 	gfs2_setup_inode(inode);
782 	ip = GFS2_I(inode);
783 
784 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
785 	if (error)
786 		goto fail_gunlock;
787 
788 	error = gfs2_qa_get(ip);
789 	if (error)
790 		goto fail_free_acls;
791 
792 	inode->i_mode = mode;
793 	set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
794 	inode->i_rdev = dev;
795 	inode->i_size = size;
796 	simple_inode_init_ts(inode);
797 	munge_mode_uid_gid(dip, inode);
798 	check_and_update_goal(dip);
799 	ip->i_goal = dip->i_goal;
800 	ip->i_diskflags = 0;
801 	ip->i_eattr = 0;
802 	ip->i_height = 0;
803 	ip->i_depth = 0;
804 	ip->i_entries = 0;
805 	ip->i_no_addr = 0; /* Temporarily zero until real addr is assigned */
806 
807 	switch(mode & S_IFMT) {
808 	case S_IFREG:
809 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
810 		    gfs2_tune_get(sdp, gt_new_files_jdata))
811 			ip->i_diskflags |= GFS2_DIF_JDATA;
812 		gfs2_set_aops(inode);
813 		break;
814 	case S_IFDIR:
815 		ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
816 		ip->i_diskflags |= GFS2_DIF_JDATA;
817 		ip->i_entries = 2;
818 		break;
819 	}
820 
821 	/* Force SYSTEM flag on all files and subdirs of a SYSTEM directory */
822 	if (dip->i_diskflags & GFS2_DIF_SYSTEM)
823 		ip->i_diskflags |= GFS2_DIF_SYSTEM;
824 
825 	gfs2_set_inode_flags(inode);
826 
827 	if ((GFS2_I(d_inode(sdp->sd_root_dir)) == dip) ||
828 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
829 		aflags |= GFS2_AF_ORLOV;
830 
831 	if (default_acl || acl)
832 		blocks++;
833 
834 	error = alloc_dinode(ip, aflags, &blocks);
835 	if (error)
836 		goto fail_free_inode;
837 
838 	gfs2_set_inode_blocks(inode, blocks);
839 
840 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &gl);
841 	if (error)
842 		goto fail_dealloc_inode;
843 	rcu_assign_pointer(ip->i_gl, gl);
844 
845 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
846 	if (error)
847 		goto fail_dealloc_inode;
848 	gfs2_cancel_delete_work(io_gl);
849 	io_gl->gl_no_formal_ino = ip->i_no_formal_ino;
850 
851 retry:
852 	error = insert_inode_locked4(inode, ip->i_no_addr, iget_test, &ip->i_no_addr);
853 	if (error == -EBUSY)
854 		goto retry;
855 	if (error)
856 		goto fail_gunlock2;
857 
858 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT | GL_NOPID,
859 				   &ip->i_iopen_gh);
860 	if (error)
861 		goto fail_gunlock2;
862 
863 	error = gfs2_glock_nq_init(gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
864 	if (error)
865 		goto fail_gunlock3;
866 	clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
867 
868 	error = gfs2_trans_begin(sdp, blocks, 0);
869 	if (error)
870 		goto fail_gunlock3;
871 
872 	if (blocks > 1) {
873 		gfs2_init_xattr(ip);
874 		xattr_initialized = true;
875 	}
876 	init_dinode(dip, ip, symname);
877 	gfs2_trans_end(sdp);
878 
879 	glock_set_object(gl, ip);
880 	glock_set_object(io_gl, ip);
881 	gfs2_set_iop(inode);
882 
883 	if (default_acl) {
884 		error = __gfs2_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
885 		if (error)
886 			goto fail_gunlock4;
887 		posix_acl_release(default_acl);
888 		default_acl = NULL;
889 	}
890 	if (acl) {
891 		error = __gfs2_set_acl(inode, acl, ACL_TYPE_ACCESS);
892 		if (error)
893 			goto fail_gunlock4;
894 		posix_acl_release(acl);
895 		acl = NULL;
896 	}
897 
898 	error = security_inode_init_security(&ip->i_inode, &dip->i_inode, name,
899 					     &gfs2_initxattrs, NULL);
900 	if (error)
901 		goto fail_gunlock4;
902 
903 	error = link_dinode(dip, name, ip, &da);
904 	if (error)
905 		goto fail_gunlock4;
906 
907 	mark_inode_dirty(inode);
908 	d_instantiate_new(dentry, inode);
909 	/* After instantiate, errors should result in evict which will destroy
910 	 * both inode and iopen glocks properly. */
911 	if (file) {
912 		file->f_mode |= FMODE_CREATED;
913 		error = finish_open(file, dentry, gfs2_open_common);
914 	}
915 	gfs2_glock_dq_uninit(&d_gh);
916 	gfs2_qa_put(ip);
917 	gfs2_glock_dq_uninit(&gh);
918 	gfs2_glock_put(io_gl);
919 	gfs2_qa_put(dip);
920 	return error;
921 
922 fail_gunlock4:
923 	glock_clear_object(gl, ip);
924 	glock_clear_object(io_gl, ip);
925 fail_gunlock3:
926 	gfs2_glock_dq_uninit(&ip->i_iopen_gh);
927 fail_gunlock2:
928 	gfs2_glock_put(io_gl);
929 fail_dealloc_inode:
930 	dealloc_error = 0;
931 	if (ip->i_eattr)
932 		dealloc_error = gfs2_ea_dealloc(ip, xattr_initialized);
933 	clear_nlink(inode);
934 	mark_inode_dirty(inode);
935 	if (!dealloc_error)
936 		dealloc_error = gfs2_dinode_dealloc(ip);
937 	if (dealloc_error)
938 		fs_warn(sdp, "%s: %d\n", __func__, dealloc_error);
939 	ip->i_no_addr = 0;
940 fail_free_inode:
941 	if (gl) {
942 		gfs2_glock_put(gl);
943 		rcu_assign_pointer(ip->i_gl, NULL);
944 	}
945 	gfs2_rs_deltree(&ip->i_res);
946 	gfs2_qa_put(ip);
947 fail_free_acls:
948 	posix_acl_release(default_acl);
949 	posix_acl_release(acl);
950 fail_gunlock:
951 	gfs2_dir_no_add(&da);
952 	gfs2_glock_dq_uninit(&d_gh);
953 	if (!IS_ERR_OR_NULL(inode)) {
954 		if (inode_state_read_once(inode) & I_NEW)
955 			iget_failed(inode);
956 		else
957 			iput(inode);
958 	}
959 	if (gfs2_holder_initialized(&gh))
960 		gfs2_glock_dq_uninit(&gh);
961 fail:
962 	gfs2_qa_put(dip);
963 	return error;
964 }
965 
966 /**
967  * gfs2_create - Create a file
968  * @idmap: idmap of the mount the inode was found from
969  * @dir: The directory in which to create the file
970  * @dentry: The dentry of the new file
971  * @mode: The mode of the new file
972  *
973  * Returns: errno
974  */
975 
976 static int gfs2_create(const struct mnt_idmap *idmap, struct inode *dir,
977 		       struct dentry *dentry, umode_t mode)
978 {
979 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, 1);
980 }
981 
982 /**
983  * __gfs2_lookup - Look up a filename in a directory and return its inode
984  * @dir: The directory inode
985  * @dentry: The dentry of the new inode
986  * @file: File to be opened
987  *
988  *
989  * Returns: errno
990  */
991 
992 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
993 				    struct file *file)
994 {
995 	struct inode *inode;
996 	struct dentry *d;
997 	struct gfs2_holder gh;
998 	struct gfs2_glock *gl;
999 	int error;
1000 
1001 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
1002 	if (inode == NULL || IS_ERR(inode))
1003 		return d_splice_alias(inode, dentry);
1004 
1005 	gl = gfs2_inode_glock(inode);
1006 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1007 	if (error) {
1008 		iput(inode);
1009 		return ERR_PTR(error);
1010 	}
1011 
1012 	d = d_splice_alias(inode, dentry);
1013 	if (IS_ERR(d)) {
1014 		gfs2_glock_dq_uninit(&gh);
1015 		return d;
1016 	}
1017 	if (file && S_ISREG(inode->i_mode))
1018 		error = finish_open(file, dentry, gfs2_open_common);
1019 
1020 	gfs2_glock_dq_uninit(&gh);
1021 	if (error) {
1022 		dput(d);
1023 		return ERR_PTR(error);
1024 	}
1025 	return d;
1026 }
1027 
1028 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
1029 				  unsigned flags)
1030 {
1031 	return __gfs2_lookup(dir, dentry, NULL);
1032 }
1033 
1034 /**
1035  * gfs2_link - Link to a file
1036  * @old_dentry: The inode to link
1037  * @dir: Add link to this directory
1038  * @dentry: The name of the link
1039  *
1040  * Link the inode in "old_dentry" into the directory "dir" with the
1041  * name in "dentry".
1042  *
1043  * Returns: errno
1044  */
1045 
1046 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
1047 		     struct dentry *dentry)
1048 {
1049 	struct gfs2_inode *dip = GFS2_I(dir);
1050 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1051 	struct inode *inode = d_inode(old_dentry);
1052 	struct gfs2_inode *ip = GFS2_I(inode);
1053 	struct gfs2_holder d_gh, gh;
1054 	struct buffer_head *dibh;
1055 	struct gfs2_diradd da = { .bh = NULL, .save_loc = 1, };
1056 	int error;
1057 
1058 	if (S_ISDIR(inode->i_mode))
1059 		return -EPERM;
1060 
1061 	error = gfs2_qa_get(dip);
1062 	if (error)
1063 		return error;
1064 
1065 	gfs2_holder_init(gfs2_inode_glock(dir), LM_ST_EXCLUSIVE, 0, &d_gh);
1066 	gfs2_holder_init(gfs2_inode_glock(inode), LM_ST_EXCLUSIVE, 0, &gh);
1067 
1068 	error = gfs2_glock_nq(&d_gh);
1069 	if (error)
1070 		goto out_parent;
1071 
1072 	error = gfs2_glock_nq(&gh);
1073 	if (error)
1074 		goto out_child;
1075 
1076 	error = -ENOENT;
1077 	if (inode->i_nlink == 0)
1078 		goto out_gunlock;
1079 
1080 	error = gfs2_permission(&nop_mnt_idmap, dir, MAY_WRITE | MAY_EXEC);
1081 	if (error)
1082 		goto out_gunlock;
1083 
1084 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
1085 	switch (error) {
1086 	case -ENOENT:
1087 		break;
1088 	case 0:
1089 		error = -EEXIST;
1090 		goto out_gunlock;
1091 	default:
1092 		goto out_gunlock;
1093 	}
1094 
1095 	error = -EINVAL;
1096 	if (!dip->i_inode.i_nlink)
1097 		goto out_gunlock;
1098 	error = -EFBIG;
1099 	if (dip->i_entries == (u32)-1)
1100 		goto out_gunlock;
1101 	error = -EPERM;
1102 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1103 		goto out_gunlock;
1104 	error = -EMLINK;
1105 	if (ip->i_inode.i_nlink == (u32)-1)
1106 		goto out_gunlock;
1107 
1108 	error = gfs2_diradd_alloc_required(dir, &dentry->d_name, &da);
1109 	if (error < 0)
1110 		goto out_gunlock;
1111 
1112 	if (da.nr_blocks) {
1113 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1114 		error = gfs2_quota_lock_check(dip, &ap);
1115 		if (error)
1116 			goto out_gunlock;
1117 
1118 		error = gfs2_inplace_reserve(dip, &ap);
1119 		if (error)
1120 			goto out_gunlock_q;
1121 
1122 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, &da, 2), 0);
1123 		if (error)
1124 			goto out_ipres;
1125 	} else {
1126 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
1127 		if (error)
1128 			goto out_ipres;
1129 	}
1130 
1131 	error = gfs2_meta_inode_buffer(ip, &dibh);
1132 	if (error)
1133 		goto out_end_trans;
1134 
1135 	error = gfs2_dir_add(dir, &dentry->d_name, ip, &da);
1136 	if (error)
1137 		goto out_brelse;
1138 
1139 	gfs2_trans_add_meta(gfs2_inode_glock(inode), dibh);
1140 	inc_nlink(&ip->i_inode);
1141 	inode_set_ctime_current(&ip->i_inode);
1142 	ihold(inode);
1143 	d_instantiate(dentry, inode);
1144 	mark_inode_dirty(inode);
1145 
1146 out_brelse:
1147 	brelse(dibh);
1148 out_end_trans:
1149 	gfs2_trans_end(sdp);
1150 out_ipres:
1151 	if (da.nr_blocks)
1152 		gfs2_inplace_release(dip);
1153 out_gunlock_q:
1154 	if (da.nr_blocks)
1155 		gfs2_quota_unlock(dip);
1156 out_gunlock:
1157 	gfs2_dir_no_add(&da);
1158 	gfs2_glock_dq(&gh);
1159 out_child:
1160 	gfs2_glock_dq(&d_gh);
1161 out_parent:
1162 	gfs2_qa_put(dip);
1163 	gfs2_holder_uninit(&d_gh);
1164 	gfs2_holder_uninit(&gh);
1165 	return error;
1166 }
1167 
1168 /*
1169  * gfs2_unlink_ok - check to see that a inode is still in a directory
1170  * @dip: the directory
1171  * @name: the name of the file
1172  * @ip: the inode
1173  *
1174  * Assumes that the lock on (at least) @dip is held.
1175  *
1176  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1177  */
1178 
1179 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
1180 			  const struct gfs2_inode *ip)
1181 {
1182 	int error;
1183 
1184 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
1185 		return -EPERM;
1186 
1187 	if ((dip->i_inode.i_mode & S_ISVTX) &&
1188 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
1189 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
1190 		return -EPERM;
1191 
1192 	if (IS_APPEND(&dip->i_inode))
1193 		return -EPERM;
1194 
1195 	error = gfs2_permission(&nop_mnt_idmap, &dip->i_inode,
1196 				MAY_WRITE | MAY_EXEC);
1197 	if (error)
1198 		return error;
1199 
1200 	return gfs2_dir_check(&dip->i_inode, name, ip);
1201 }
1202 
1203 /**
1204  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1205  * @dip: The parent directory
1206  * @dentry: The dentry to unlink
1207  *
1208  * Called with all the locks and in a transaction. This will only be
1209  * called for a directory after it has been checked to ensure it is empty.
1210  *
1211  * Returns: 0 on success, or an error
1212  */
1213 
1214 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1215 			     const struct dentry *dentry)
1216 {
1217 	struct inode *inode = d_inode(dentry);
1218 	struct gfs2_inode *ip = GFS2_I(inode);
1219 	int error;
1220 
1221 	error = gfs2_dir_del(dip, dentry);
1222 	if (error)
1223 		return error;
1224 
1225 	ip->i_entries = 0;
1226 	inode_set_ctime_current(inode);
1227 	if (S_ISDIR(inode->i_mode))
1228 		clear_nlink(inode);
1229 	else
1230 		drop_nlink(inode);
1231 	mark_inode_dirty(inode);
1232 	if (inode->i_nlink == 0)
1233 		gfs2_unlink_di(inode);
1234 	return 0;
1235 }
1236 
1237 
1238 /**
1239  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1240  * @dir: The inode of the directory containing the inode to unlink
1241  * @dentry: The file itself
1242  *
1243  * This routine uses the type of the inode as a flag to figure out
1244  * whether this is an unlink or an rmdir.
1245  *
1246  * Returns: errno
1247  */
1248 
1249 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1250 {
1251 	struct gfs2_inode *dip = GFS2_I(dir);
1252 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1253 	struct inode *inode = d_inode(dentry);
1254 	struct gfs2_inode *ip = GFS2_I(inode);
1255 	struct gfs2_holder d_gh, r_gh, gh;
1256 	struct gfs2_rgrpd *rgd;
1257 	int error;
1258 
1259 	error = gfs2_rindex_update(sdp);
1260 	if (error)
1261 		return error;
1262 
1263 	error = -EROFS;
1264 
1265 	gfs2_holder_init(gfs2_inode_glock(dir), LM_ST_EXCLUSIVE, 0, &d_gh);
1266 	gfs2_holder_init(gfs2_inode_glock(inode),  LM_ST_EXCLUSIVE, 0, &gh);
1267 
1268 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1269 	if (!rgd)
1270 		goto out_inodes;
1271 
1272 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, LM_FLAG_NODE_SCOPE, &r_gh);
1273 
1274 
1275 	error = gfs2_glock_nq(&d_gh);
1276 	if (error)
1277 		goto out_parent;
1278 
1279 	error = gfs2_glock_nq(&gh);
1280 	if (error)
1281 		goto out_child;
1282 
1283 	error = -ENOENT;
1284 	if (inode->i_nlink == 0)
1285 		goto out_rgrp;
1286 
1287 	if (S_ISDIR(inode->i_mode)) {
1288 		error = -ENOTEMPTY;
1289 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1290 			goto out_rgrp;
1291 	}
1292 
1293 	error = gfs2_glock_nq(&r_gh); /* rgrp */
1294 	if (error)
1295 		goto out_rgrp;
1296 
1297 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1298 	if (error)
1299 		goto out_gunlock;
1300 
1301 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1302 	if (error)
1303 		goto out_gunlock;
1304 
1305 	error = gfs2_unlink_inode(dip, dentry);
1306 	gfs2_trans_end(sdp);
1307 
1308 out_gunlock:
1309 	gfs2_glock_dq(&r_gh);
1310 out_rgrp:
1311 	gfs2_glock_dq(&gh);
1312 out_child:
1313 	gfs2_glock_dq(&d_gh);
1314 out_parent:
1315 	gfs2_holder_uninit(&r_gh);
1316 out_inodes:
1317 	gfs2_holder_uninit(&gh);
1318 	gfs2_holder_uninit(&d_gh);
1319 	return error;
1320 }
1321 
1322 /**
1323  * gfs2_symlink - Create a symlink
1324  * @idmap: idmap of the mount the inode was found from
1325  * @dir: The directory to create the symlink in
1326  * @dentry: The dentry to put the symlink in
1327  * @symname: The thing which the link points to
1328  *
1329  * Returns: errno
1330  */
1331 
1332 static int gfs2_symlink(const struct mnt_idmap *idmap, struct inode *dir,
1333 			struct dentry *dentry, const char *symname)
1334 {
1335 	unsigned int size;
1336 
1337 	size = strlen(symname);
1338 	if (size >= gfs2_max_stuffed_size(GFS2_I(dir)))
1339 		return -ENAMETOOLONG;
1340 
1341 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 1);
1342 }
1343 
1344 /**
1345  * gfs2_mkdir - Make a directory
1346  * @idmap: idmap of the mount the inode was found from
1347  * @dir: The parent directory of the new one
1348  * @dentry: The dentry of the new directory
1349  * @mode: The mode of the new directory
1350  *
1351  * Returns: the dentry, or ERR_PTR(errno)
1352  */
1353 
1354 static struct dentry *gfs2_mkdir(const struct mnt_idmap *idmap, struct inode *dir,
1355 				 struct dentry *dentry, umode_t mode)
1356 {
1357 	unsigned dsize = gfs2_max_stuffed_size(GFS2_I(dir));
1358 
1359 	return ERR_PTR(gfs2_create_inode(dir, dentry, NULL, mode, 0, NULL, dsize, 1));
1360 }
1361 
1362 /**
1363  * gfs2_mknod - Make a special file
1364  * @idmap: idmap of the mount the inode was found from
1365  * @dir: The directory in which the special file will reside
1366  * @dentry: The dentry of the special file
1367  * @mode: The mode of the special file
1368  * @dev: The device specification of the special file
1369  *
1370  */
1371 
1372 static int gfs2_mknod(const struct mnt_idmap *idmap, struct inode *dir,
1373 		      struct dentry *dentry, umode_t mode, dev_t dev)
1374 {
1375 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 1);
1376 }
1377 
1378 /**
1379  * gfs2_atomic_open - Atomically open a file
1380  * @dir: The directory
1381  * @dentry: The proposed new entry
1382  * @file: The proposed new struct file
1383  * @flags: open flags
1384  * @mode: File mode
1385  *
1386  * Returns: error code or 0 for success
1387  */
1388 
1389 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1390 			    struct file *file, unsigned flags,
1391 			    umode_t mode)
1392 {
1393 	bool excl = !!(flags & O_EXCL);
1394 
1395 	if (d_in_lookup(dentry)) {
1396 		struct dentry *d = __gfs2_lookup(dir, dentry, file);
1397 		if (file->f_mode & FMODE_OPENED) {
1398 			if (IS_ERR(d))
1399 				return PTR_ERR(d);
1400 			dput(d);
1401 			return excl && (flags & O_CREAT) ? -EEXIST : 0;
1402 		}
1403 		if (d || d_really_is_positive(dentry))
1404 			return finish_no_open(file, d);
1405 	}
1406 	if (!(flags & O_CREAT))
1407 		return -ENOENT;
1408 
1409 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl);
1410 }
1411 
1412 /*
1413  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1414  * @this: move this
1415  * @to: to here
1416  *
1417  * Follow @to back to the root and make sure we don't encounter @this
1418  * Assumes we already hold the rename lock.
1419  *
1420  * Returns: errno
1421  */
1422 
1423 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1424 {
1425 	struct inode *dir = &to->i_inode;
1426 	struct super_block *sb = dir->i_sb;
1427 	struct inode *tmp;
1428 	int error = 0;
1429 
1430 	igrab(dir);
1431 
1432 	for (;;) {
1433 		if (dir == &this->i_inode) {
1434 			error = -EINVAL;
1435 			break;
1436 		}
1437 		if (dir == d_inode(sb->s_root)) {
1438 			error = 0;
1439 			break;
1440 		}
1441 
1442 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1443 		if (!tmp) {
1444 			error = -ENOENT;
1445 			break;
1446 		}
1447 		if (IS_ERR(tmp)) {
1448 			error = PTR_ERR(tmp);
1449 			break;
1450 		}
1451 
1452 		iput(dir);
1453 		dir = tmp;
1454 	}
1455 
1456 	iput(dir);
1457 
1458 	return error;
1459 }
1460 
1461 /**
1462  * update_moved_ino - Update an inode that's being moved
1463  * @ip: The inode being moved
1464  * @ndip: The parent directory of the new filename
1465  * @dir_rename: True of ip is a directory
1466  *
1467  * Returns: errno
1468  */
1469 
1470 static int update_moved_ino(struct gfs2_inode *ip, struct gfs2_inode *ndip,
1471 			    int dir_rename)
1472 {
1473 	if (dir_rename)
1474 		return gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1475 
1476 	inode_set_ctime_current(&ip->i_inode);
1477 	mark_inode_dirty_sync(&ip->i_inode);
1478 	return 0;
1479 }
1480 
1481 
1482 /**
1483  * gfs2_rename - Rename a file
1484  * @odir: Parent directory of old file name
1485  * @odentry: The old dentry of the file
1486  * @ndir: Parent directory of new file name
1487  * @ndentry: The new dentry of the file
1488  *
1489  * Returns: errno
1490  */
1491 
1492 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1493 		       struct inode *ndir, struct dentry *ndentry)
1494 {
1495 	struct gfs2_inode *odip = GFS2_I(odir);
1496 	struct gfs2_inode *ndip = GFS2_I(ndir);
1497 	struct gfs2_inode *ip = GFS2_I(d_inode(odentry));
1498 	struct gfs2_inode *nip = NULL;
1499 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1500 	struct gfs2_holder ghs[4], r_gh, rd_gh;
1501 	struct gfs2_rgrpd *nrgd;
1502 	unsigned int num_gh;
1503 	int dir_rename = 0;
1504 	struct gfs2_diradd da = { .nr_blocks = 0, .save_loc = 0, };
1505 	unsigned int retries = 0, x;
1506 	int error;
1507 
1508 	gfs2_holder_mark_uninitialized(&r_gh);
1509 	gfs2_holder_mark_uninitialized(&rd_gh);
1510 	if (d_really_is_positive(ndentry)) {
1511 		nip = GFS2_I(d_inode(ndentry));
1512 		if (ip == nip)
1513 			return 0;
1514 	}
1515 
1516 	error = gfs2_rindex_update(sdp);
1517 	if (error)
1518 		return error;
1519 
1520 	error = gfs2_qa_get(ndip);
1521 	if (error)
1522 		return error;
1523 
1524 	if (odip != ndip) {
1525 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1526 					   0, &r_gh);
1527 		if (error)
1528 			goto out;
1529 
1530 		if (S_ISDIR(ip->i_inode.i_mode)) {
1531 			dir_rename = 1;
1532 			/* don't move a directory into its subdir */
1533 			error = gfs2_ok_to_move(ip, ndip);
1534 			if (error)
1535 				goto out_gunlock_r;
1536 		}
1537 	}
1538 
1539 	num_gh = 1;
1540 	gfs2_holder_init(gfs2_inode_glock(odir), LM_ST_EXCLUSIVE, GL_ASYNC, ghs);
1541 	if (odip != ndip) {
1542 		gfs2_holder_init(gfs2_inode_glock(ndir), LM_ST_EXCLUSIVE,GL_ASYNC,
1543 				 ghs + num_gh);
1544 		num_gh++;
1545 	}
1546 	gfs2_holder_init(gfs2_inode_glock(&ip->i_inode), LM_ST_EXCLUSIVE,
1547 			 GL_ASYNC, ghs + num_gh);
1548 	num_gh++;
1549 
1550 	if (nip) {
1551 		gfs2_holder_init(gfs2_inode_glock(&nip->i_inode),
1552 				 LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh);
1553 		num_gh++;
1554 	}
1555 
1556 again:
1557 	for (x = 0; x < num_gh; x++) {
1558 		error = gfs2_glock_nq(ghs + x);
1559 		if (error)
1560 			goto out_gunlock;
1561 	}
1562 	error = gfs2_glock_async_wait(num_gh, ghs, retries);
1563 	if (error == -ESTALE) {
1564 		retries++;
1565 		goto again;
1566 	}
1567 	if (error)
1568 		goto out_gunlock;
1569 
1570 	if (nip) {
1571 		/* Grab the resource group glock for unlink flag twiddling.
1572 		 * This is the case where the target dinode already exists
1573 		 * so we unlink before doing the rename.
1574 		 */
1575 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1576 		if (!nrgd) {
1577 			error = -ENOENT;
1578 			goto out_gunlock;
1579 		}
1580 		error = gfs2_glock_nq_init(nrgd->rd_gl, LM_ST_EXCLUSIVE,
1581 					   LM_FLAG_NODE_SCOPE, &rd_gh);
1582 		if (error)
1583 			goto out_gunlock;
1584 	}
1585 
1586 	error = -ENOENT;
1587 	if (ip->i_inode.i_nlink == 0)
1588 		goto out_gunlock;
1589 
1590 	/* Check out the old directory */
1591 
1592 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1593 	if (error)
1594 		goto out_gunlock;
1595 
1596 	/* Check out the new directory */
1597 
1598 	if (nip) {
1599 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1600 		if (error)
1601 			goto out_gunlock;
1602 
1603 		if (nip->i_inode.i_nlink == 0) {
1604 			error = -EAGAIN;
1605 			goto out_gunlock;
1606 		}
1607 
1608 		if (S_ISDIR(nip->i_inode.i_mode)) {
1609 			if (nip->i_entries < 2) {
1610 				gfs2_consist_inode(nip);
1611 				error = -EIO;
1612 				goto out_gunlock;
1613 			}
1614 			if (nip->i_entries > 2) {
1615 				error = -ENOTEMPTY;
1616 				goto out_gunlock;
1617 			}
1618 		}
1619 	} else {
1620 		error = gfs2_permission(&nop_mnt_idmap, ndir,
1621 					MAY_WRITE | MAY_EXEC);
1622 		if (error)
1623 			goto out_gunlock;
1624 
1625 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1626 		switch (error) {
1627 		case -ENOENT:
1628 			error = 0;
1629 			break;
1630 		case 0:
1631 			error = -EEXIST;
1632 			goto out_gunlock;
1633 		default:
1634 			goto out_gunlock;
1635 		}
1636 
1637 		if (odip != ndip) {
1638 			if (!ndip->i_inode.i_nlink) {
1639 				error = -ENOENT;
1640 				goto out_gunlock;
1641 			}
1642 			if (ndip->i_entries == (u32)-1) {
1643 				error = -EFBIG;
1644 				goto out_gunlock;
1645 			}
1646 			if (S_ISDIR(ip->i_inode.i_mode) &&
1647 			    ndip->i_inode.i_nlink == (u32)-1) {
1648 				error = -EMLINK;
1649 				goto out_gunlock;
1650 			}
1651 		}
1652 	}
1653 
1654 	/* Check out the dir to be renamed */
1655 
1656 	if (dir_rename) {
1657 		error = gfs2_permission(&nop_mnt_idmap, d_inode(odentry),
1658 					MAY_WRITE);
1659 		if (error)
1660 			goto out_gunlock;
1661 	}
1662 
1663 	if (nip == NULL) {
1664 		error = gfs2_diradd_alloc_required(ndir, &ndentry->d_name, &da);
1665 		if (error)
1666 			goto out_gunlock;
1667 	}
1668 
1669 	if (da.nr_blocks) {
1670 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1671 		error = gfs2_quota_lock_check(ndip, &ap);
1672 		if (error)
1673 			goto out_gunlock;
1674 
1675 		error = gfs2_inplace_reserve(ndip, &ap);
1676 		if (error)
1677 			goto out_gunlock_q;
1678 
1679 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(ndip, &da, 4) +
1680 					 4 * RES_LEAF + 4, 0);
1681 		if (error)
1682 			goto out_ipreserv;
1683 	} else {
1684 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1685 					 5 * RES_LEAF + 4, 0);
1686 		if (error)
1687 			goto out_gunlock;
1688 	}
1689 
1690 	/* Remove the target file, if it exists */
1691 
1692 	if (nip)
1693 		error = gfs2_unlink_inode(ndip, ndentry);
1694 
1695 	error = update_moved_ino(ip, ndip, dir_rename);
1696 	if (error)
1697 		goto out_end_trans;
1698 
1699 	error = gfs2_dir_del(odip, odentry);
1700 	if (error)
1701 		goto out_end_trans;
1702 
1703 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip, &da);
1704 	if (error)
1705 		goto out_end_trans;
1706 
1707 out_end_trans:
1708 	gfs2_trans_end(sdp);
1709 out_ipreserv:
1710 	if (da.nr_blocks)
1711 		gfs2_inplace_release(ndip);
1712 out_gunlock_q:
1713 	if (da.nr_blocks)
1714 		gfs2_quota_unlock(ndip);
1715 out_gunlock:
1716 	gfs2_dir_no_add(&da);
1717 	if (gfs2_holder_initialized(&rd_gh))
1718 		gfs2_glock_dq_uninit(&rd_gh);
1719 
1720 	while (x--) {
1721 		if (gfs2_holder_queued(ghs + x))
1722 			gfs2_glock_dq(ghs + x);
1723 		gfs2_holder_uninit(ghs + x);
1724 	}
1725 out_gunlock_r:
1726 	if (gfs2_holder_initialized(&r_gh))
1727 		gfs2_glock_dq_uninit(&r_gh);
1728 out:
1729 	gfs2_qa_put(ndip);
1730 	return error;
1731 }
1732 
1733 /**
1734  * gfs2_exchange - exchange two files
1735  * @odir: Parent directory of old file name
1736  * @odentry: The old dentry of the file
1737  * @ndir: Parent directory of new file name
1738  * @ndentry: The new dentry of the file
1739  * @flags: The rename flags
1740  *
1741  * Returns: errno
1742  */
1743 
1744 static int gfs2_exchange(struct inode *odir, struct dentry *odentry,
1745 			 struct inode *ndir, struct dentry *ndentry,
1746 			 unsigned int flags)
1747 {
1748 	struct gfs2_inode *odip = GFS2_I(odir);
1749 	struct gfs2_inode *ndip = GFS2_I(ndir);
1750 	struct gfs2_inode *oip = GFS2_I(odentry->d_inode);
1751 	struct gfs2_inode *nip = GFS2_I(ndentry->d_inode);
1752 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1753 	struct gfs2_holder ghs[4], r_gh;
1754 	unsigned int num_gh;
1755 	unsigned int retries = 0, x;
1756 	umode_t old_mode = oip->i_inode.i_mode;
1757 	umode_t new_mode = nip->i_inode.i_mode;
1758 	int error;
1759 
1760 	gfs2_holder_mark_uninitialized(&r_gh);
1761 	error = gfs2_rindex_update(sdp);
1762 	if (error)
1763 		return error;
1764 
1765 	if (odip != ndip) {
1766 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1767 					   0, &r_gh);
1768 		if (error)
1769 			goto out;
1770 
1771 		if (S_ISDIR(old_mode)) {
1772 			/* don't move a directory into its subdir */
1773 			error = gfs2_ok_to_move(oip, ndip);
1774 			if (error)
1775 				goto out_gunlock_r;
1776 		}
1777 
1778 		if (S_ISDIR(new_mode)) {
1779 			/* don't move a directory into its subdir */
1780 			error = gfs2_ok_to_move(nip, odip);
1781 			if (error)
1782 				goto out_gunlock_r;
1783 		}
1784 	}
1785 
1786 	num_gh = 1;
1787 	gfs2_holder_init(gfs2_inode_glock(odir), LM_ST_EXCLUSIVE, GL_ASYNC,
1788 			 ghs);
1789 	if (odip != ndip) {
1790 		gfs2_holder_init(gfs2_inode_glock(ndir), LM_ST_EXCLUSIVE,
1791 				 GL_ASYNC, ghs + num_gh);
1792 		num_gh++;
1793 	}
1794 	gfs2_holder_init(gfs2_inode_glock(&oip->i_inode), LM_ST_EXCLUSIVE,
1795 			 GL_ASYNC, ghs + num_gh);
1796 	num_gh++;
1797 
1798 	gfs2_holder_init(gfs2_inode_glock(&nip->i_inode), LM_ST_EXCLUSIVE,
1799 			 GL_ASYNC, ghs + num_gh);
1800 	num_gh++;
1801 
1802 again:
1803 	for (x = 0; x < num_gh; x++) {
1804 		error = gfs2_glock_nq(ghs + x);
1805 		if (error)
1806 			goto out_gunlock;
1807 	}
1808 
1809 	error = gfs2_glock_async_wait(num_gh, ghs, retries);
1810 	if (error == -ESTALE) {
1811 		retries++;
1812 		goto again;
1813 	}
1814 	if (error)
1815 		goto out_gunlock;
1816 
1817 	error = -ENOENT;
1818 	if (oip->i_inode.i_nlink == 0 || nip->i_inode.i_nlink == 0)
1819 		goto out_gunlock;
1820 
1821 	error = gfs2_unlink_ok(odip, &odentry->d_name, oip);
1822 	if (error)
1823 		goto out_gunlock;
1824 	error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1825 	if (error)
1826 		goto out_gunlock;
1827 
1828 	if (S_ISDIR(old_mode)) {
1829 		error = gfs2_permission(&nop_mnt_idmap, odentry->d_inode,
1830 					MAY_WRITE);
1831 		if (error)
1832 			goto out_gunlock;
1833 	}
1834 	if (S_ISDIR(new_mode)) {
1835 		error = gfs2_permission(&nop_mnt_idmap, ndentry->d_inode,
1836 					MAY_WRITE);
1837 		if (error)
1838 			goto out_gunlock;
1839 	}
1840 	error = gfs2_trans_begin(sdp, 4 * RES_DINODE + 4 * RES_LEAF, 0);
1841 	if (error)
1842 		goto out_gunlock;
1843 
1844 	error = update_moved_ino(oip, ndip, S_ISDIR(old_mode));
1845 	if (error)
1846 		goto out_end_trans;
1847 
1848 	error = update_moved_ino(nip, odip, S_ISDIR(new_mode));
1849 	if (error)
1850 		goto out_end_trans;
1851 
1852 	error = gfs2_dir_mvino(ndip, &ndentry->d_name, oip,
1853 			       IF2DT(old_mode));
1854 	if (error)
1855 		goto out_end_trans;
1856 
1857 	error = gfs2_dir_mvino(odip, &odentry->d_name, nip,
1858 			       IF2DT(new_mode));
1859 	if (error)
1860 		goto out_end_trans;
1861 
1862 	if (odip != ndip) {
1863 		if (S_ISDIR(new_mode) && !S_ISDIR(old_mode)) {
1864 			inc_nlink(&odip->i_inode);
1865 			drop_nlink(&ndip->i_inode);
1866 		} else if (S_ISDIR(old_mode) && !S_ISDIR(new_mode)) {
1867 			inc_nlink(&ndip->i_inode);
1868 			drop_nlink(&odip->i_inode);
1869 		}
1870 	}
1871 	mark_inode_dirty(&ndip->i_inode);
1872 	if (odip != ndip)
1873 		mark_inode_dirty(&odip->i_inode);
1874 
1875 out_end_trans:
1876 	gfs2_trans_end(sdp);
1877 out_gunlock:
1878 	while (x--) {
1879 		if (gfs2_holder_queued(ghs + x))
1880 			gfs2_glock_dq(ghs + x);
1881 		gfs2_holder_uninit(ghs + x);
1882 	}
1883 out_gunlock_r:
1884 	if (gfs2_holder_initialized(&r_gh))
1885 		gfs2_glock_dq_uninit(&r_gh);
1886 out:
1887 	return error;
1888 }
1889 
1890 static int gfs2_rename2(const struct mnt_idmap *idmap, struct inode *odir,
1891 			struct dentry *odentry, struct inode *ndir,
1892 			struct dentry *ndentry, unsigned int flags)
1893 {
1894 	flags &= ~RENAME_NOREPLACE;
1895 
1896 	if (flags & ~RENAME_EXCHANGE)
1897 		return -EINVAL;
1898 
1899 	if (flags & RENAME_EXCHANGE)
1900 		return gfs2_exchange(odir, odentry, ndir, ndentry, flags);
1901 
1902 	return gfs2_rename(odir, odentry, ndir, ndentry);
1903 }
1904 
1905 /**
1906  * gfs2_get_link - Follow a symbolic link
1907  * @dentry: The dentry of the link
1908  * @inode: The inode of the link
1909  * @done: destructor for return value
1910  *
1911  * This can handle symlinks of any size.
1912  *
1913  * Returns: the link target on success, an ERR_PTR() on failure
1914  */
1915 
1916 static const char *gfs2_get_link(struct dentry *dentry,
1917 				 struct inode *inode,
1918 				 struct delayed_call *done)
1919 {
1920 	struct gfs2_glock *gl = gfs2_inode_glock(inode);
1921 	struct gfs2_inode *ip = GFS2_I(inode);
1922 	struct gfs2_holder i_gh;
1923 	struct buffer_head *dibh;
1924 	unsigned int size;
1925 	char *buf;
1926 	int error;
1927 
1928 	if (!dentry)
1929 		return ERR_PTR(-ECHILD);
1930 
1931 	gfs2_holder_init(gl, LM_ST_SHARED, 0, &i_gh);
1932 	error = gfs2_glock_nq(&i_gh);
1933 	if (error) {
1934 		gfs2_holder_uninit(&i_gh);
1935 		return ERR_PTR(error);
1936 	}
1937 
1938 	size = (unsigned int)i_size_read(&ip->i_inode);
1939 	if (size == 0) {
1940 		gfs2_consist_inode(ip);
1941 		buf = ERR_PTR(-EIO);
1942 		goto out;
1943 	}
1944 
1945 	error = gfs2_meta_inode_buffer(ip, &dibh);
1946 	if (error) {
1947 		buf = ERR_PTR(error);
1948 		goto out;
1949 	}
1950 
1951 	buf = kzalloc(size + 1, GFP_NOFS);
1952 	if (!buf)
1953 		buf = ERR_PTR(-ENOMEM);
1954 	else
1955 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1956 	brelse(dibh);
1957 out:
1958 	gfs2_glock_dq_uninit(&i_gh);
1959 	if (!IS_ERR(buf))
1960 		set_delayed_call(done, kfree_link, buf);
1961 	return buf;
1962 }
1963 
1964 /**
1965  * gfs2_permission
1966  * @idmap: idmap of the mount the inode was found from
1967  * @inode: The inode
1968  * @mask: The mask to be tested
1969  *
1970  * This may be called from the VFS directly, or from within GFS2 with the
1971  * inode locked, so we look to see if the glock is already locked and only
1972  * lock the glock if its not already been done.
1973  *
1974  * Returns: errno
1975  */
1976 
1977 int gfs2_permission(const struct mnt_idmap *idmap, struct inode *inode,
1978 		    int mask)
1979 {
1980 	int may_not_block = mask & MAY_NOT_BLOCK;
1981 	struct gfs2_inode *ip;
1982 	struct gfs2_holder i_gh;
1983 	struct gfs2_glock *gl;
1984 	int error;
1985 
1986 	gfs2_holder_mark_uninitialized(&i_gh);
1987 	ip = GFS2_I(inode);
1988 	gl = rcu_dereference_check(ip->i_gl, !may_not_block);
1989 	if (unlikely(!gl)) {
1990 		/* inode is getting torn down, must be RCU mode */
1991 		WARN_ON_ONCE(!may_not_block);
1992 		return -ECHILD;
1993         }
1994 	if (gfs2_glock_is_locked_by_me(gl) == NULL) {
1995 		if (may_not_block)
1996 			return -ECHILD;
1997 		error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1998 		if (error)
1999 			return error;
2000 	}
2001 
2002 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
2003 		error = -EPERM;
2004 	else
2005 		error = generic_permission(&nop_mnt_idmap, inode, mask);
2006 	if (gfs2_holder_initialized(&i_gh))
2007 		gfs2_glock_dq_uninit(&i_gh);
2008 
2009 	return error;
2010 }
2011 
2012 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
2013 {
2014 	setattr_copy(&nop_mnt_idmap, inode, attr);
2015 	mark_inode_dirty(inode);
2016 	return 0;
2017 }
2018 
2019 static int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
2020 {
2021 	int error;
2022 
2023 	if (current->journal_info)
2024 		return __gfs2_setattr_simple(inode, attr);
2025 
2026 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
2027 	if (error)
2028 		return error;
2029 
2030 	error = __gfs2_setattr_simple(inode, attr);
2031 	gfs2_trans_end(GFS2_SB(inode));
2032 	return error;
2033 }
2034 
2035 static int setattr_chown(struct inode *inode, struct iattr *attr)
2036 {
2037 	struct gfs2_inode *ip = GFS2_I(inode);
2038 	struct gfs2_sbd *sdp = GFS2_SB(inode);
2039 	kuid_t ouid, nuid;
2040 	kgid_t ogid, ngid;
2041 	int error;
2042 	struct gfs2_alloc_parms ap = {};
2043 
2044 	ouid = inode->i_uid;
2045 	ogid = inode->i_gid;
2046 	nuid = attr->ia_uid;
2047 	ngid = attr->ia_gid;
2048 
2049 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
2050 		ouid = nuid = NO_UID_QUOTA_CHANGE;
2051 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
2052 		ogid = ngid = NO_GID_QUOTA_CHANGE;
2053 	error = gfs2_qa_get(ip);
2054 	if (error)
2055 		return error;
2056 
2057 	error = gfs2_rindex_update(sdp);
2058 	if (error)
2059 		goto out;
2060 
2061 	error = gfs2_quota_lock(ip, nuid, ngid);
2062 	if (error)
2063 		goto out;
2064 
2065 	ap.target = gfs2_get_inode_blocks(&ip->i_inode);
2066 
2067 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
2068 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
2069 		error = gfs2_quota_check(ip, nuid, ngid, &ap);
2070 		if (error)
2071 			goto out_gunlock_q;
2072 	}
2073 
2074 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
2075 	if (error)
2076 		goto out_gunlock_q;
2077 
2078 	error = gfs2_setattr_simple(inode, attr);
2079 	if (error)
2080 		goto out_end_trans;
2081 
2082 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
2083 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
2084 		gfs2_quota_change(ip, -(s64)ap.target, ouid, ogid);
2085 		gfs2_quota_change(ip, ap.target, nuid, ngid);
2086 	}
2087 
2088 out_end_trans:
2089 	gfs2_trans_end(sdp);
2090 out_gunlock_q:
2091 	gfs2_quota_unlock(ip);
2092 out:
2093 	gfs2_qa_put(ip);
2094 	return error;
2095 }
2096 
2097 /**
2098  * gfs2_setattr - Change attributes on an inode
2099  * @idmap: idmap of the mount the inode was found from
2100  * @dentry: The dentry which is changing
2101  * @attr: The structure describing the change
2102  *
2103  * The VFS layer wants to change one or more of an inodes attributes.  Write
2104  * that change out to disk.
2105  *
2106  * Returns: errno
2107  */
2108 
2109 static int gfs2_setattr(const struct mnt_idmap *idmap,
2110 			struct dentry *dentry, struct iattr *attr)
2111 {
2112 	struct inode *inode = d_inode(dentry);
2113 	struct gfs2_glock *gl = gfs2_inode_glock(inode);
2114 	struct gfs2_inode *ip = GFS2_I(inode);
2115 	struct gfs2_holder i_gh;
2116 	int error;
2117 
2118 	error = gfs2_qa_get(ip);
2119 	if (error)
2120 		return error;
2121 
2122 	error = gfs2_glock_nq_init(gl, LM_ST_EXCLUSIVE, 0, &i_gh);
2123 	if (error)
2124 		goto out;
2125 
2126 	error = may_setattr(&nop_mnt_idmap, inode, attr->ia_valid);
2127 	if (error)
2128 		goto error;
2129 
2130 	error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
2131 	if (error)
2132 		goto error;
2133 
2134 	if (attr->ia_valid & ATTR_SIZE)
2135 		error = gfs2_setattr_size(inode, attr->ia_size);
2136 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
2137 		error = setattr_chown(inode, attr);
2138 	else {
2139 		error = gfs2_setattr_simple(inode, attr);
2140 		if (!error && attr->ia_valid & ATTR_MODE)
2141 			error = posix_acl_chmod(&nop_mnt_idmap, dentry,
2142 						inode->i_mode);
2143 	}
2144 
2145 error:
2146 	if (!error)
2147 		mark_inode_dirty(inode);
2148 	gfs2_glock_dq_uninit(&i_gh);
2149 out:
2150 	gfs2_qa_put(ip);
2151 	return error;
2152 }
2153 
2154 /**
2155  * gfs2_getattr - Read out an inode's attributes
2156  * @idmap: idmap of the mount the inode was found from
2157  * @path: Object to query
2158  * @stat: The inode's stats
2159  * @request_mask: Mask of STATX_xxx flags indicating the caller's interests
2160  * @flags: AT_STATX_xxx setting
2161  *
2162  * This may be called from the VFS directly, or from within GFS2 with the
2163  * inode locked, so we look to see if the glock is already locked and only
2164  * lock the glock if its not already been done. Note that its the NFS
2165  * readdirplus operation which causes this to be called (from filldir)
2166  * with the glock already held.
2167  *
2168  * Returns: errno
2169  */
2170 
2171 static int gfs2_getattr(const struct mnt_idmap *idmap,
2172 			const struct path *path, struct kstat *stat,
2173 			u32 request_mask, unsigned int flags)
2174 {
2175 	struct inode *inode = d_inode(path->dentry);
2176 	struct gfs2_glock *gl = gfs2_inode_glock(inode);
2177 	struct gfs2_inode *ip = GFS2_I(inode);
2178 	struct gfs2_holder gh;
2179 	u32 gfsflags;
2180 	int error;
2181 
2182 	gfs2_holder_mark_uninitialized(&gh);
2183 	if (gfs2_glock_is_locked_by_me(gl) == NULL) {
2184 		error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
2185 		if (error)
2186 			return error;
2187 	}
2188 
2189 	gfsflags = ip->i_diskflags;
2190 	if (gfsflags & GFS2_DIF_APPENDONLY)
2191 		stat->attributes |= STATX_ATTR_APPEND;
2192 	if (gfsflags & GFS2_DIF_IMMUTABLE)
2193 		stat->attributes |= STATX_ATTR_IMMUTABLE;
2194 
2195 	stat->attributes_mask |= (STATX_ATTR_APPEND |
2196 				  STATX_ATTR_COMPRESSED |
2197 				  STATX_ATTR_ENCRYPTED |
2198 				  STATX_ATTR_IMMUTABLE |
2199 				  STATX_ATTR_NODUMP);
2200 
2201 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
2202 
2203 	if (gfs2_holder_initialized(&gh))
2204 		gfs2_glock_dq_uninit(&gh);
2205 
2206 	return 0;
2207 }
2208 
2209 static bool fault_in_fiemap(struct fiemap_extent_info *fi)
2210 {
2211 	struct fiemap_extent __user *dest = fi->fi_extents_start;
2212 	size_t size = sizeof(*dest) * fi->fi_extents_max;
2213 
2214 	return fault_in_safe_writeable((char __user *)dest, size) == 0;
2215 }
2216 
2217 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2218 		       u64 start, u64 len)
2219 {
2220 	struct gfs2_glock *gl = gfs2_inode_glock(inode);
2221 	struct gfs2_holder gh;
2222 	int ret;
2223 
2224 	inode_lock_shared(inode);
2225 
2226 retry:
2227 	ret = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &gh);
2228 	if (ret)
2229 		goto out;
2230 
2231 	pagefault_disable();
2232 	ret = iomap_fiemap(inode, fieinfo, start, len, &gfs2_iomap_ops);
2233 	pagefault_enable();
2234 
2235 	gfs2_glock_dq_uninit(&gh);
2236 
2237 	if (ret == -EFAULT && fault_in_fiemap(fieinfo)) {
2238 		fieinfo->fi_extents_mapped = 0;
2239 		goto retry;
2240 	}
2241 
2242 out:
2243 	inode_unlock_shared(inode);
2244 	return ret;
2245 }
2246 
2247 loff_t gfs2_seek_data(struct file *file, loff_t offset)
2248 {
2249 	struct inode *inode = file->f_mapping->host;
2250 	struct gfs2_glock *gl = gfs2_inode_glock(inode);
2251 	struct gfs2_holder gh;
2252 	loff_t ret;
2253 
2254 	inode_lock_shared(inode);
2255 	ret = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &gh);
2256 	if (!ret)
2257 		ret = iomap_seek_data(inode, offset, &gfs2_iomap_ops);
2258 	gfs2_glock_dq_uninit(&gh);
2259 	inode_unlock_shared(inode);
2260 
2261 	if (ret < 0)
2262 		return ret;
2263 	return vfs_setpos(file, ret, inode->i_sb->s_maxbytes);
2264 }
2265 
2266 loff_t gfs2_seek_hole(struct file *file, loff_t offset)
2267 {
2268 	struct inode *inode = file->f_mapping->host;
2269 	struct gfs2_glock *gl = gfs2_inode_glock(inode);
2270 	struct gfs2_holder gh;
2271 	loff_t ret;
2272 
2273 	inode_lock_shared(inode);
2274 	ret = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &gh);
2275 	if (!ret)
2276 		ret = iomap_seek_hole(inode, offset, &gfs2_iomap_ops);
2277 	gfs2_glock_dq_uninit(&gh);
2278 	inode_unlock_shared(inode);
2279 
2280 	if (ret < 0)
2281 		return ret;
2282 	return vfs_setpos(file, ret, inode->i_sb->s_maxbytes);
2283 }
2284 
2285 static int gfs2_update_time(struct inode *inode, enum fs_update_time type,
2286 		unsigned int flags)
2287 {
2288 	struct gfs2_glock *gl = gfs2_inode_glock(inode);
2289 	struct gfs2_holder *gh;
2290 	int error;
2291 
2292 	if (flags & IOCB_NOWAIT)
2293 		return -EAGAIN;
2294 
2295 	gh = gfs2_glock_is_locked_by_me(gl);
2296 	if (gh && gl->gl_state != LM_ST_EXCLUSIVE) {
2297 		gfs2_glock_dq(gh);
2298 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, gh);
2299 		error = gfs2_glock_nq(gh);
2300 		if (error)
2301 			return error;
2302 	}
2303 	return generic_update_time(inode, type, flags);
2304 }
2305 
2306 static const struct inode_operations gfs2_file_iops = {
2307 	.permission = gfs2_permission,
2308 	.setattr = gfs2_setattr,
2309 	.getattr = gfs2_getattr,
2310 	.listxattr = gfs2_listxattr,
2311 	.fiemap = gfs2_fiemap,
2312 	.get_inode_acl = gfs2_get_acl,
2313 	.set_acl = gfs2_set_acl,
2314 	.update_time = gfs2_update_time,
2315 	.fileattr_get = gfs2_fileattr_get,
2316 	.fileattr_set = gfs2_fileattr_set,
2317 };
2318 
2319 static const struct inode_operations gfs2_dir_iops = {
2320 	.create = gfs2_create,
2321 	.lookup = gfs2_lookup,
2322 	.link = gfs2_link,
2323 	.unlink = gfs2_unlink,
2324 	.symlink = gfs2_symlink,
2325 	.mkdir = gfs2_mkdir,
2326 	.rmdir = gfs2_unlink,
2327 	.mknod = gfs2_mknod,
2328 	.rename = gfs2_rename2,
2329 	.permission = gfs2_permission,
2330 	.setattr = gfs2_setattr,
2331 	.getattr = gfs2_getattr,
2332 	.listxattr = gfs2_listxattr,
2333 	.fiemap = gfs2_fiemap,
2334 	.get_inode_acl = gfs2_get_acl,
2335 	.set_acl = gfs2_set_acl,
2336 	.update_time = gfs2_update_time,
2337 	.atomic_open = gfs2_atomic_open,
2338 	.fileattr_get = gfs2_fileattr_get,
2339 	.fileattr_set = gfs2_fileattr_set,
2340 };
2341 
2342 static const struct inode_operations gfs2_symlink_iops = {
2343 	.get_link = gfs2_get_link,
2344 	.permission = gfs2_permission,
2345 	.setattr = gfs2_setattr,
2346 	.getattr = gfs2_getattr,
2347 	.listxattr = gfs2_listxattr,
2348 	.fiemap = gfs2_fiemap,
2349 };
2350 
2351