xref: /linux/fs/gfs2/inode.c (revision 60063497a95e716c9a689af3be2687d261f115b4)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2011 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/namei.h>
15 #include <linux/mm.h>
16 #include <linux/xattr.h>
17 #include <linux/posix_acl.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/crc32.h>
20 #include <linux/fiemap.h>
21 #include <linux/security.h>
22 #include <asm/uaccess.h>
23 
24 #include "gfs2.h"
25 #include "incore.h"
26 #include "acl.h"
27 #include "bmap.h"
28 #include "dir.h"
29 #include "xattr.h"
30 #include "glock.h"
31 #include "inode.h"
32 #include "meta_io.h"
33 #include "quota.h"
34 #include "rgrp.h"
35 #include "trans.h"
36 #include "util.h"
37 #include "super.h"
38 #include "glops.h"
39 
40 struct gfs2_skip_data {
41 	u64 no_addr;
42 	int skipped;
43 	int non_block;
44 };
45 
46 static int iget_test(struct inode *inode, void *opaque)
47 {
48 	struct gfs2_inode *ip = GFS2_I(inode);
49 	struct gfs2_skip_data *data = opaque;
50 
51 	if (ip->i_no_addr == data->no_addr) {
52 		if (data->non_block &&
53 		    inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
54 			data->skipped = 1;
55 			return 0;
56 		}
57 		return 1;
58 	}
59 	return 0;
60 }
61 
62 static int iget_set(struct inode *inode, void *opaque)
63 {
64 	struct gfs2_inode *ip = GFS2_I(inode);
65 	struct gfs2_skip_data *data = opaque;
66 
67 	if (data->skipped)
68 		return -ENOENT;
69 	inode->i_ino = (unsigned long)(data->no_addr);
70 	ip->i_no_addr = data->no_addr;
71 	return 0;
72 }
73 
74 struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr, int non_block)
75 {
76 	unsigned long hash = (unsigned long)no_addr;
77 	struct gfs2_skip_data data;
78 
79 	data.no_addr = no_addr;
80 	data.skipped = 0;
81 	data.non_block = non_block;
82 	return ilookup5(sb, hash, iget_test, &data);
83 }
84 
85 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr,
86 			       int non_block)
87 {
88 	struct gfs2_skip_data data;
89 	unsigned long hash = (unsigned long)no_addr;
90 
91 	data.no_addr = no_addr;
92 	data.skipped = 0;
93 	data.non_block = non_block;
94 	return iget5_locked(sb, hash, iget_test, iget_set, &data);
95 }
96 
97 /**
98  * gfs2_set_iop - Sets inode operations
99  * @inode: The inode with correct i_mode filled in
100  *
101  * GFS2 lookup code fills in vfs inode contents based on info obtained
102  * from directory entry inside gfs2_inode_lookup().
103  */
104 
105 static void gfs2_set_iop(struct inode *inode)
106 {
107 	struct gfs2_sbd *sdp = GFS2_SB(inode);
108 	umode_t mode = inode->i_mode;
109 
110 	if (S_ISREG(mode)) {
111 		inode->i_op = &gfs2_file_iops;
112 		if (gfs2_localflocks(sdp))
113 			inode->i_fop = &gfs2_file_fops_nolock;
114 		else
115 			inode->i_fop = &gfs2_file_fops;
116 	} else if (S_ISDIR(mode)) {
117 		inode->i_op = &gfs2_dir_iops;
118 		if (gfs2_localflocks(sdp))
119 			inode->i_fop = &gfs2_dir_fops_nolock;
120 		else
121 			inode->i_fop = &gfs2_dir_fops;
122 	} else if (S_ISLNK(mode)) {
123 		inode->i_op = &gfs2_symlink_iops;
124 	} else {
125 		inode->i_op = &gfs2_file_iops;
126 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
127 	}
128 }
129 
130 /**
131  * gfs2_inode_lookup - Lookup an inode
132  * @sb: The super block
133  * @no_addr: The inode number
134  * @type: The type of the inode
135  * non_block: Can we block on inodes that are being freed?
136  *
137  * Returns: A VFS inode, or an error
138  */
139 
140 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
141 				u64 no_addr, u64 no_formal_ino, int non_block)
142 {
143 	struct inode *inode;
144 	struct gfs2_inode *ip;
145 	struct gfs2_glock *io_gl = NULL;
146 	int error;
147 
148 	inode = gfs2_iget(sb, no_addr, non_block);
149 	ip = GFS2_I(inode);
150 
151 	if (!inode)
152 		return ERR_PTR(-ENOBUFS);
153 
154 	if (inode->i_state & I_NEW) {
155 		struct gfs2_sbd *sdp = GFS2_SB(inode);
156 		ip->i_no_formal_ino = no_formal_ino;
157 
158 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
159 		if (unlikely(error))
160 			goto fail;
161 		ip->i_gl->gl_object = ip;
162 
163 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
164 		if (unlikely(error))
165 			goto fail_put;
166 
167 		set_bit(GIF_INVALID, &ip->i_flags);
168 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
169 		if (unlikely(error))
170 			goto fail_iopen;
171 
172 		ip->i_iopen_gh.gh_gl->gl_object = ip;
173 		gfs2_glock_put(io_gl);
174 		io_gl = NULL;
175 
176 		if (type == DT_UNKNOWN) {
177 			/* Inode glock must be locked already */
178 			error = gfs2_inode_refresh(GFS2_I(inode));
179 			if (error)
180 				goto fail_refresh;
181 		} else {
182 			inode->i_mode = DT2IF(type);
183 		}
184 
185 		gfs2_set_iop(inode);
186 		unlock_new_inode(inode);
187 	}
188 
189 	return inode;
190 
191 fail_refresh:
192 	ip->i_iopen_gh.gh_gl->gl_object = NULL;
193 	gfs2_glock_dq_uninit(&ip->i_iopen_gh);
194 fail_iopen:
195 	if (io_gl)
196 		gfs2_glock_put(io_gl);
197 fail_put:
198 	ip->i_gl->gl_object = NULL;
199 	gfs2_glock_put(ip->i_gl);
200 fail:
201 	iget_failed(inode);
202 	return ERR_PTR(error);
203 }
204 
205 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
206 				  u64 *no_formal_ino, unsigned int blktype)
207 {
208 	struct super_block *sb = sdp->sd_vfs;
209 	struct gfs2_holder i_gh;
210 	struct inode *inode = NULL;
211 	int error;
212 
213 	/* Must not read in block until block type is verified */
214 	error = gfs2_glock_nq_num(sdp, no_addr, &gfs2_inode_glops,
215 				  LM_ST_EXCLUSIVE, GL_SKIP, &i_gh);
216 	if (error)
217 		return ERR_PTR(error);
218 
219 	error = gfs2_check_blk_type(sdp, no_addr, blktype);
220 	if (error)
221 		goto fail;
222 
223 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0, 1);
224 	if (IS_ERR(inode))
225 		goto fail;
226 
227 	/* Two extra checks for NFS only */
228 	if (no_formal_ino) {
229 		error = -ESTALE;
230 		if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
231 			goto fail_iput;
232 
233 		error = -EIO;
234 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
235 			goto fail_iput;
236 
237 		error = 0;
238 	}
239 
240 fail:
241 	gfs2_glock_dq_uninit(&i_gh);
242 	return error ? ERR_PTR(error) : inode;
243 fail_iput:
244 	iput(inode);
245 	goto fail;
246 }
247 
248 
249 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
250 {
251 	struct qstr qstr;
252 	struct inode *inode;
253 	gfs2_str2qstr(&qstr, name);
254 	inode = gfs2_lookupi(dip, &qstr, 1);
255 	/* gfs2_lookupi has inconsistent callers: vfs
256 	 * related routines expect NULL for no entry found,
257 	 * gfs2_lookup_simple callers expect ENOENT
258 	 * and do not check for NULL.
259 	 */
260 	if (inode == NULL)
261 		return ERR_PTR(-ENOENT);
262 	else
263 		return inode;
264 }
265 
266 
267 /**
268  * gfs2_lookupi - Look up a filename in a directory and return its inode
269  * @d_gh: An initialized holder for the directory glock
270  * @name: The name of the inode to look for
271  * @is_root: If 1, ignore the caller's permissions
272  * @i_gh: An uninitialized holder for the new inode glock
273  *
274  * This can be called via the VFS filldir function when NFS is doing
275  * a readdirplus and the inode which its intending to stat isn't
276  * already in cache. In this case we must not take the directory glock
277  * again, since the readdir call will have already taken that lock.
278  *
279  * Returns: errno
280  */
281 
282 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
283 			   int is_root)
284 {
285 	struct super_block *sb = dir->i_sb;
286 	struct gfs2_inode *dip = GFS2_I(dir);
287 	struct gfs2_holder d_gh;
288 	int error = 0;
289 	struct inode *inode = NULL;
290 	int unlock = 0;
291 
292 	if (!name->len || name->len > GFS2_FNAMESIZE)
293 		return ERR_PTR(-ENAMETOOLONG);
294 
295 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
296 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
297 	     dir == sb->s_root->d_inode)) {
298 		igrab(dir);
299 		return dir;
300 	}
301 
302 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
303 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
304 		if (error)
305 			return ERR_PTR(error);
306 		unlock = 1;
307 	}
308 
309 	if (!is_root) {
310 		error = gfs2_permission(dir, MAY_EXEC);
311 		if (error)
312 			goto out;
313 	}
314 
315 	inode = gfs2_dir_search(dir, name);
316 	if (IS_ERR(inode))
317 		error = PTR_ERR(inode);
318 out:
319 	if (unlock)
320 		gfs2_glock_dq_uninit(&d_gh);
321 	if (error == -ENOENT)
322 		return NULL;
323 	return inode ? inode : ERR_PTR(error);
324 }
325 
326 /**
327  * create_ok - OK to create a new on-disk inode here?
328  * @dip:  Directory in which dinode is to be created
329  * @name:  Name of new dinode
330  * @mode:
331  *
332  * Returns: errno
333  */
334 
335 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
336 		     unsigned int mode)
337 {
338 	int error;
339 
340 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
341 	if (error)
342 		return error;
343 
344 	/*  Don't create entries in an unlinked directory  */
345 	if (!dip->i_inode.i_nlink)
346 		return -ENOENT;
347 
348 	error = gfs2_dir_check(&dip->i_inode, name, NULL);
349 	switch (error) {
350 	case -ENOENT:
351 		error = 0;
352 		break;
353 	case 0:
354 		return -EEXIST;
355 	default:
356 		return error;
357 	}
358 
359 	if (dip->i_entries == (u32)-1)
360 		return -EFBIG;
361 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
362 		return -EMLINK;
363 
364 	return 0;
365 }
366 
367 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
368 			       unsigned int *uid, unsigned int *gid)
369 {
370 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
371 	    (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
372 		if (S_ISDIR(*mode))
373 			*mode |= S_ISUID;
374 		else if (dip->i_inode.i_uid != current_fsuid())
375 			*mode &= ~07111;
376 		*uid = dip->i_inode.i_uid;
377 	} else
378 		*uid = current_fsuid();
379 
380 	if (dip->i_inode.i_mode & S_ISGID) {
381 		if (S_ISDIR(*mode))
382 			*mode |= S_ISGID;
383 		*gid = dip->i_inode.i_gid;
384 	} else
385 		*gid = current_fsgid();
386 }
387 
388 static int alloc_dinode(struct gfs2_inode *dip, u64 *no_addr, u64 *generation)
389 {
390 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
391 	int error;
392 
393 	if (gfs2_alloc_get(dip) == NULL)
394 		return -ENOMEM;
395 
396 	dip->i_alloc->al_requested = RES_DINODE;
397 	error = gfs2_inplace_reserve(dip);
398 	if (error)
399 		goto out;
400 
401 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
402 	if (error)
403 		goto out_ipreserv;
404 
405 	error = gfs2_alloc_di(dip, no_addr, generation);
406 
407 	gfs2_trans_end(sdp);
408 
409 out_ipreserv:
410 	gfs2_inplace_release(dip);
411 out:
412 	gfs2_alloc_put(dip);
413 	return error;
414 }
415 
416 static void gfs2_init_dir(struct buffer_head *dibh,
417 			  const struct gfs2_inode *parent)
418 {
419 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
420 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
421 
422 	gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
423 	dent->de_inum = di->di_num; /* already GFS2 endian */
424 	dent->de_type = cpu_to_be16(DT_DIR);
425 
426 	dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
427 	gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
428 	gfs2_inum_out(parent, dent);
429 	dent->de_type = cpu_to_be16(DT_DIR);
430 
431 }
432 
433 /**
434  * init_dinode - Fill in a new dinode structure
435  * @dip: The directory this inode is being created in
436  * @gl: The glock covering the new inode
437  * @inum: The inode number
438  * @mode: The file permissions
439  * @uid: The uid of the new inode
440  * @gid: The gid of the new inode
441  * @generation: The generation number of the new inode
442  * @dev: The device number (if a device node)
443  * @symname: The symlink destination (if a symlink)
444  * @size: The inode size (ignored for directories)
445  * @bhp: The buffer head (returned to caller)
446  *
447  */
448 
449 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
450 			const struct gfs2_inum_host *inum, unsigned int mode,
451 			unsigned int uid, unsigned int gid,
452 			const u64 *generation, dev_t dev, const char *symname,
453 			unsigned size, struct buffer_head **bhp)
454 {
455 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
456 	struct gfs2_dinode *di;
457 	struct buffer_head *dibh;
458 	struct timespec tv = CURRENT_TIME;
459 
460 	dibh = gfs2_meta_new(gl, inum->no_addr);
461 	gfs2_trans_add_bh(gl, dibh, 1);
462 	gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
463 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
464 	di = (struct gfs2_dinode *)dibh->b_data;
465 
466 	di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
467 	di->di_num.no_addr = cpu_to_be64(inum->no_addr);
468 	di->di_mode = cpu_to_be32(mode);
469 	di->di_uid = cpu_to_be32(uid);
470 	di->di_gid = cpu_to_be32(gid);
471 	di->di_nlink = 0;
472 	di->di_size = cpu_to_be64(size);
473 	di->di_blocks = cpu_to_be64(1);
474 	di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(tv.tv_sec);
475 	di->di_major = cpu_to_be32(MAJOR(dev));
476 	di->di_minor = cpu_to_be32(MINOR(dev));
477 	di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
478 	di->di_generation = cpu_to_be64(*generation);
479 	di->di_flags = 0;
480 	di->__pad1 = 0;
481 	di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
482 	di->di_height = 0;
483 	di->__pad2 = 0;
484 	di->__pad3 = 0;
485 	di->di_depth = 0;
486 	di->di_entries = 0;
487 	memset(&di->__pad4, 0, sizeof(di->__pad4));
488 	di->di_eattr = 0;
489 	di->di_atime_nsec = cpu_to_be32(tv.tv_nsec);
490 	di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec);
491 	di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec);
492 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
493 
494 	switch(mode & S_IFMT) {
495 	case S_IFREG:
496 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
497 		    gfs2_tune_get(sdp, gt_new_files_jdata))
498 			di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
499 		break;
500 	case S_IFDIR:
501 		di->di_flags |= cpu_to_be32(dip->i_diskflags &
502 					    GFS2_DIF_INHERIT_JDATA);
503 		di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
504 		di->di_size = cpu_to_be64(sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode));
505 		di->di_entries = cpu_to_be32(2);
506 		gfs2_init_dir(dibh, dip);
507 		break;
508 	case S_IFLNK:
509 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, size);
510 		break;
511 	}
512 
513 	set_buffer_uptodate(dibh);
514 
515 	*bhp = dibh;
516 }
517 
518 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
519 		       unsigned int mode, const struct gfs2_inum_host *inum,
520 		       const u64 *generation, dev_t dev, const char *symname,
521 		       unsigned int size, struct buffer_head **bhp)
522 {
523 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
524 	unsigned int uid, gid;
525 	int error;
526 
527 	munge_mode_uid_gid(dip, &mode, &uid, &gid);
528 	if (!gfs2_alloc_get(dip))
529 		return -ENOMEM;
530 
531 	error = gfs2_quota_lock(dip, uid, gid);
532 	if (error)
533 		goto out;
534 
535 	error = gfs2_quota_check(dip, uid, gid);
536 	if (error)
537 		goto out_quota;
538 
539 	error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
540 	if (error)
541 		goto out_quota;
542 
543 	init_dinode(dip, gl, inum, mode, uid, gid, generation, dev, symname, size, bhp);
544 	gfs2_quota_change(dip, +1, uid, gid);
545 	gfs2_trans_end(sdp);
546 
547 out_quota:
548 	gfs2_quota_unlock(dip);
549 out:
550 	gfs2_alloc_put(dip);
551 	return error;
552 }
553 
554 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
555 		       struct gfs2_inode *ip)
556 {
557 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
558 	struct gfs2_alloc *al;
559 	int alloc_required;
560 	struct buffer_head *dibh;
561 	int error;
562 
563 	al = gfs2_alloc_get(dip);
564 	if (!al)
565 		return -ENOMEM;
566 
567 	error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
568 	if (error)
569 		goto fail;
570 
571 	error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
572 	if (alloc_required < 0)
573 		goto fail_quota_locks;
574 	if (alloc_required) {
575 		error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
576 		if (error)
577 			goto fail_quota_locks;
578 
579 		al->al_requested = sdp->sd_max_dirres;
580 
581 		error = gfs2_inplace_reserve(dip);
582 		if (error)
583 			goto fail_quota_locks;
584 
585 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
586 					 al->al_rgd->rd_length +
587 					 2 * RES_DINODE +
588 					 RES_STATFS + RES_QUOTA, 0);
589 		if (error)
590 			goto fail_ipreserv;
591 	} else {
592 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
593 		if (error)
594 			goto fail_quota_locks;
595 	}
596 
597 	error = gfs2_dir_add(&dip->i_inode, name, ip);
598 	if (error)
599 		goto fail_end_trans;
600 
601 	error = gfs2_meta_inode_buffer(ip, &dibh);
602 	if (error)
603 		goto fail_end_trans;
604 	inc_nlink(&ip->i_inode);
605 	if (S_ISDIR(ip->i_inode.i_mode))
606 		inc_nlink(&ip->i_inode);
607 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
608 	gfs2_dinode_out(ip, dibh->b_data);
609 	brelse(dibh);
610 	return 0;
611 
612 fail_end_trans:
613 	gfs2_trans_end(sdp);
614 
615 fail_ipreserv:
616 	if (dip->i_alloc->al_rgd)
617 		gfs2_inplace_release(dip);
618 
619 fail_quota_locks:
620 	gfs2_quota_unlock(dip);
621 
622 fail:
623 	gfs2_alloc_put(dip);
624 	return error;
625 }
626 
627 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
628 			      const struct qstr *qstr)
629 {
630 	int err;
631 	size_t len;
632 	void *value;
633 	char *name;
634 
635 	err = security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
636 					   &name, &value, &len);
637 
638 	if (err) {
639 		if (err == -EOPNOTSUPP)
640 			return 0;
641 		return err;
642 	}
643 
644 	err = __gfs2_xattr_set(&ip->i_inode, name, value, len, 0,
645 			       GFS2_EATYPE_SECURITY);
646 	kfree(value);
647 	kfree(name);
648 
649 	return err;
650 }
651 
652 /**
653  * gfs2_create_inode - Create a new inode
654  * @dir: The parent directory
655  * @dentry: The new dentry
656  * @mode: The permissions on the new inode
657  * @dev: For device nodes, this is the device number
658  * @symname: For symlinks, this is the link destination
659  * @size: The initial size of the inode (ignored for directories)
660  *
661  * Returns: 0 on success, or error code
662  */
663 
664 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
665 			     unsigned int mode, dev_t dev, const char *symname,
666 			     unsigned int size)
667 {
668 	const struct qstr *name = &dentry->d_name;
669 	struct gfs2_holder ghs[2];
670 	struct inode *inode = NULL;
671 	struct gfs2_inode *dip = GFS2_I(dir);
672 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
673 	struct gfs2_inum_host inum = { .no_addr = 0, .no_formal_ino = 0 };
674 	int error;
675 	u64 generation;
676 	struct buffer_head *bh = NULL;
677 
678 	if (!name->len || name->len > GFS2_FNAMESIZE)
679 		return -ENAMETOOLONG;
680 
681 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
682 	if (error)
683 		goto fail;
684 
685 	error = create_ok(dip, name, mode);
686 	if (error)
687 		goto fail_gunlock;
688 
689 	error = alloc_dinode(dip, &inum.no_addr, &generation);
690 	if (error)
691 		goto fail_gunlock;
692 	inum.no_formal_ino = generation;
693 
694 	error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops,
695 				  LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
696 	if (error)
697 		goto fail_gunlock;
698 
699 	error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev, symname, size, &bh);
700 	if (error)
701 		goto fail_gunlock2;
702 
703 	inode = gfs2_inode_lookup(dir->i_sb, IF2DT(mode), inum.no_addr,
704 				  inum.no_formal_ino, 0);
705 	if (IS_ERR(inode))
706 		goto fail_gunlock2;
707 
708 	error = gfs2_inode_refresh(GFS2_I(inode));
709 	if (error)
710 		goto fail_gunlock2;
711 
712 	error = gfs2_acl_create(dip, inode);
713 	if (error)
714 		goto fail_gunlock2;
715 
716 	error = gfs2_security_init(dip, GFS2_I(inode), name);
717 	if (error)
718 		goto fail_gunlock2;
719 
720 	error = link_dinode(dip, name, GFS2_I(inode));
721 	if (error)
722 		goto fail_gunlock2;
723 
724 	if (bh)
725 		brelse(bh);
726 
727 	gfs2_trans_end(sdp);
728 	if (dip->i_alloc->al_rgd)
729 		gfs2_inplace_release(dip);
730 	gfs2_quota_unlock(dip);
731 	gfs2_alloc_put(dip);
732 	gfs2_glock_dq_uninit_m(2, ghs);
733 	mark_inode_dirty(inode);
734 	d_instantiate(dentry, inode);
735 	return 0;
736 
737 fail_gunlock2:
738 	gfs2_glock_dq_uninit(ghs + 1);
739 	if (inode && !IS_ERR(inode))
740 		iput(inode);
741 fail_gunlock:
742 	gfs2_glock_dq_uninit(ghs);
743 fail:
744 	if (bh)
745 		brelse(bh);
746 	return error;
747 }
748 
749 /**
750  * gfs2_create - Create a file
751  * @dir: The directory in which to create the file
752  * @dentry: The dentry of the new file
753  * @mode: The mode of the new file
754  *
755  * Returns: errno
756  */
757 
758 static int gfs2_create(struct inode *dir, struct dentry *dentry,
759 		       int mode, struct nameidata *nd)
760 {
761 	struct inode *inode;
762 	int ret;
763 
764 	for (;;) {
765 		ret = gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0);
766 		if (ret != -EEXIST || (nd && (nd->flags & LOOKUP_EXCL)))
767 			return ret;
768 
769 		inode = gfs2_lookupi(dir, &dentry->d_name, 0);
770 		if (inode) {
771 			if (!IS_ERR(inode))
772 				break;
773 			return PTR_ERR(inode);
774 		}
775 	}
776 
777 	d_instantiate(dentry, inode);
778 	return 0;
779 }
780 
781 /**
782  * gfs2_lookup - Look up a filename in a directory and return its inode
783  * @dir: The directory inode
784  * @dentry: The dentry of the new inode
785  * @nd: passed from Linux VFS, ignored by us
786  *
787  * Called by the VFS layer. Lock dir and call gfs2_lookupi()
788  *
789  * Returns: errno
790  */
791 
792 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
793 				  struct nameidata *nd)
794 {
795 	struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
796 	if (inode && !IS_ERR(inode)) {
797 		struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
798 		struct gfs2_holder gh;
799 		int error;
800 		error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
801 		if (error) {
802 			iput(inode);
803 			return ERR_PTR(error);
804 		}
805 		gfs2_glock_dq_uninit(&gh);
806 	}
807 	return d_splice_alias(inode, dentry);
808 }
809 
810 /**
811  * gfs2_link - Link to a file
812  * @old_dentry: The inode to link
813  * @dir: Add link to this directory
814  * @dentry: The name of the link
815  *
816  * Link the inode in "old_dentry" into the directory "dir" with the
817  * name in "dentry".
818  *
819  * Returns: errno
820  */
821 
822 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
823 		     struct dentry *dentry)
824 {
825 	struct gfs2_inode *dip = GFS2_I(dir);
826 	struct gfs2_sbd *sdp = GFS2_SB(dir);
827 	struct inode *inode = old_dentry->d_inode;
828 	struct gfs2_inode *ip = GFS2_I(inode);
829 	struct gfs2_holder ghs[2];
830 	struct buffer_head *dibh;
831 	int alloc_required;
832 	int error;
833 
834 	if (S_ISDIR(inode->i_mode))
835 		return -EPERM;
836 
837 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
838 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
839 
840 	error = gfs2_glock_nq(ghs); /* parent */
841 	if (error)
842 		goto out_parent;
843 
844 	error = gfs2_glock_nq(ghs + 1); /* child */
845 	if (error)
846 		goto out_child;
847 
848 	error = -ENOENT;
849 	if (inode->i_nlink == 0)
850 		goto out_gunlock;
851 
852 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
853 	if (error)
854 		goto out_gunlock;
855 
856 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
857 	switch (error) {
858 	case -ENOENT:
859 		break;
860 	case 0:
861 		error = -EEXIST;
862 	default:
863 		goto out_gunlock;
864 	}
865 
866 	error = -EINVAL;
867 	if (!dip->i_inode.i_nlink)
868 		goto out_gunlock;
869 	error = -EFBIG;
870 	if (dip->i_entries == (u32)-1)
871 		goto out_gunlock;
872 	error = -EPERM;
873 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
874 		goto out_gunlock;
875 	error = -EINVAL;
876 	if (!ip->i_inode.i_nlink)
877 		goto out_gunlock;
878 	error = -EMLINK;
879 	if (ip->i_inode.i_nlink == (u32)-1)
880 		goto out_gunlock;
881 
882 	alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
883 	if (error < 0)
884 		goto out_gunlock;
885 	error = 0;
886 
887 	if (alloc_required) {
888 		struct gfs2_alloc *al = gfs2_alloc_get(dip);
889 		if (!al) {
890 			error = -ENOMEM;
891 			goto out_gunlock;
892 		}
893 
894 		error = gfs2_quota_lock_check(dip);
895 		if (error)
896 			goto out_alloc;
897 
898 		al->al_requested = sdp->sd_max_dirres;
899 
900 		error = gfs2_inplace_reserve(dip);
901 		if (error)
902 			goto out_gunlock_q;
903 
904 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
905 					 gfs2_rg_blocks(al) +
906 					 2 * RES_DINODE + RES_STATFS +
907 					 RES_QUOTA, 0);
908 		if (error)
909 			goto out_ipres;
910 	} else {
911 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
912 		if (error)
913 			goto out_ipres;
914 	}
915 
916 	error = gfs2_meta_inode_buffer(ip, &dibh);
917 	if (error)
918 		goto out_end_trans;
919 
920 	error = gfs2_dir_add(dir, &dentry->d_name, ip);
921 	if (error)
922 		goto out_brelse;
923 
924 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
925 	inc_nlink(&ip->i_inode);
926 	ip->i_inode.i_ctime = CURRENT_TIME;
927 	gfs2_dinode_out(ip, dibh->b_data);
928 	mark_inode_dirty(&ip->i_inode);
929 
930 out_brelse:
931 	brelse(dibh);
932 out_end_trans:
933 	gfs2_trans_end(sdp);
934 out_ipres:
935 	if (alloc_required)
936 		gfs2_inplace_release(dip);
937 out_gunlock_q:
938 	if (alloc_required)
939 		gfs2_quota_unlock(dip);
940 out_alloc:
941 	if (alloc_required)
942 		gfs2_alloc_put(dip);
943 out_gunlock:
944 	gfs2_glock_dq(ghs + 1);
945 out_child:
946 	gfs2_glock_dq(ghs);
947 out_parent:
948 	gfs2_holder_uninit(ghs);
949 	gfs2_holder_uninit(ghs + 1);
950 	if (!error) {
951 		ihold(inode);
952 		d_instantiate(dentry, inode);
953 		mark_inode_dirty(inode);
954 	}
955 	return error;
956 }
957 
958 /*
959  * gfs2_unlink_ok - check to see that a inode is still in a directory
960  * @dip: the directory
961  * @name: the name of the file
962  * @ip: the inode
963  *
964  * Assumes that the lock on (at least) @dip is held.
965  *
966  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
967  */
968 
969 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
970 			  const struct gfs2_inode *ip)
971 {
972 	int error;
973 
974 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
975 		return -EPERM;
976 
977 	if ((dip->i_inode.i_mode & S_ISVTX) &&
978 	    dip->i_inode.i_uid != current_fsuid() &&
979 	    ip->i_inode.i_uid != current_fsuid() && !capable(CAP_FOWNER))
980 		return -EPERM;
981 
982 	if (IS_APPEND(&dip->i_inode))
983 		return -EPERM;
984 
985 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
986 	if (error)
987 		return error;
988 
989 	error = gfs2_dir_check(&dip->i_inode, name, ip);
990 	if (error)
991 		return error;
992 
993 	return 0;
994 }
995 
996 /**
997  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
998  * @dip: The parent directory
999  * @name: The name of the entry in the parent directory
1000  * @bh: The inode buffer for the inode to be removed
1001  * @inode: The inode to be removed
1002  *
1003  * Called with all the locks and in a transaction. This will only be
1004  * called for a directory after it has been checked to ensure it is empty.
1005  *
1006  * Returns: 0 on success, or an error
1007  */
1008 
1009 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1010 			     const struct dentry *dentry,
1011 			     struct buffer_head *bh)
1012 {
1013 	struct inode *inode = dentry->d_inode;
1014 	struct gfs2_inode *ip = GFS2_I(inode);
1015 	int error;
1016 
1017 	error = gfs2_dir_del(dip, dentry);
1018 	if (error)
1019 		return error;
1020 
1021 	ip->i_entries = 0;
1022 	inode->i_ctime = CURRENT_TIME;
1023 	if (S_ISDIR(inode->i_mode))
1024 		clear_nlink(inode);
1025 	else
1026 		drop_nlink(inode);
1027 	gfs2_trans_add_bh(ip->i_gl, bh, 1);
1028 	gfs2_dinode_out(ip, bh->b_data);
1029 	mark_inode_dirty(inode);
1030 	if (inode->i_nlink == 0)
1031 		gfs2_unlink_di(inode);
1032 	return 0;
1033 }
1034 
1035 
1036 /**
1037  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1038  * @dir: The inode of the directory containing the inode to unlink
1039  * @dentry: The file itself
1040  *
1041  * This routine uses the type of the inode as a flag to figure out
1042  * whether this is an unlink or an rmdir.
1043  *
1044  * Returns: errno
1045  */
1046 
1047 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1048 {
1049 	struct gfs2_inode *dip = GFS2_I(dir);
1050 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1051 	struct inode *inode = dentry->d_inode;
1052 	struct gfs2_inode *ip = GFS2_I(inode);
1053 	struct buffer_head *bh;
1054 	struct gfs2_holder ghs[3];
1055 	struct gfs2_rgrpd *rgd;
1056 	struct gfs2_holder ri_gh;
1057 	int error;
1058 
1059 	error = gfs2_rindex_hold(sdp, &ri_gh);
1060 	if (error)
1061 		return error;
1062 
1063 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1064 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1065 
1066 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
1067 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1068 
1069 
1070 	error = gfs2_glock_nq(ghs); /* parent */
1071 	if (error)
1072 		goto out_parent;
1073 
1074 	error = gfs2_glock_nq(ghs + 1); /* child */
1075 	if (error)
1076 		goto out_child;
1077 
1078 	error = -ENOENT;
1079 	if (inode->i_nlink == 0)
1080 		goto out_rgrp;
1081 
1082 	if (S_ISDIR(inode->i_mode)) {
1083 		error = -ENOTEMPTY;
1084 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1085 			goto out_rgrp;
1086 	}
1087 
1088 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1089 	if (error)
1090 		goto out_rgrp;
1091 
1092 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1093 	if (error)
1094 		goto out_gunlock;
1095 
1096 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1097 	if (error)
1098 		goto out_gunlock;
1099 
1100 	error = gfs2_meta_inode_buffer(ip, &bh);
1101 	if (error)
1102 		goto out_end_trans;
1103 
1104 	error = gfs2_unlink_inode(dip, dentry, bh);
1105 	brelse(bh);
1106 
1107 out_end_trans:
1108 	gfs2_trans_end(sdp);
1109 out_gunlock:
1110 	gfs2_glock_dq(ghs + 2);
1111 out_rgrp:
1112 	gfs2_holder_uninit(ghs + 2);
1113 	gfs2_glock_dq(ghs + 1);
1114 out_child:
1115 	gfs2_holder_uninit(ghs + 1);
1116 	gfs2_glock_dq(ghs);
1117 out_parent:
1118 	gfs2_holder_uninit(ghs);
1119 	gfs2_glock_dq_uninit(&ri_gh);
1120 	return error;
1121 }
1122 
1123 /**
1124  * gfs2_symlink - Create a symlink
1125  * @dir: The directory to create the symlink in
1126  * @dentry: The dentry to put the symlink in
1127  * @symname: The thing which the link points to
1128  *
1129  * Returns: errno
1130  */
1131 
1132 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1133 			const char *symname)
1134 {
1135 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1136 	unsigned int size;
1137 
1138 	size = strlen(symname);
1139 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1140 		return -ENAMETOOLONG;
1141 
1142 	return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size);
1143 }
1144 
1145 /**
1146  * gfs2_mkdir - Make a directory
1147  * @dir: The parent directory of the new one
1148  * @dentry: The dentry of the new directory
1149  * @mode: The mode of the new directory
1150  *
1151  * Returns: errno
1152  */
1153 
1154 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1155 {
1156 	return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, 0);
1157 }
1158 
1159 /**
1160  * gfs2_mknod - Make a special file
1161  * @dir: The directory in which the special file will reside
1162  * @dentry: The dentry of the special file
1163  * @mode: The mode of the special file
1164  * @dev: The device specification of the special file
1165  *
1166  */
1167 
1168 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, int mode,
1169 		      dev_t dev)
1170 {
1171 	return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0);
1172 }
1173 
1174 /*
1175  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1176  * @this: move this
1177  * @to: to here
1178  *
1179  * Follow @to back to the root and make sure we don't encounter @this
1180  * Assumes we already hold the rename lock.
1181  *
1182  * Returns: errno
1183  */
1184 
1185 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1186 {
1187 	struct inode *dir = &to->i_inode;
1188 	struct super_block *sb = dir->i_sb;
1189 	struct inode *tmp;
1190 	int error = 0;
1191 
1192 	igrab(dir);
1193 
1194 	for (;;) {
1195 		if (dir == &this->i_inode) {
1196 			error = -EINVAL;
1197 			break;
1198 		}
1199 		if (dir == sb->s_root->d_inode) {
1200 			error = 0;
1201 			break;
1202 		}
1203 
1204 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1205 		if (IS_ERR(tmp)) {
1206 			error = PTR_ERR(tmp);
1207 			break;
1208 		}
1209 
1210 		iput(dir);
1211 		dir = tmp;
1212 	}
1213 
1214 	iput(dir);
1215 
1216 	return error;
1217 }
1218 
1219 /**
1220  * gfs2_rename - Rename a file
1221  * @odir: Parent directory of old file name
1222  * @odentry: The old dentry of the file
1223  * @ndir: Parent directory of new file name
1224  * @ndentry: The new dentry of the file
1225  *
1226  * Returns: errno
1227  */
1228 
1229 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1230 		       struct inode *ndir, struct dentry *ndentry)
1231 {
1232 	struct gfs2_inode *odip = GFS2_I(odir);
1233 	struct gfs2_inode *ndip = GFS2_I(ndir);
1234 	struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1235 	struct gfs2_inode *nip = NULL;
1236 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1237 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, }, ri_gh;
1238 	struct gfs2_rgrpd *nrgd;
1239 	unsigned int num_gh;
1240 	int dir_rename = 0;
1241 	int alloc_required = 0;
1242 	unsigned int x;
1243 	int error;
1244 
1245 	if (ndentry->d_inode) {
1246 		nip = GFS2_I(ndentry->d_inode);
1247 		if (ip == nip)
1248 			return 0;
1249 	}
1250 
1251 	error = gfs2_rindex_hold(sdp, &ri_gh);
1252 	if (error)
1253 		return error;
1254 
1255 	if (odip != ndip) {
1256 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1257 					   0, &r_gh);
1258 		if (error)
1259 			goto out;
1260 
1261 		if (S_ISDIR(ip->i_inode.i_mode)) {
1262 			dir_rename = 1;
1263 			/* don't move a dirctory into it's subdir */
1264 			error = gfs2_ok_to_move(ip, ndip);
1265 			if (error)
1266 				goto out_gunlock_r;
1267 		}
1268 	}
1269 
1270 	num_gh = 1;
1271 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1272 	if (odip != ndip) {
1273 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1274 		num_gh++;
1275 	}
1276 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1277 	num_gh++;
1278 
1279 	if (nip) {
1280 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1281 		num_gh++;
1282 		/* grab the resource lock for unlink flag twiddling
1283 		 * this is the case of the target file already existing
1284 		 * so we unlink before doing the rename
1285 		 */
1286 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr);
1287 		if (nrgd)
1288 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1289 	}
1290 
1291 	for (x = 0; x < num_gh; x++) {
1292 		error = gfs2_glock_nq(ghs + x);
1293 		if (error)
1294 			goto out_gunlock;
1295 	}
1296 
1297 	error = -ENOENT;
1298 	if (ip->i_inode.i_nlink == 0)
1299 		goto out_gunlock;
1300 
1301 	/* Check out the old directory */
1302 
1303 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1304 	if (error)
1305 		goto out_gunlock;
1306 
1307 	/* Check out the new directory */
1308 
1309 	if (nip) {
1310 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1311 		if (error)
1312 			goto out_gunlock;
1313 
1314 		if (nip->i_inode.i_nlink == 0) {
1315 			error = -EAGAIN;
1316 			goto out_gunlock;
1317 		}
1318 
1319 		if (S_ISDIR(nip->i_inode.i_mode)) {
1320 			if (nip->i_entries < 2) {
1321 				gfs2_consist_inode(nip);
1322 				error = -EIO;
1323 				goto out_gunlock;
1324 			}
1325 			if (nip->i_entries > 2) {
1326 				error = -ENOTEMPTY;
1327 				goto out_gunlock;
1328 			}
1329 		}
1330 	} else {
1331 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1332 		if (error)
1333 			goto out_gunlock;
1334 
1335 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1336 		switch (error) {
1337 		case -ENOENT:
1338 			error = 0;
1339 			break;
1340 		case 0:
1341 			error = -EEXIST;
1342 		default:
1343 			goto out_gunlock;
1344 		};
1345 
1346 		if (odip != ndip) {
1347 			if (!ndip->i_inode.i_nlink) {
1348 				error = -ENOENT;
1349 				goto out_gunlock;
1350 			}
1351 			if (ndip->i_entries == (u32)-1) {
1352 				error = -EFBIG;
1353 				goto out_gunlock;
1354 			}
1355 			if (S_ISDIR(ip->i_inode.i_mode) &&
1356 			    ndip->i_inode.i_nlink == (u32)-1) {
1357 				error = -EMLINK;
1358 				goto out_gunlock;
1359 			}
1360 		}
1361 	}
1362 
1363 	/* Check out the dir to be renamed */
1364 
1365 	if (dir_rename) {
1366 		error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1367 		if (error)
1368 			goto out_gunlock;
1369 	}
1370 
1371 	if (nip == NULL)
1372 		alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1373 	error = alloc_required;
1374 	if (error < 0)
1375 		goto out_gunlock;
1376 	error = 0;
1377 
1378 	if (alloc_required) {
1379 		struct gfs2_alloc *al = gfs2_alloc_get(ndip);
1380 		if (!al) {
1381 			error = -ENOMEM;
1382 			goto out_gunlock;
1383 		}
1384 
1385 		error = gfs2_quota_lock_check(ndip);
1386 		if (error)
1387 			goto out_alloc;
1388 
1389 		al->al_requested = sdp->sd_max_dirres;
1390 
1391 		error = gfs2_inplace_reserve_ri(ndip);
1392 		if (error)
1393 			goto out_gunlock_q;
1394 
1395 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1396 					 gfs2_rg_blocks(al) +
1397 					 4 * RES_DINODE + 4 * RES_LEAF +
1398 					 RES_STATFS + RES_QUOTA + 4, 0);
1399 		if (error)
1400 			goto out_ipreserv;
1401 	} else {
1402 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1403 					 5 * RES_LEAF + 4, 0);
1404 		if (error)
1405 			goto out_gunlock;
1406 	}
1407 
1408 	/* Remove the target file, if it exists */
1409 
1410 	if (nip) {
1411 		struct buffer_head *bh;
1412 		error = gfs2_meta_inode_buffer(nip, &bh);
1413 		if (error)
1414 			goto out_end_trans;
1415 		error = gfs2_unlink_inode(ndip, ndentry, bh);
1416 		brelse(bh);
1417 	}
1418 
1419 	if (dir_rename) {
1420 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1421 		if (error)
1422 			goto out_end_trans;
1423 	} else {
1424 		struct buffer_head *dibh;
1425 		error = gfs2_meta_inode_buffer(ip, &dibh);
1426 		if (error)
1427 			goto out_end_trans;
1428 		ip->i_inode.i_ctime = CURRENT_TIME;
1429 		gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1430 		gfs2_dinode_out(ip, dibh->b_data);
1431 		brelse(dibh);
1432 	}
1433 
1434 	error = gfs2_dir_del(odip, odentry);
1435 	if (error)
1436 		goto out_end_trans;
1437 
1438 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1439 	if (error)
1440 		goto out_end_trans;
1441 
1442 out_end_trans:
1443 	gfs2_trans_end(sdp);
1444 out_ipreserv:
1445 	if (alloc_required)
1446 		gfs2_inplace_release(ndip);
1447 out_gunlock_q:
1448 	if (alloc_required)
1449 		gfs2_quota_unlock(ndip);
1450 out_alloc:
1451 	if (alloc_required)
1452 		gfs2_alloc_put(ndip);
1453 out_gunlock:
1454 	while (x--) {
1455 		gfs2_glock_dq(ghs + x);
1456 		gfs2_holder_uninit(ghs + x);
1457 	}
1458 out_gunlock_r:
1459 	if (r_gh.gh_gl)
1460 		gfs2_glock_dq_uninit(&r_gh);
1461 out:
1462 	gfs2_glock_dq_uninit(&ri_gh);
1463 	return error;
1464 }
1465 
1466 /**
1467  * gfs2_follow_link - Follow a symbolic link
1468  * @dentry: The dentry of the link
1469  * @nd: Data that we pass to vfs_follow_link()
1470  *
1471  * This can handle symlinks of any size.
1472  *
1473  * Returns: 0 on success or error code
1474  */
1475 
1476 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1477 {
1478 	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1479 	struct gfs2_holder i_gh;
1480 	struct buffer_head *dibh;
1481 	unsigned int size;
1482 	char *buf;
1483 	int error;
1484 
1485 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1486 	error = gfs2_glock_nq(&i_gh);
1487 	if (error) {
1488 		gfs2_holder_uninit(&i_gh);
1489 		nd_set_link(nd, ERR_PTR(error));
1490 		return NULL;
1491 	}
1492 
1493 	size = (unsigned int)i_size_read(&ip->i_inode);
1494 	if (size == 0) {
1495 		gfs2_consist_inode(ip);
1496 		buf = ERR_PTR(-EIO);
1497 		goto out;
1498 	}
1499 
1500 	error = gfs2_meta_inode_buffer(ip, &dibh);
1501 	if (error) {
1502 		buf = ERR_PTR(error);
1503 		goto out;
1504 	}
1505 
1506 	buf = kzalloc(size + 1, GFP_NOFS);
1507 	if (!buf)
1508 		buf = ERR_PTR(-ENOMEM);
1509 	else
1510 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1511 	brelse(dibh);
1512 out:
1513 	gfs2_glock_dq_uninit(&i_gh);
1514 	nd_set_link(nd, buf);
1515 	return NULL;
1516 }
1517 
1518 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1519 {
1520 	char *s = nd_get_link(nd);
1521 	if (!IS_ERR(s))
1522 		kfree(s);
1523 }
1524 
1525 /**
1526  * gfs2_permission -
1527  * @inode: The inode
1528  * @mask: The mask to be tested
1529  * @flags: Indicates whether this is an RCU path walk or not
1530  *
1531  * This may be called from the VFS directly, or from within GFS2 with the
1532  * inode locked, so we look to see if the glock is already locked and only
1533  * lock the glock if its not already been done.
1534  *
1535  * Returns: errno
1536  */
1537 
1538 int gfs2_permission(struct inode *inode, int mask)
1539 {
1540 	struct gfs2_inode *ip;
1541 	struct gfs2_holder i_gh;
1542 	int error;
1543 	int unlock = 0;
1544 
1545 
1546 	ip = GFS2_I(inode);
1547 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1548 		if (mask & MAY_NOT_BLOCK)
1549 			return -ECHILD;
1550 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1551 		if (error)
1552 			return error;
1553 		unlock = 1;
1554 	}
1555 
1556 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1557 		error = -EACCES;
1558 	else
1559 		error = generic_permission(inode, mask);
1560 	if (unlock)
1561 		gfs2_glock_dq_uninit(&i_gh);
1562 
1563 	return error;
1564 }
1565 
1566 static int __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1567 {
1568 	struct inode *inode = &ip->i_inode;
1569 	struct buffer_head *dibh;
1570 	int error;
1571 
1572 	error = gfs2_meta_inode_buffer(ip, &dibh);
1573 	if (error)
1574 		return error;
1575 
1576 	setattr_copy(inode, attr);
1577 	mark_inode_dirty(inode);
1578 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1579 	gfs2_dinode_out(ip, dibh->b_data);
1580 	brelse(dibh);
1581 	return 0;
1582 }
1583 
1584 /**
1585  * gfs2_setattr_simple -
1586  * @ip:
1587  * @attr:
1588  *
1589  * Returns: errno
1590  */
1591 
1592 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1593 {
1594 	int error;
1595 
1596 	if (current->journal_info)
1597 		return __gfs2_setattr_simple(ip, attr);
1598 
1599 	error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0);
1600 	if (error)
1601 		return error;
1602 
1603 	error = __gfs2_setattr_simple(ip, attr);
1604 	gfs2_trans_end(GFS2_SB(&ip->i_inode));
1605 	return error;
1606 }
1607 
1608 static int setattr_chown(struct inode *inode, struct iattr *attr)
1609 {
1610 	struct gfs2_inode *ip = GFS2_I(inode);
1611 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1612 	u32 ouid, ogid, nuid, ngid;
1613 	int error;
1614 
1615 	ouid = inode->i_uid;
1616 	ogid = inode->i_gid;
1617 	nuid = attr->ia_uid;
1618 	ngid = attr->ia_gid;
1619 
1620 	if (!(attr->ia_valid & ATTR_UID) || ouid == nuid)
1621 		ouid = nuid = NO_QUOTA_CHANGE;
1622 	if (!(attr->ia_valid & ATTR_GID) || ogid == ngid)
1623 		ogid = ngid = NO_QUOTA_CHANGE;
1624 
1625 	if (!gfs2_alloc_get(ip))
1626 		return -ENOMEM;
1627 
1628 	error = gfs2_quota_lock(ip, nuid, ngid);
1629 	if (error)
1630 		goto out_alloc;
1631 
1632 	if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1633 		error = gfs2_quota_check(ip, nuid, ngid);
1634 		if (error)
1635 			goto out_gunlock_q;
1636 	}
1637 
1638 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1639 	if (error)
1640 		goto out_gunlock_q;
1641 
1642 	error = gfs2_setattr_simple(ip, attr);
1643 	if (error)
1644 		goto out_end_trans;
1645 
1646 	if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1647 		u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1648 		gfs2_quota_change(ip, -blocks, ouid, ogid);
1649 		gfs2_quota_change(ip, blocks, nuid, ngid);
1650 	}
1651 
1652 out_end_trans:
1653 	gfs2_trans_end(sdp);
1654 out_gunlock_q:
1655 	gfs2_quota_unlock(ip);
1656 out_alloc:
1657 	gfs2_alloc_put(ip);
1658 	return error;
1659 }
1660 
1661 /**
1662  * gfs2_setattr - Change attributes on an inode
1663  * @dentry: The dentry which is changing
1664  * @attr: The structure describing the change
1665  *
1666  * The VFS layer wants to change one or more of an inodes attributes.  Write
1667  * that change out to disk.
1668  *
1669  * Returns: errno
1670  */
1671 
1672 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1673 {
1674 	struct inode *inode = dentry->d_inode;
1675 	struct gfs2_inode *ip = GFS2_I(inode);
1676 	struct gfs2_holder i_gh;
1677 	int error;
1678 
1679 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1680 	if (error)
1681 		return error;
1682 
1683 	error = -EPERM;
1684 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1685 		goto out;
1686 
1687 	error = inode_change_ok(inode, attr);
1688 	if (error)
1689 		goto out;
1690 
1691 	if (attr->ia_valid & ATTR_SIZE)
1692 		error = gfs2_setattr_size(inode, attr->ia_size);
1693 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1694 		error = setattr_chown(inode, attr);
1695 	else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1696 		error = gfs2_acl_chmod(ip, attr);
1697 	else
1698 		error = gfs2_setattr_simple(ip, attr);
1699 
1700 out:
1701 	gfs2_glock_dq_uninit(&i_gh);
1702 	if (!error)
1703 		mark_inode_dirty(inode);
1704 	return error;
1705 }
1706 
1707 /**
1708  * gfs2_getattr - Read out an inode's attributes
1709  * @mnt: The vfsmount the inode is being accessed from
1710  * @dentry: The dentry to stat
1711  * @stat: The inode's stats
1712  *
1713  * This may be called from the VFS directly, or from within GFS2 with the
1714  * inode locked, so we look to see if the glock is already locked and only
1715  * lock the glock if its not already been done. Note that its the NFS
1716  * readdirplus operation which causes this to be called (from filldir)
1717  * with the glock already held.
1718  *
1719  * Returns: errno
1720  */
1721 
1722 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1723 			struct kstat *stat)
1724 {
1725 	struct inode *inode = dentry->d_inode;
1726 	struct gfs2_inode *ip = GFS2_I(inode);
1727 	struct gfs2_holder gh;
1728 	int error;
1729 	int unlock = 0;
1730 
1731 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1732 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1733 		if (error)
1734 			return error;
1735 		unlock = 1;
1736 	}
1737 
1738 	generic_fillattr(inode, stat);
1739 	if (unlock)
1740 		gfs2_glock_dq_uninit(&gh);
1741 
1742 	return 0;
1743 }
1744 
1745 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1746 			 const void *data, size_t size, int flags)
1747 {
1748 	struct inode *inode = dentry->d_inode;
1749 	struct gfs2_inode *ip = GFS2_I(inode);
1750 	struct gfs2_holder gh;
1751 	int ret;
1752 
1753 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1754 	ret = gfs2_glock_nq(&gh);
1755 	if (ret == 0) {
1756 		ret = generic_setxattr(dentry, name, data, size, flags);
1757 		gfs2_glock_dq(&gh);
1758 	}
1759 	gfs2_holder_uninit(&gh);
1760 	return ret;
1761 }
1762 
1763 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1764 			     void *data, size_t size)
1765 {
1766 	struct inode *inode = dentry->d_inode;
1767 	struct gfs2_inode *ip = GFS2_I(inode);
1768 	struct gfs2_holder gh;
1769 	int ret;
1770 
1771 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1772 	ret = gfs2_glock_nq(&gh);
1773 	if (ret == 0) {
1774 		ret = generic_getxattr(dentry, name, data, size);
1775 		gfs2_glock_dq(&gh);
1776 	}
1777 	gfs2_holder_uninit(&gh);
1778 	return ret;
1779 }
1780 
1781 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1782 {
1783 	struct inode *inode = dentry->d_inode;
1784 	struct gfs2_inode *ip = GFS2_I(inode);
1785 	struct gfs2_holder gh;
1786 	int ret;
1787 
1788 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1789 	ret = gfs2_glock_nq(&gh);
1790 	if (ret == 0) {
1791 		ret = generic_removexattr(dentry, name);
1792 		gfs2_glock_dq(&gh);
1793 	}
1794 	gfs2_holder_uninit(&gh);
1795 	return ret;
1796 }
1797 
1798 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1799 		       u64 start, u64 len)
1800 {
1801 	struct gfs2_inode *ip = GFS2_I(inode);
1802 	struct gfs2_holder gh;
1803 	int ret;
1804 
1805 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1806 	if (ret)
1807 		return ret;
1808 
1809 	mutex_lock(&inode->i_mutex);
1810 
1811 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1812 	if (ret)
1813 		goto out;
1814 
1815 	if (gfs2_is_stuffed(ip)) {
1816 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1817 		u64 size = i_size_read(inode);
1818 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1819 			    FIEMAP_EXTENT_DATA_INLINE;
1820 		phys += sizeof(struct gfs2_dinode);
1821 		phys += start;
1822 		if (start + len > size)
1823 			len = size - start;
1824 		if (start < size)
1825 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1826 						      len, flags);
1827 		if (ret == 1)
1828 			ret = 0;
1829 	} else {
1830 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1831 					     gfs2_block_map);
1832 	}
1833 
1834 	gfs2_glock_dq_uninit(&gh);
1835 out:
1836 	mutex_unlock(&inode->i_mutex);
1837 	return ret;
1838 }
1839 
1840 const struct inode_operations gfs2_file_iops = {
1841 	.permission = gfs2_permission,
1842 	.setattr = gfs2_setattr,
1843 	.getattr = gfs2_getattr,
1844 	.setxattr = gfs2_setxattr,
1845 	.getxattr = gfs2_getxattr,
1846 	.listxattr = gfs2_listxattr,
1847 	.removexattr = gfs2_removexattr,
1848 	.fiemap = gfs2_fiemap,
1849 	.get_acl = gfs2_get_acl,
1850 };
1851 
1852 const struct inode_operations gfs2_dir_iops = {
1853 	.create = gfs2_create,
1854 	.lookup = gfs2_lookup,
1855 	.link = gfs2_link,
1856 	.unlink = gfs2_unlink,
1857 	.symlink = gfs2_symlink,
1858 	.mkdir = gfs2_mkdir,
1859 	.rmdir = gfs2_unlink,
1860 	.mknod = gfs2_mknod,
1861 	.rename = gfs2_rename,
1862 	.permission = gfs2_permission,
1863 	.setattr = gfs2_setattr,
1864 	.getattr = gfs2_getattr,
1865 	.setxattr = gfs2_setxattr,
1866 	.getxattr = gfs2_getxattr,
1867 	.listxattr = gfs2_listxattr,
1868 	.removexattr = gfs2_removexattr,
1869 	.fiemap = gfs2_fiemap,
1870 	.get_acl = gfs2_get_acl,
1871 };
1872 
1873 const struct inode_operations gfs2_symlink_iops = {
1874 	.readlink = generic_readlink,
1875 	.follow_link = gfs2_follow_link,
1876 	.put_link = gfs2_put_link,
1877 	.permission = gfs2_permission,
1878 	.setattr = gfs2_setattr,
1879 	.getattr = gfs2_getattr,
1880 	.setxattr = gfs2_setxattr,
1881 	.getxattr = gfs2_getxattr,
1882 	.listxattr = gfs2_listxattr,
1883 	.removexattr = gfs2_removexattr,
1884 	.fiemap = gfs2_fiemap,
1885 	.get_acl = gfs2_get_acl,
1886 };
1887 
1888