xref: /linux/fs/xfs/xfs_handle.c (revision 7fc2cd2e4b398c57c9cf961cfea05eadbf34c05c)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2022-2024 Oracle.
5  * All rights reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_shared.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_bmap_btree.h"
15 #include "xfs_inode.h"
16 #include "xfs_error.h"
17 #include "xfs_trace.h"
18 #include "xfs_trans.h"
19 #include "xfs_da_format.h"
20 #include "xfs_da_btree.h"
21 #include "xfs_attr.h"
22 #include "xfs_ioctl.h"
23 #include "xfs_parent.h"
24 #include "xfs_handle.h"
25 #include "xfs_health.h"
26 #include "xfs_icache.h"
27 #include "xfs_export.h"
28 #include "xfs_xattr.h"
29 #include "xfs_acl.h"
30 
31 #include <linux/namei.h>
32 
33 static inline size_t
34 xfs_filehandle_fid_len(void)
35 {
36 	struct xfs_handle	*handle = NULL;
37 
38 	return sizeof(struct xfs_fid) - sizeof(handle->ha_fid.fid_len);
39 }
40 
41 static inline size_t
42 xfs_filehandle_init(
43 	struct xfs_mount	*mp,
44 	xfs_ino_t		ino,
45 	uint32_t		gen,
46 	struct xfs_handle	*handle)
47 {
48 	memcpy(&handle->ha_fsid, mp->m_fixedfsid, sizeof(struct xfs_fsid));
49 
50 	handle->ha_fid.fid_len = xfs_filehandle_fid_len();
51 	handle->ha_fid.fid_pad = 0;
52 	handle->ha_fid.fid_gen = gen;
53 	handle->ha_fid.fid_ino = ino;
54 
55 	return sizeof(struct xfs_handle);
56 }
57 
58 static inline size_t
59 xfs_fshandle_init(
60 	struct xfs_mount	*mp,
61 	struct xfs_handle	*handle)
62 {
63 	memcpy(&handle->ha_fsid, mp->m_fixedfsid, sizeof(struct xfs_fsid));
64 	memset(&handle->ha_fid, 0, sizeof(handle->ha_fid));
65 
66 	return sizeof(struct xfs_fsid);
67 }
68 
69 /*
70  * xfs_find_handle maps from userspace xfs_fsop_handlereq structure to
71  * a file or fs handle.
72  *
73  * XFS_IOC_PATH_TO_FSHANDLE
74  *    returns fs handle for a mount point or path within that mount point
75  * XFS_IOC_FD_TO_HANDLE
76  *    returns full handle for a FD opened in user space
77  * XFS_IOC_PATH_TO_HANDLE
78  *    returns full handle for a path
79  */
80 int
81 xfs_find_handle(
82 	unsigned int		cmd,
83 	xfs_fsop_handlereq_t	*hreq)
84 {
85 	int			hsize;
86 	xfs_handle_t		handle;
87 	struct inode		*inode;
88 	struct path		path;
89 	int			error;
90 	struct xfs_inode	*ip;
91 
92 	if (cmd == XFS_IOC_FD_TO_HANDLE) {
93 		CLASS(fd, f)(hreq->fd);
94 
95 		if (fd_empty(f))
96 			return -EBADF;
97 		path = fd_file(f)->f_path;
98 		path_get(&path);
99 	} else {
100 		error = user_path_at(AT_FDCWD, hreq->path, 0, &path);
101 		if (error)
102 			return error;
103 	}
104 	inode = d_inode(path.dentry);
105 	ip = XFS_I(inode);
106 
107 	/*
108 	 * We can only generate handles for inodes residing on a XFS filesystem,
109 	 * and only for regular files, directories or symbolic links.
110 	 */
111 	error = -EINVAL;
112 	if (inode->i_sb->s_magic != XFS_SB_MAGIC)
113 		goto out_put;
114 
115 	error = -EBADF;
116 	if (!S_ISREG(inode->i_mode) &&
117 	    !S_ISDIR(inode->i_mode) &&
118 	    !S_ISLNK(inode->i_mode))
119 		goto out_put;
120 
121 
122 	memcpy(&handle.ha_fsid, ip->i_mount->m_fixedfsid, sizeof(xfs_fsid_t));
123 
124 	if (cmd == XFS_IOC_PATH_TO_FSHANDLE)
125 		hsize = xfs_fshandle_init(ip->i_mount, &handle);
126 	else
127 		hsize = xfs_filehandle_init(ip->i_mount, ip->i_ino,
128 				inode->i_generation, &handle);
129 
130 	error = -EFAULT;
131 	if (copy_to_user(hreq->ohandle, &handle, hsize) ||
132 	    copy_to_user(hreq->ohandlen, &hsize, sizeof(__s32)))
133 		goto out_put;
134 
135 	error = 0;
136 
137  out_put:
138 	path_put(&path);
139 	return error;
140 }
141 
142 /*
143  * No need to do permission checks on the various pathname components
144  * as the handle operations are privileged.
145  */
146 STATIC int
147 xfs_handle_acceptable(
148 	void			*context,
149 	struct dentry		*dentry)
150 {
151 	return 1;
152 }
153 
154 /* Convert handle already copied to kernel space into a dentry. */
155 static struct dentry *
156 xfs_khandle_to_dentry(
157 	struct file		*file,
158 	struct xfs_handle	*handle)
159 {
160 	struct xfs_fid64        fid = {
161 		.ino		= handle->ha_fid.fid_ino,
162 		.gen		= handle->ha_fid.fid_gen,
163 	};
164 
165 	/*
166 	 * Only allow handle opens under a directory.
167 	 */
168 	if (!S_ISDIR(file_inode(file)->i_mode))
169 		return ERR_PTR(-ENOTDIR);
170 
171 	if (handle->ha_fid.fid_len != xfs_filehandle_fid_len())
172 		return ERR_PTR(-EINVAL);
173 
174 	return exportfs_decode_fh(file->f_path.mnt, (struct fid *)&fid, 3,
175 			FILEID_INO32_GEN | XFS_FILEID_TYPE_64FLAG,
176 			xfs_handle_acceptable, NULL);
177 }
178 
179 /* Convert handle already copied to kernel space into an xfs_inode. */
180 static struct xfs_inode *
181 xfs_khandle_to_inode(
182 	struct file		*file,
183 	struct xfs_handle	*handle)
184 {
185 	struct xfs_inode	*ip = XFS_I(file_inode(file));
186 	struct xfs_mount	*mp = ip->i_mount;
187 	struct inode		*inode;
188 
189 	if (!S_ISDIR(VFS_I(ip)->i_mode))
190 		return ERR_PTR(-ENOTDIR);
191 
192 	if (handle->ha_fid.fid_len != xfs_filehandle_fid_len())
193 		return ERR_PTR(-EINVAL);
194 
195 	inode = xfs_nfs_get_inode(mp->m_super, handle->ha_fid.fid_ino,
196 			handle->ha_fid.fid_gen);
197 	if (IS_ERR(inode))
198 		return ERR_CAST(inode);
199 
200 	return XFS_I(inode);
201 }
202 
203 /*
204  * Convert userspace handle data into a dentry.
205  */
206 struct dentry *
207 xfs_handle_to_dentry(
208 	struct file		*parfilp,
209 	void __user		*uhandle,
210 	u32			hlen)
211 {
212 	xfs_handle_t		handle;
213 
214 	if (hlen != sizeof(xfs_handle_t))
215 		return ERR_PTR(-EINVAL);
216 	if (copy_from_user(&handle, uhandle, hlen))
217 		return ERR_PTR(-EFAULT);
218 
219 	return xfs_khandle_to_dentry(parfilp, &handle);
220 }
221 
222 STATIC struct dentry *
223 xfs_handlereq_to_dentry(
224 	struct file		*parfilp,
225 	xfs_fsop_handlereq_t	*hreq)
226 {
227 	return xfs_handle_to_dentry(parfilp, hreq->ihandle, hreq->ihandlen);
228 }
229 
230 int
231 xfs_open_by_handle(
232 	struct file		*parfilp,
233 	xfs_fsop_handlereq_t	*hreq)
234 {
235 	const struct cred	*cred = current_cred();
236 	int			permflag;
237 	struct inode		*inode;
238 	struct dentry		*dentry;
239 	fmode_t			fmode;
240 	struct path		path __free(path_put) = {};
241 
242 	if (!capable(CAP_SYS_ADMIN))
243 		return -EPERM;
244 
245 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
246 	if (IS_ERR(dentry))
247 		return PTR_ERR(dentry);
248 	inode = d_inode(dentry);
249 	path.dentry = dentry;
250 
251 	/* Restrict xfs_open_by_handle to directories & regular files. */
252 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
253 		return -EPERM;
254 
255 #if BITS_PER_LONG != 32
256 	hreq->oflags |= O_LARGEFILE;
257 #endif
258 
259 	permflag = hreq->oflags;
260 	fmode = OPEN_FMODE(permflag);
261 	if ((!(permflag & O_APPEND) || (permflag & O_TRUNC)) &&
262 	    (fmode & FMODE_WRITE) && IS_APPEND(inode))
263 		return -EPERM;
264 
265 	if ((fmode & FMODE_WRITE) && IS_IMMUTABLE(inode))
266 		return -EPERM;
267 
268 	/* Can't write directories. */
269 	if (S_ISDIR(inode->i_mode) && (fmode & FMODE_WRITE))
270 		return -EISDIR;
271 
272 	path.mnt = mntget(parfilp->f_path.mnt);
273 
274 	FD_PREPARE(fdf, 0, dentry_open(&path, hreq->oflags, cred));
275 	if (fdf.err)
276 		return fdf.err;
277 
278 	if (S_ISREG(inode->i_mode)) {
279 		struct file *filp = fd_prepare_file(fdf);
280 
281 		filp->f_flags |= O_NOATIME;
282 		filp->f_mode |= FMODE_NOCMTIME;
283 	}
284 
285 	return fd_publish(fdf);
286 }
287 
288 int
289 xfs_readlink_by_handle(
290 	struct file		*parfilp,
291 	xfs_fsop_handlereq_t	*hreq)
292 {
293 	struct dentry		*dentry;
294 	__u32			olen;
295 	int			error;
296 
297 	if (!capable(CAP_SYS_ADMIN))
298 		return -EPERM;
299 
300 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
301 	if (IS_ERR(dentry))
302 		return PTR_ERR(dentry);
303 
304 	/* Restrict this handle operation to symlinks only. */
305 	if (!d_is_symlink(dentry)) {
306 		error = -EINVAL;
307 		goto out_dput;
308 	}
309 
310 	if (copy_from_user(&olen, hreq->ohandlen, sizeof(__u32))) {
311 		error = -EFAULT;
312 		goto out_dput;
313 	}
314 
315 	error = vfs_readlink(dentry, hreq->ohandle, olen);
316 
317  out_dput:
318 	dput(dentry);
319 	return error;
320 }
321 
322 /*
323  * Format an attribute and copy it out to the user's buffer.
324  * Take care to check values and protect against them changing later,
325  * we may be reading them directly out of a user buffer.
326  */
327 static void
328 xfs_ioc_attr_put_listent(
329 	struct xfs_attr_list_context *context,
330 	int			flags,
331 	unsigned char		*name,
332 	int			namelen,
333 	void			*value,
334 	int			valuelen)
335 {
336 	struct xfs_attrlist	*alist = context->buffer;
337 	struct xfs_attrlist_ent	*aep;
338 	int			arraytop;
339 
340 	ASSERT(!context->seen_enough);
341 	ASSERT(context->count >= 0);
342 	ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
343 	ASSERT(context->firstu >= sizeof(*alist));
344 	ASSERT(context->firstu <= context->bufsize);
345 
346 	/*
347 	 * Only list entries in the right namespace.
348 	 */
349 	if (context->attr_filter != (flags & XFS_ATTR_NSP_ONDISK_MASK))
350 		return;
351 
352 	arraytop = sizeof(*alist) +
353 			context->count * sizeof(alist->al_offset[0]);
354 
355 	/* decrement by the actual bytes used by the attr */
356 	context->firstu -= round_up(offsetof(struct xfs_attrlist_ent, a_name) +
357 			namelen + 1, sizeof(uint32_t));
358 	if (context->firstu < arraytop) {
359 		trace_xfs_attr_list_full(context);
360 		alist->al_more = 1;
361 		context->seen_enough = 1;
362 		return;
363 	}
364 
365 	aep = context->buffer + context->firstu;
366 	aep->a_valuelen = valuelen;
367 	memcpy(aep->a_name, name, namelen);
368 	aep->a_name[namelen] = 0;
369 	alist->al_offset[context->count++] = context->firstu;
370 	alist->al_count = context->count;
371 	trace_xfs_attr_list_add(context);
372 }
373 
374 static unsigned int
375 xfs_attr_filter(
376 	u32			ioc_flags)
377 {
378 	if (ioc_flags & XFS_IOC_ATTR_ROOT)
379 		return XFS_ATTR_ROOT;
380 	if (ioc_flags & XFS_IOC_ATTR_SECURE)
381 		return XFS_ATTR_SECURE;
382 	return 0;
383 }
384 
385 static inline enum xfs_attr_update
386 xfs_xattr_flags(
387 	u32			ioc_flags,
388 	void			*value)
389 {
390 	if (!value)
391 		return XFS_ATTRUPDATE_REMOVE;
392 	if (ioc_flags & XFS_IOC_ATTR_CREATE)
393 		return XFS_ATTRUPDATE_CREATE;
394 	if (ioc_flags & XFS_IOC_ATTR_REPLACE)
395 		return XFS_ATTRUPDATE_REPLACE;
396 	return XFS_ATTRUPDATE_UPSERT;
397 }
398 
399 int
400 xfs_ioc_attr_list(
401 	struct xfs_inode		*dp,
402 	void __user			*ubuf,
403 	size_t				bufsize,
404 	int				flags,
405 	struct xfs_attrlist_cursor __user *ucursor)
406 {
407 	struct xfs_attr_list_context	context = { };
408 	struct xfs_attrlist		*alist;
409 	void				*buffer;
410 	int				error;
411 
412 	if (bufsize < sizeof(struct xfs_attrlist) ||
413 	    bufsize > XFS_XATTR_LIST_MAX)
414 		return -EINVAL;
415 
416 	/*
417 	 * Reject flags, only allow namespaces.
418 	 */
419 	if (flags & ~(XFS_IOC_ATTR_ROOT | XFS_IOC_ATTR_SECURE))
420 		return -EINVAL;
421 	if (flags == (XFS_IOC_ATTR_ROOT | XFS_IOC_ATTR_SECURE))
422 		return -EINVAL;
423 
424 	/*
425 	 * Validate the cursor.
426 	 */
427 	if (copy_from_user(&context.cursor, ucursor, sizeof(context.cursor)))
428 		return -EFAULT;
429 	if (context.cursor.pad1 || context.cursor.pad2)
430 		return -EINVAL;
431 	if (!context.cursor.initted &&
432 	    (context.cursor.hashval || context.cursor.blkno ||
433 	     context.cursor.offset))
434 		return -EINVAL;
435 
436 	buffer = kvzalloc(bufsize, GFP_KERNEL);
437 	if (!buffer)
438 		return -ENOMEM;
439 
440 	/*
441 	 * Initialize the output buffer.
442 	 */
443 	context.dp = dp;
444 	context.resynch = 1;
445 	context.attr_filter = xfs_attr_filter(flags);
446 	context.buffer = buffer;
447 	context.bufsize = round_down(bufsize, sizeof(uint32_t));
448 	context.firstu = context.bufsize;
449 	context.put_listent = xfs_ioc_attr_put_listent;
450 
451 	alist = context.buffer;
452 	alist->al_count = 0;
453 	alist->al_more = 0;
454 	alist->al_offset[0] = context.bufsize;
455 
456 	error = xfs_attr_list(&context);
457 	if (error)
458 		goto out_free;
459 
460 	if (copy_to_user(ubuf, buffer, bufsize) ||
461 	    copy_to_user(ucursor, &context.cursor, sizeof(context.cursor)))
462 		error = -EFAULT;
463 out_free:
464 	kvfree(buffer);
465 	return error;
466 }
467 
468 int
469 xfs_attrlist_by_handle(
470 	struct file		*parfilp,
471 	struct xfs_fsop_attrlist_handlereq __user *p)
472 {
473 	struct xfs_fsop_attrlist_handlereq al_hreq;
474 	struct dentry		*dentry;
475 	int			error = -ENOMEM;
476 
477 	if (!capable(CAP_SYS_ADMIN))
478 		return -EPERM;
479 	if (copy_from_user(&al_hreq, p, sizeof(al_hreq)))
480 		return -EFAULT;
481 
482 	dentry = xfs_handlereq_to_dentry(parfilp, &al_hreq.hreq);
483 	if (IS_ERR(dentry))
484 		return PTR_ERR(dentry);
485 
486 	error = xfs_ioc_attr_list(XFS_I(d_inode(dentry)), al_hreq.buffer,
487 				  al_hreq.buflen, al_hreq.flags, &p->pos);
488 	dput(dentry);
489 	return error;
490 }
491 
492 static int
493 xfs_attrmulti_attr_get(
494 	struct inode		*inode,
495 	unsigned char		*name,
496 	unsigned char		__user *ubuf,
497 	uint32_t		*len,
498 	uint32_t		flags)
499 {
500 	struct xfs_da_args	args = {
501 		.dp		= XFS_I(inode),
502 		.attr_filter	= xfs_attr_filter(flags),
503 		.name		= name,
504 		.namelen	= strlen(name),
505 		.valuelen	= *len,
506 	};
507 	int			error;
508 
509 	if (*len > XFS_XATTR_SIZE_MAX)
510 		return -EINVAL;
511 
512 	error = xfs_attr_get(&args);
513 	if (error)
514 		goto out_kfree;
515 
516 	*len = args.valuelen;
517 	if (copy_to_user(ubuf, args.value, args.valuelen))
518 		error = -EFAULT;
519 
520 out_kfree:
521 	kvfree(args.value);
522 	return error;
523 }
524 
525 static int
526 xfs_attrmulti_attr_set(
527 	struct inode		*inode,
528 	unsigned char		*name,
529 	const unsigned char	__user *ubuf,
530 	uint32_t		len,
531 	uint32_t		flags)
532 {
533 	struct xfs_da_args	args = {
534 		.dp		= XFS_I(inode),
535 		.attr_filter	= xfs_attr_filter(flags),
536 		.name		= name,
537 		.namelen	= strlen(name),
538 	};
539 	int			error;
540 
541 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
542 		return -EPERM;
543 
544 	if (ubuf) {
545 		if (len > XFS_XATTR_SIZE_MAX)
546 			return -EINVAL;
547 		args.value = memdup_user(ubuf, len);
548 		if (IS_ERR(args.value))
549 			return PTR_ERR(args.value);
550 		args.valuelen = len;
551 	}
552 
553 	error = xfs_attr_change(&args, xfs_xattr_flags(flags, args.value));
554 	if (!error && (flags & XFS_IOC_ATTR_ROOT))
555 		xfs_forget_acl(inode, name);
556 	kfree(args.value);
557 	return error;
558 }
559 
560 int
561 xfs_ioc_attrmulti_one(
562 	struct file		*parfilp,
563 	struct inode		*inode,
564 	uint32_t		opcode,
565 	void __user		*uname,
566 	void __user		*value,
567 	uint32_t		*len,
568 	uint32_t		flags)
569 {
570 	unsigned char		*name;
571 	int			error;
572 
573 	if ((flags & XFS_IOC_ATTR_ROOT) && (flags & XFS_IOC_ATTR_SECURE))
574 		return -EINVAL;
575 
576 	name = strndup_user(uname, MAXNAMELEN);
577 	if (IS_ERR(name))
578 		return PTR_ERR(name);
579 
580 	switch (opcode) {
581 	case ATTR_OP_GET:
582 		error = xfs_attrmulti_attr_get(inode, name, value, len, flags);
583 		break;
584 	case ATTR_OP_REMOVE:
585 		value = NULL;
586 		*len = 0;
587 		fallthrough;
588 	case ATTR_OP_SET:
589 		error = mnt_want_write_file(parfilp);
590 		if (error)
591 			break;
592 		error = xfs_attrmulti_attr_set(inode, name, value, *len, flags);
593 		mnt_drop_write_file(parfilp);
594 		break;
595 	default:
596 		error = -EINVAL;
597 		break;
598 	}
599 
600 	kfree(name);
601 	return error;
602 }
603 
604 int
605 xfs_attrmulti_by_handle(
606 	struct file		*parfilp,
607 	void			__user *arg)
608 {
609 	int			error;
610 	xfs_attr_multiop_t	*ops;
611 	xfs_fsop_attrmulti_handlereq_t am_hreq;
612 	struct dentry		*dentry;
613 	unsigned int		i, size;
614 
615 	if (!capable(CAP_SYS_ADMIN))
616 		return -EPERM;
617 	if (copy_from_user(&am_hreq, arg, sizeof(xfs_fsop_attrmulti_handlereq_t)))
618 		return -EFAULT;
619 
620 	/* overflow check */
621 	if (am_hreq.opcount >= INT_MAX / sizeof(xfs_attr_multiop_t))
622 		return -E2BIG;
623 
624 	dentry = xfs_handlereq_to_dentry(parfilp, &am_hreq.hreq);
625 	if (IS_ERR(dentry))
626 		return PTR_ERR(dentry);
627 
628 	error = -E2BIG;
629 	size = am_hreq.opcount * sizeof(xfs_attr_multiop_t);
630 	if (!size || size > 16 * PAGE_SIZE)
631 		goto out_dput;
632 
633 	ops = memdup_user(am_hreq.ops, size);
634 	if (IS_ERR(ops)) {
635 		error = PTR_ERR(ops);
636 		goto out_dput;
637 	}
638 
639 	error = 0;
640 	for (i = 0; i < am_hreq.opcount; i++) {
641 		ops[i].am_error = xfs_ioc_attrmulti_one(parfilp,
642 				d_inode(dentry), ops[i].am_opcode,
643 				ops[i].am_attrname, ops[i].am_attrvalue,
644 				&ops[i].am_length, ops[i].am_flags);
645 	}
646 
647 	if (copy_to_user(am_hreq.ops, ops, size))
648 		error = -EFAULT;
649 
650 	kfree(ops);
651  out_dput:
652 	dput(dentry);
653 	return error;
654 }
655 
656 struct xfs_getparents_ctx {
657 	struct xfs_attr_list_context	context;
658 	struct xfs_getparents_by_handle	gph;
659 
660 	/* File to target */
661 	struct xfs_inode		*ip;
662 
663 	/* Internal buffer where we format records */
664 	void				*krecords;
665 
666 	/* Last record filled out */
667 	struct xfs_getparents_rec	*lastrec;
668 
669 	unsigned int			count;
670 };
671 
672 static inline unsigned int
673 xfs_getparents_rec_sizeof(
674 	unsigned int		namelen)
675 {
676 	return round_up(sizeof(struct xfs_getparents_rec) + namelen + 1,
677 			sizeof(uint64_t));
678 }
679 
680 static void
681 xfs_getparents_put_listent(
682 	struct xfs_attr_list_context	*context,
683 	int				flags,
684 	unsigned char			*name,
685 	int				namelen,
686 	void				*value,
687 	int				valuelen)
688 {
689 	struct xfs_getparents_ctx	*gpx =
690 		container_of(context, struct xfs_getparents_ctx, context);
691 	struct xfs_inode		*ip = context->dp;
692 	struct xfs_mount		*mp = ip->i_mount;
693 	struct xfs_getparents		*gp = &gpx->gph.gph_request;
694 	struct xfs_getparents_rec	*gpr = gpx->krecords + context->firstu;
695 	unsigned short			reclen =
696 		xfs_getparents_rec_sizeof(namelen);
697 	xfs_ino_t			ino;
698 	uint32_t			gen;
699 	int				error;
700 
701 	if (!(flags & XFS_ATTR_PARENT))
702 		return;
703 
704 	error = xfs_parent_from_attr(mp, flags, name, namelen, value, valuelen,
705 			&ino, &gen);
706 	if (error) {
707 		xfs_inode_mark_sick(ip, XFS_SICK_INO_PARENT);
708 		context->seen_enough = -EFSCORRUPTED;
709 		return;
710 	}
711 
712 	/*
713 	 * We found a parent pointer, but we've filled up the buffer.  Signal
714 	 * to the caller that we did /not/ reach the end of the parent pointer
715 	 * recordset.
716 	 */
717 	if (context->firstu > context->bufsize - reclen) {
718 		context->seen_enough = 1;
719 		return;
720 	}
721 
722 	/* Format the parent pointer directly into the caller buffer. */
723 	gpr->gpr_reclen = reclen;
724 	xfs_filehandle_init(mp, ino, gen, &gpr->gpr_parent);
725 	memcpy(gpr->gpr_name, name, namelen);
726 	gpr->gpr_name[namelen] = 0;
727 
728 	trace_xfs_getparents_put_listent(ip, gp, context, gpr);
729 
730 	context->firstu += reclen;
731 	gpx->count++;
732 	gpx->lastrec = gpr;
733 }
734 
735 /* Expand the last record to fill the rest of the caller's buffer. */
736 static inline void
737 xfs_getparents_expand_lastrec(
738 	struct xfs_getparents_ctx	*gpx)
739 {
740 	struct xfs_getparents		*gp = &gpx->gph.gph_request;
741 	struct xfs_getparents_rec	*gpr = gpx->lastrec;
742 
743 	if (!gpx->lastrec)
744 		gpr = gpx->krecords;
745 
746 	gpr->gpr_reclen = gp->gp_bufsize - ((void *)gpr - gpx->krecords);
747 
748 	trace_xfs_getparents_expand_lastrec(gpx->ip, gp, &gpx->context, gpr);
749 }
750 
751 /* Retrieve the parent pointers for a given inode. */
752 STATIC int
753 xfs_getparents(
754 	struct xfs_getparents_ctx	*gpx)
755 {
756 	struct xfs_getparents		*gp = &gpx->gph.gph_request;
757 	struct xfs_inode		*ip = gpx->ip;
758 	struct xfs_mount		*mp = ip->i_mount;
759 	size_t				bufsize;
760 	int				error;
761 
762 	/* Check size of buffer requested by user */
763 	if (gp->gp_bufsize > XFS_XATTR_LIST_MAX)
764 		return -ENOMEM;
765 	if (gp->gp_bufsize < xfs_getparents_rec_sizeof(1))
766 		return -EINVAL;
767 
768 	if (gp->gp_iflags & ~XFS_GETPARENTS_IFLAGS_ALL)
769 		return -EINVAL;
770 	if (gp->gp_reserved)
771 		return -EINVAL;
772 
773 	bufsize = round_down(gp->gp_bufsize, sizeof(uint64_t));
774 	gpx->krecords = kvzalloc(bufsize, GFP_KERNEL);
775 	if (!gpx->krecords) {
776 		bufsize = min(bufsize, PAGE_SIZE);
777 		gpx->krecords = kvzalloc(bufsize, GFP_KERNEL);
778 		if (!gpx->krecords)
779 			return -ENOMEM;
780 	}
781 
782 	gpx->context.dp = ip;
783 	gpx->context.resynch = 1;
784 	gpx->context.put_listent = xfs_getparents_put_listent;
785 	gpx->context.bufsize = bufsize;
786 	/* firstu is used to track the bytes filled in the buffer */
787 	gpx->context.firstu = 0;
788 
789 	/* Copy the cursor provided by caller */
790 	memcpy(&gpx->context.cursor, &gp->gp_cursor,
791 			sizeof(struct xfs_attrlist_cursor));
792 	gpx->count = 0;
793 	gp->gp_oflags = 0;
794 
795 	trace_xfs_getparents_begin(ip, gp, &gpx->context.cursor);
796 
797 	error = xfs_attr_list(&gpx->context);
798 	if (error)
799 		goto out_free_buf;
800 	if (gpx->context.seen_enough < 0) {
801 		error = gpx->context.seen_enough;
802 		goto out_free_buf;
803 	}
804 	xfs_getparents_expand_lastrec(gpx);
805 
806 	/* Update the caller with the current cursor position */
807 	memcpy(&gp->gp_cursor, &gpx->context.cursor,
808 			sizeof(struct xfs_attrlist_cursor));
809 
810 	/* Is this the root directory? */
811 	if (ip->i_ino == mp->m_sb.sb_rootino)
812 		gp->gp_oflags |= XFS_GETPARENTS_OFLAG_ROOT;
813 
814 	if (gpx->context.seen_enough == 0) {
815 		/*
816 		 * If we did not run out of buffer space, then we reached the
817 		 * end of the pptr recordset, so set the DONE flag.
818 		 */
819 		gp->gp_oflags |= XFS_GETPARENTS_OFLAG_DONE;
820 	} else if (gpx->count == 0) {
821 		/*
822 		 * If we ran out of buffer space before copying any parent
823 		 * pointers at all, the caller's buffer was too short.  Tell
824 		 * userspace that, erm, the message is too long.
825 		 */
826 		error = -EMSGSIZE;
827 		goto out_free_buf;
828 	}
829 
830 	trace_xfs_getparents_end(ip, gp, &gpx->context.cursor);
831 
832 	ASSERT(gpx->context.firstu <= gpx->gph.gph_request.gp_bufsize);
833 
834 	/* Copy the records to userspace. */
835 	if (copy_to_user(u64_to_user_ptr(gpx->gph.gph_request.gp_buffer),
836 				gpx->krecords, gpx->context.firstu))
837 		error = -EFAULT;
838 
839 out_free_buf:
840 	kvfree(gpx->krecords);
841 	gpx->krecords = NULL;
842 	return error;
843 }
844 
845 /* Retrieve the parents of this file and pass them back to userspace. */
846 int
847 xfs_ioc_getparents(
848 	struct file			*file,
849 	struct xfs_getparents __user	*ureq)
850 {
851 	struct xfs_getparents_ctx	gpx = {
852 		.ip			= XFS_I(file_inode(file)),
853 	};
854 	struct xfs_getparents		*kreq = &gpx.gph.gph_request;
855 	struct xfs_mount		*mp = gpx.ip->i_mount;
856 	int				error;
857 
858 	if (!capable(CAP_SYS_ADMIN))
859 		return -EPERM;
860 	if (!xfs_has_parent(mp))
861 		return -EOPNOTSUPP;
862 	if (copy_from_user(kreq, ureq, sizeof(*kreq)))
863 		return -EFAULT;
864 
865 	error = xfs_getparents(&gpx);
866 	if (error)
867 		return error;
868 
869 	if (copy_to_user(ureq, kreq, sizeof(*kreq)))
870 		return -EFAULT;
871 
872 	return 0;
873 }
874 
875 /* Retrieve the parents of this file handle and pass them back to userspace. */
876 int
877 xfs_ioc_getparents_by_handle(
878 	struct file			*file,
879 	struct xfs_getparents_by_handle __user	*ureq)
880 {
881 	struct xfs_getparents_ctx	gpx = { };
882 	struct xfs_inode		*ip = XFS_I(file_inode(file));
883 	struct xfs_mount		*mp = ip->i_mount;
884 	struct xfs_getparents_by_handle	*kreq = &gpx.gph;
885 	struct xfs_handle		*handle = &kreq->gph_handle;
886 	int				error;
887 
888 	if (!capable(CAP_SYS_ADMIN))
889 		return -EPERM;
890 	if (!xfs_has_parent(mp))
891 		return -EOPNOTSUPP;
892 	if (copy_from_user(kreq, ureq, sizeof(*kreq)))
893 		return -EFAULT;
894 
895 	/*
896 	 * We don't use exportfs_decode_fh because it does too much work here.
897 	 * If the handle refers to a directory, the exportfs code will walk
898 	 * upwards through the directory tree to connect the dentries to the
899 	 * root directory dentry.  For GETPARENTS we don't care about that
900 	 * because we're not actually going to open a file descriptor; we only
901 	 * want to open an inode and read its parent pointers.
902 	 *
903 	 * Note that xfs_scrub uses GETPARENTS to log that it will try to fix a
904 	 * corrupted file's metadata.  For this usecase we would really rather
905 	 * userspace single-step the path reconstruction to avoid loops or
906 	 * other strange things if the directory tree is corrupt.
907 	 */
908 	gpx.ip = xfs_khandle_to_inode(file, handle);
909 	if (IS_ERR(gpx.ip))
910 		return PTR_ERR(gpx.ip);
911 
912 	error = xfs_getparents(&gpx);
913 	if (error)
914 		goto out_rele;
915 
916 	if (copy_to_user(ureq, kreq, sizeof(*kreq)))
917 		error = -EFAULT;
918 
919 out_rele:
920 	xfs_irele(gpx.ip);
921 	return error;
922 }
923