Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_inode.h"
14#include "xfs_acl.h"
15#include "xfs_quota.h"
16#include "xfs_da_format.h"
17#include "xfs_da_btree.h"
18#include "xfs_attr.h"
19#include "xfs_trans.h"
20#include "xfs_trace.h"
21#include "xfs_icache.h"
22#include "xfs_symlink.h"
23#include "xfs_dir2.h"
24#include "xfs_iomap.h"
25#include "xfs_error.h"
26#include "xfs_ioctl.h"
27#include "xfs_xattr.h"
28
29#include <linux/posix_acl.h>
30#include <linux/security.h>
31#include <linux/iversion.h>
32#include <linux/fiemap.h>
33
34/*
35 * Directories have different lock order w.r.t. mmap_lock compared to regular
36 * files. This is due to readdir potentially triggering page faults on a user
37 * buffer inside filldir(), and this happens with the ilock on the directory
38 * held. For regular files, the lock order is the other way around - the
39 * mmap_lock is taken during the page fault, and then we lock the ilock to do
40 * block mapping. Hence we need a different class for the directory ilock so
41 * that lockdep can tell them apart.
42 */
43static struct lock_class_key xfs_nondir_ilock_class;
44static struct lock_class_key xfs_dir_ilock_class;
45
46static int
47xfs_initxattrs(
48 struct inode *inode,
49 const struct xattr *xattr_array,
50 void *fs_info)
51{
52 const struct xattr *xattr;
53 struct xfs_inode *ip = XFS_I(inode);
54 int error = 0;
55
56 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
57 struct xfs_da_args args = {
58 .dp = ip,
59 .attr_filter = XFS_ATTR_SECURE,
60 .name = xattr->name,
61 .namelen = strlen(xattr->name),
62 .value = xattr->value,
63 .valuelen = xattr->value_len,
64 };
65 error = xfs_attr_change(&args);
66 if (error < 0)
67 break;
68 }
69 return error;
70}
71
72/*
73 * Hook in SELinux. This is not quite correct yet, what we really need
74 * here (as we do for default ACLs) is a mechanism by which creation of
75 * these attrs can be journalled at inode creation time (along with the
76 * inode, of course, such that log replay can't cause these to be lost).
77 */
78int
79xfs_inode_init_security(
80 struct inode *inode,
81 struct inode *dir,
82 const struct qstr *qstr)
83{
84 return security_inode_init_security(inode, dir, qstr,
85 &xfs_initxattrs, NULL);
86}
87
88static void
89xfs_dentry_to_name(
90 struct xfs_name *namep,
91 struct dentry *dentry)
92{
93 namep->name = dentry->d_name.name;
94 namep->len = dentry->d_name.len;
95 namep->type = XFS_DIR3_FT_UNKNOWN;
96}
97
98static int
99xfs_dentry_mode_to_name(
100 struct xfs_name *namep,
101 struct dentry *dentry,
102 int mode)
103{
104 namep->name = dentry->d_name.name;
105 namep->len = dentry->d_name.len;
106 namep->type = xfs_mode_to_ftype(mode);
107
108 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
109 return -EFSCORRUPTED;
110
111 return 0;
112}
113
114STATIC void
115xfs_cleanup_inode(
116 struct inode *dir,
117 struct inode *inode,
118 struct dentry *dentry)
119{
120 struct xfs_name teardown;
121
122 /* Oh, the horror.
123 * If we can't add the ACL or we fail in
124 * xfs_inode_init_security we must back out.
125 * ENOSPC can hit here, among other things.
126 */
127 xfs_dentry_to_name(&teardown, dentry);
128
129 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
130}
131
132/*
133 * Check to see if we are likely to need an extended attribute to be added to
134 * the inode we are about to allocate. This allows the attribute fork to be
135 * created during the inode allocation, reducing the number of transactions we
136 * need to do in this fast path.
137 *
138 * The security checks are optimistic, but not guaranteed. The two LSMs that
139 * require xattrs to be added here (selinux and smack) are also the only two
140 * LSMs that add a sb->s_security structure to the superblock. Hence if security
141 * is enabled and sb->s_security is set, we have a pretty good idea that we are
142 * going to be asked to add a security xattr immediately after allocating the
143 * xfs inode and instantiating the VFS inode.
144 */
145static inline bool
146xfs_create_need_xattr(
147 struct inode *dir,
148 struct posix_acl *default_acl,
149 struct posix_acl *acl)
150{
151 if (acl)
152 return true;
153 if (default_acl)
154 return true;
155#if IS_ENABLED(CONFIG_SECURITY)
156 if (dir->i_sb->s_security)
157 return true;
158#endif
159 return false;
160}
161
162
163STATIC int
164xfs_generic_create(
165 struct mnt_idmap *idmap,
166 struct inode *dir,
167 struct dentry *dentry,
168 umode_t mode,
169 dev_t rdev,
170 struct file *tmpfile) /* unnamed file */
171{
172 struct inode *inode;
173 struct xfs_inode *ip = NULL;
174 struct posix_acl *default_acl, *acl;
175 struct xfs_name name;
176 int error;
177
178 /*
179 * Irix uses Missed'em'V split, but doesn't want to see
180 * the upper 5 bits of (14bit) major.
181 */
182 if (S_ISCHR(mode) || S_ISBLK(mode)) {
183 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
184 return -EINVAL;
185 } else {
186 rdev = 0;
187 }
188
189 error = posix_acl_create(dir, &mode, &default_acl, &acl);
190 if (error)
191 return error;
192
193 /* Verify mode is valid also for tmpfile case */
194 error = xfs_dentry_mode_to_name(&name, dentry, mode);
195 if (unlikely(error))
196 goto out_free_acl;
197
198 if (!tmpfile) {
199 error = xfs_create(idmap, XFS_I(dir), &name, mode, rdev,
200 xfs_create_need_xattr(dir, default_acl, acl),
201 &ip);
202 } else {
203 error = xfs_create_tmpfile(idmap, XFS_I(dir), mode, &ip);
204 }
205 if (unlikely(error))
206 goto out_free_acl;
207
208 inode = VFS_I(ip);
209
210 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
211 if (unlikely(error))
212 goto out_cleanup_inode;
213
214 if (default_acl) {
215 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
216 if (error)
217 goto out_cleanup_inode;
218 }
219 if (acl) {
220 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
221 if (error)
222 goto out_cleanup_inode;
223 }
224
225 xfs_setup_iops(ip);
226
227 if (tmpfile) {
228 /*
229 * The VFS requires that any inode fed to d_tmpfile must have
230 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
231 * However, we created the temp file with nlink == 0 because
232 * we're not allowed to put an inode with nlink > 0 on the
233 * unlinked list. Therefore we have to set nlink to 1 so that
234 * d_tmpfile can immediately set it back to zero.
235 */
236 set_nlink(inode, 1);
237 d_tmpfile(tmpfile, inode);
238 } else
239 d_instantiate(dentry, inode);
240
241 xfs_finish_inode_setup(ip);
242
243 out_free_acl:
244 posix_acl_release(default_acl);
245 posix_acl_release(acl);
246 return error;
247
248 out_cleanup_inode:
249 xfs_finish_inode_setup(ip);
250 if (!tmpfile)
251 xfs_cleanup_inode(dir, inode, dentry);
252 xfs_irele(ip);
253 goto out_free_acl;
254}
255
256STATIC int
257xfs_vn_mknod(
258 struct mnt_idmap *idmap,
259 struct inode *dir,
260 struct dentry *dentry,
261 umode_t mode,
262 dev_t rdev)
263{
264 return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL);
265}
266
267STATIC int
268xfs_vn_create(
269 struct mnt_idmap *idmap,
270 struct inode *dir,
271 struct dentry *dentry,
272 umode_t mode,
273 bool flags)
274{
275 return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL);
276}
277
278STATIC int
279xfs_vn_mkdir(
280 struct mnt_idmap *idmap,
281 struct inode *dir,
282 struct dentry *dentry,
283 umode_t mode)
284{
285 return xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL);
286}
287
288STATIC struct dentry *
289xfs_vn_lookup(
290 struct inode *dir,
291 struct dentry *dentry,
292 unsigned int flags)
293{
294 struct inode *inode;
295 struct xfs_inode *cip;
296 struct xfs_name name;
297 int error;
298
299 if (dentry->d_name.len >= MAXNAMELEN)
300 return ERR_PTR(-ENAMETOOLONG);
301
302 xfs_dentry_to_name(&name, dentry);
303 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
304 if (likely(!error))
305 inode = VFS_I(cip);
306 else if (likely(error == -ENOENT))
307 inode = NULL;
308 else
309 inode = ERR_PTR(error);
310 return d_splice_alias(inode, dentry);
311}
312
313STATIC struct dentry *
314xfs_vn_ci_lookup(
315 struct inode *dir,
316 struct dentry *dentry,
317 unsigned int flags)
318{
319 struct xfs_inode *ip;
320 struct xfs_name xname;
321 struct xfs_name ci_name;
322 struct qstr dname;
323 int error;
324
325 if (dentry->d_name.len >= MAXNAMELEN)
326 return ERR_PTR(-ENAMETOOLONG);
327
328 xfs_dentry_to_name(&xname, dentry);
329 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
330 if (unlikely(error)) {
331 if (unlikely(error != -ENOENT))
332 return ERR_PTR(error);
333 /*
334 * call d_add(dentry, NULL) here when d_drop_negative_children
335 * is called in xfs_vn_mknod (ie. allow negative dentries
336 * with CI filesystems).
337 */
338 return NULL;
339 }
340
341 /* if exact match, just splice and exit */
342 if (!ci_name.name)
343 return d_splice_alias(VFS_I(ip), dentry);
344
345 /* else case-insensitive match... */
346 dname.name = ci_name.name;
347 dname.len = ci_name.len;
348 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
349 kmem_free(ci_name.name);
350 return dentry;
351}
352
353STATIC int
354xfs_vn_link(
355 struct dentry *old_dentry,
356 struct inode *dir,
357 struct dentry *dentry)
358{
359 struct inode *inode = d_inode(old_dentry);
360 struct xfs_name name;
361 int error;
362
363 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
364 if (unlikely(error))
365 return error;
366
367 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
368 if (unlikely(error))
369 return error;
370
371 ihold(inode);
372 d_instantiate(dentry, inode);
373 return 0;
374}
375
376STATIC int
377xfs_vn_unlink(
378 struct inode *dir,
379 struct dentry *dentry)
380{
381 struct xfs_name name;
382 int error;
383
384 xfs_dentry_to_name(&name, dentry);
385
386 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
387 if (error)
388 return error;
389
390 /*
391 * With unlink, the VFS makes the dentry "negative": no inode,
392 * but still hashed. This is incompatible with case-insensitive
393 * mode, so invalidate (unhash) the dentry in CI-mode.
394 */
395 if (xfs_has_asciici(XFS_M(dir->i_sb)))
396 d_invalidate(dentry);
397 return 0;
398}
399
400STATIC int
401xfs_vn_symlink(
402 struct mnt_idmap *idmap,
403 struct inode *dir,
404 struct dentry *dentry,
405 const char *symname)
406{
407 struct inode *inode;
408 struct xfs_inode *cip = NULL;
409 struct xfs_name name;
410 int error;
411 umode_t mode;
412
413 mode = S_IFLNK |
414 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
415 error = xfs_dentry_mode_to_name(&name, dentry, mode);
416 if (unlikely(error))
417 goto out;
418
419 error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip);
420 if (unlikely(error))
421 goto out;
422
423 inode = VFS_I(cip);
424
425 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
426 if (unlikely(error))
427 goto out_cleanup_inode;
428
429 xfs_setup_iops(cip);
430
431 d_instantiate(dentry, inode);
432 xfs_finish_inode_setup(cip);
433 return 0;
434
435 out_cleanup_inode:
436 xfs_finish_inode_setup(cip);
437 xfs_cleanup_inode(dir, inode, dentry);
438 xfs_irele(cip);
439 out:
440 return error;
441}
442
443STATIC int
444xfs_vn_rename(
445 struct mnt_idmap *idmap,
446 struct inode *odir,
447 struct dentry *odentry,
448 struct inode *ndir,
449 struct dentry *ndentry,
450 unsigned int flags)
451{
452 struct inode *new_inode = d_inode(ndentry);
453 int omode = 0;
454 int error;
455 struct xfs_name oname;
456 struct xfs_name nname;
457
458 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
459 return -EINVAL;
460
461 /* if we are exchanging files, we need to set i_mode of both files */
462 if (flags & RENAME_EXCHANGE)
463 omode = d_inode(ndentry)->i_mode;
464
465 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
466 if (omode && unlikely(error))
467 return error;
468
469 error = xfs_dentry_mode_to_name(&nname, ndentry,
470 d_inode(odentry)->i_mode);
471 if (unlikely(error))
472 return error;
473
474 return xfs_rename(idmap, XFS_I(odir), &oname,
475 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
476 new_inode ? XFS_I(new_inode) : NULL, flags);
477}
478
479/*
480 * careful here - this function can get called recursively, so
481 * we need to be very careful about how much stack we use.
482 * uio is kmalloced for this reason...
483 */
484STATIC const char *
485xfs_vn_get_link(
486 struct dentry *dentry,
487 struct inode *inode,
488 struct delayed_call *done)
489{
490 char *link;
491 int error = -ENOMEM;
492
493 if (!dentry)
494 return ERR_PTR(-ECHILD);
495
496 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
497 if (!link)
498 goto out_err;
499
500 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
501 if (unlikely(error))
502 goto out_kfree;
503
504 set_delayed_call(done, kfree_link, link);
505 return link;
506
507 out_kfree:
508 kfree(link);
509 out_err:
510 return ERR_PTR(error);
511}
512
513static uint32_t
514xfs_stat_blksize(
515 struct xfs_inode *ip)
516{
517 struct xfs_mount *mp = ip->i_mount;
518
519 /*
520 * If the file blocks are being allocated from a realtime volume, then
521 * always return the realtime extent size.
522 */
523 if (XFS_IS_REALTIME_INODE(ip))
524 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip));
525
526 /*
527 * Allow large block sizes to be reported to userspace programs if the
528 * "largeio" mount option is used.
529 *
530 * If compatibility mode is specified, simply return the basic unit of
531 * caching so that we don't get inefficient read/modify/write I/O from
532 * user apps. Otherwise....
533 *
534 * If the underlying volume is a stripe, then return the stripe width in
535 * bytes as the recommended I/O size. It is not a stripe and we've set a
536 * default buffered I/O size, return that, otherwise return the compat
537 * default.
538 */
539 if (xfs_has_large_iosize(mp)) {
540 if (mp->m_swidth)
541 return XFS_FSB_TO_B(mp, mp->m_swidth);
542 if (xfs_has_allocsize(mp))
543 return 1U << mp->m_allocsize_log;
544 }
545
546 return PAGE_SIZE;
547}
548
549STATIC int
550xfs_vn_getattr(
551 struct mnt_idmap *idmap,
552 const struct path *path,
553 struct kstat *stat,
554 u32 request_mask,
555 unsigned int query_flags)
556{
557 struct inode *inode = d_inode(path->dentry);
558 struct xfs_inode *ip = XFS_I(inode);
559 struct xfs_mount *mp = ip->i_mount;
560 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
561 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
562
563 trace_xfs_getattr(ip);
564
565 if (xfs_is_shutdown(mp))
566 return -EIO;
567
568 stat->size = XFS_ISIZE(ip);
569 stat->dev = inode->i_sb->s_dev;
570 stat->mode = inode->i_mode;
571 stat->nlink = inode->i_nlink;
572 stat->uid = vfsuid_into_kuid(vfsuid);
573 stat->gid = vfsgid_into_kgid(vfsgid);
574 stat->ino = ip->i_ino;
575 stat->atime = inode_get_atime(inode);
576 stat->mtime = inode_get_mtime(inode);
577 stat->ctime = inode_get_ctime(inode);
578 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
579
580 if (xfs_has_v3inodes(mp)) {
581 if (request_mask & STATX_BTIME) {
582 stat->result_mask |= STATX_BTIME;
583 stat->btime = ip->i_crtime;
584 }
585 }
586
587 if ((request_mask & STATX_CHANGE_COOKIE) && IS_I_VERSION(inode)) {
588 stat->change_cookie = inode_query_iversion(inode);
589 stat->result_mask |= STATX_CHANGE_COOKIE;
590 }
591
592 /*
593 * Note: If you add another clause to set an attribute flag, please
594 * update attributes_mask below.
595 */
596 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
597 stat->attributes |= STATX_ATTR_IMMUTABLE;
598 if (ip->i_diflags & XFS_DIFLAG_APPEND)
599 stat->attributes |= STATX_ATTR_APPEND;
600 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
601 stat->attributes |= STATX_ATTR_NODUMP;
602
603 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
604 STATX_ATTR_APPEND |
605 STATX_ATTR_NODUMP);
606
607 switch (inode->i_mode & S_IFMT) {
608 case S_IFBLK:
609 case S_IFCHR:
610 stat->blksize = BLKDEV_IOSIZE;
611 stat->rdev = inode->i_rdev;
612 break;
613 case S_IFREG:
614 if (request_mask & STATX_DIOALIGN) {
615 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
616 struct block_device *bdev = target->bt_bdev;
617
618 stat->result_mask |= STATX_DIOALIGN;
619 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
620 stat->dio_offset_align = bdev_logical_block_size(bdev);
621 }
622 fallthrough;
623 default:
624 stat->blksize = xfs_stat_blksize(ip);
625 stat->rdev = 0;
626 break;
627 }
628
629 return 0;
630}
631
632static int
633xfs_vn_change_ok(
634 struct mnt_idmap *idmap,
635 struct dentry *dentry,
636 struct iattr *iattr)
637{
638 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
639
640 if (xfs_is_readonly(mp))
641 return -EROFS;
642
643 if (xfs_is_shutdown(mp))
644 return -EIO;
645
646 return setattr_prepare(idmap, dentry, iattr);
647}
648
649/*
650 * Set non-size attributes of an inode.
651 *
652 * Caution: The caller of this function is responsible for calling
653 * setattr_prepare() or otherwise verifying the change is fine.
654 */
655static int
656xfs_setattr_nonsize(
657 struct mnt_idmap *idmap,
658 struct dentry *dentry,
659 struct xfs_inode *ip,
660 struct iattr *iattr)
661{
662 xfs_mount_t *mp = ip->i_mount;
663 struct inode *inode = VFS_I(ip);
664 int mask = iattr->ia_valid;
665 xfs_trans_t *tp;
666 int error;
667 kuid_t uid = GLOBAL_ROOT_UID;
668 kgid_t gid = GLOBAL_ROOT_GID;
669 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
670 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL;
671
672 ASSERT((mask & ATTR_SIZE) == 0);
673
674 /*
675 * If disk quotas is on, we make sure that the dquots do exist on disk,
676 * before we start any other transactions. Trying to do this later
677 * is messy. We don't care to take a readlock to look at the ids
678 * in inode here, because we can't hold it across the trans_reserve.
679 * If the IDs do change before we take the ilock, we're covered
680 * because the i_*dquot fields will get updated anyway.
681 */
682 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
683 uint qflags = 0;
684
685 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
686 uid = from_vfsuid(idmap, i_user_ns(inode),
687 iattr->ia_vfsuid);
688 qflags |= XFS_QMOPT_UQUOTA;
689 } else {
690 uid = inode->i_uid;
691 }
692 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
693 gid = from_vfsgid(idmap, i_user_ns(inode),
694 iattr->ia_vfsgid);
695 qflags |= XFS_QMOPT_GQUOTA;
696 } else {
697 gid = inode->i_gid;
698 }
699
700 /*
701 * We take a reference when we initialize udqp and gdqp,
702 * so it is important that we never blindly double trip on
703 * the same variable. See xfs_create() for an example.
704 */
705 ASSERT(udqp == NULL);
706 ASSERT(gdqp == NULL);
707 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
708 qflags, &udqp, &gdqp, NULL);
709 if (error)
710 return error;
711 }
712
713 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
714 has_capability_noaudit(current, CAP_FOWNER), &tp);
715 if (error)
716 goto out_dqrele;
717
718 /*
719 * Register quota modifications in the transaction. Must be the owner
720 * or privileged. These IDs could have changed since we last looked at
721 * them. But, we're assured that if the ownership did change while we
722 * didn't have the inode locked, inode's dquot(s) would have changed
723 * also.
724 */
725 if (XFS_IS_UQUOTA_ON(mp) &&
726 i_uid_needs_update(idmap, iattr, inode)) {
727 ASSERT(udqp);
728 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp);
729 }
730 if (XFS_IS_GQUOTA_ON(mp) &&
731 i_gid_needs_update(idmap, iattr, inode)) {
732 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp));
733 ASSERT(gdqp);
734 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp);
735 }
736
737 setattr_copy(idmap, inode, iattr);
738 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
739
740 XFS_STATS_INC(mp, xs_ig_attrchg);
741
742 if (xfs_has_wsync(mp))
743 xfs_trans_set_sync(tp);
744 error = xfs_trans_commit(tp);
745
746 /*
747 * Release any dquot(s) the inode had kept before chown.
748 */
749 xfs_qm_dqrele(old_udqp);
750 xfs_qm_dqrele(old_gdqp);
751 xfs_qm_dqrele(udqp);
752 xfs_qm_dqrele(gdqp);
753
754 if (error)
755 return error;
756
757 /*
758 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
759 * update. We could avoid this with linked transactions
760 * and passing down the transaction pointer all the way
761 * to attr_set. No previous user of the generic
762 * Posix ACL code seems to care about this issue either.
763 */
764 if (mask & ATTR_MODE) {
765 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
766 if (error)
767 return error;
768 }
769
770 return 0;
771
772out_dqrele:
773 xfs_qm_dqrele(udqp);
774 xfs_qm_dqrele(gdqp);
775 return error;
776}
777
778/*
779 * Truncate file. Must have write permission and not be a directory.
780 *
781 * Caution: The caller of this function is responsible for calling
782 * setattr_prepare() or otherwise verifying the change is fine.
783 */
784STATIC int
785xfs_setattr_size(
786 struct mnt_idmap *idmap,
787 struct dentry *dentry,
788 struct xfs_inode *ip,
789 struct iattr *iattr)
790{
791 struct xfs_mount *mp = ip->i_mount;
792 struct inode *inode = VFS_I(ip);
793 xfs_off_t oldsize, newsize;
794 struct xfs_trans *tp;
795 int error;
796 uint lock_flags = 0;
797 bool did_zeroing = false;
798
799 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
800 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
801 ASSERT(S_ISREG(inode->i_mode));
802 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
803 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
804
805 oldsize = inode->i_size;
806 newsize = iattr->ia_size;
807
808 /*
809 * Short circuit the truncate case for zero length files.
810 */
811 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
812 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
813 return 0;
814
815 /*
816 * Use the regular setattr path to update the timestamps.
817 */
818 iattr->ia_valid &= ~ATTR_SIZE;
819 return xfs_setattr_nonsize(idmap, dentry, ip, iattr);
820 }
821
822 /*
823 * Make sure that the dquots are attached to the inode.
824 */
825 error = xfs_qm_dqattach(ip);
826 if (error)
827 return error;
828
829 /*
830 * Wait for all direct I/O to complete.
831 */
832 inode_dio_wait(inode);
833
834 /*
835 * File data changes must be complete before we start the transaction to
836 * modify the inode. This needs to be done before joining the inode to
837 * the transaction because the inode cannot be unlocked once it is a
838 * part of the transaction.
839 *
840 * Start with zeroing any data beyond EOF that we may expose on file
841 * extension, or zeroing out the rest of the block on a downward
842 * truncate.
843 */
844 if (newsize > oldsize) {
845 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
846 error = xfs_zero_range(ip, oldsize, newsize - oldsize,
847 &did_zeroing);
848 } else {
849 /*
850 * iomap won't detect a dirty page over an unwritten block (or a
851 * cow block over a hole) and subsequently skips zeroing the
852 * newly post-EOF portion of the page. Flush the new EOF to
853 * convert the block before the pagecache truncate.
854 */
855 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
856 newsize);
857 if (error)
858 return error;
859 error = xfs_truncate_page(ip, newsize, &did_zeroing);
860 }
861
862 if (error)
863 return error;
864
865 /*
866 * We've already locked out new page faults, so now we can safely remove
867 * pages from the page cache knowing they won't get refaulted until we
868 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
869 * complete. The truncate_setsize() call also cleans partial EOF page
870 * PTEs on extending truncates and hence ensures sub-page block size
871 * filesystems are correctly handled, too.
872 *
873 * We have to do all the page cache truncate work outside the
874 * transaction context as the "lock" order is page lock->log space
875 * reservation as defined by extent allocation in the writeback path.
876 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
877 * having already truncated the in-memory version of the file (i.e. made
878 * user visible changes). There's not much we can do about this, except
879 * to hope that the caller sees ENOMEM and retries the truncate
880 * operation.
881 *
882 * And we update in-core i_size and truncate page cache beyond newsize
883 * before writeback the [i_disk_size, newsize] range, so we're
884 * guaranteed not to write stale data past the new EOF on truncate down.
885 */
886 truncate_setsize(inode, newsize);
887
888 /*
889 * We are going to log the inode size change in this transaction so
890 * any previous writes that are beyond the on disk EOF and the new
891 * EOF that have not been written out need to be written here. If we
892 * do not write the data out, we expose ourselves to the null files
893 * problem. Note that this includes any block zeroing we did above;
894 * otherwise those blocks may not be zeroed after a crash.
895 */
896 if (did_zeroing ||
897 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
898 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
899 ip->i_disk_size, newsize - 1);
900 if (error)
901 return error;
902 }
903
904 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
905 if (error)
906 return error;
907
908 lock_flags |= XFS_ILOCK_EXCL;
909 xfs_ilock(ip, XFS_ILOCK_EXCL);
910 xfs_trans_ijoin(tp, ip, 0);
911
912 /*
913 * Only change the c/mtime if we are changing the size or we are
914 * explicitly asked to change it. This handles the semantic difference
915 * between truncate() and ftruncate() as implemented in the VFS.
916 *
917 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
918 * special case where we need to update the times despite not having
919 * these flags set. For all other operations the VFS set these flags
920 * explicitly if it wants a timestamp update.
921 */
922 if (newsize != oldsize &&
923 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
924 iattr->ia_ctime = iattr->ia_mtime =
925 current_time(inode);
926 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
927 }
928
929 /*
930 * The first thing we do is set the size to new_size permanently on
931 * disk. This way we don't have to worry about anyone ever being able
932 * to look at the data being freed even in the face of a crash.
933 * What we're getting around here is the case where we free a block, it
934 * is allocated to another file, it is written to, and then we crash.
935 * If the new data gets written to the file but the log buffers
936 * containing the free and reallocation don't, then we'd end up with
937 * garbage in the blocks being freed. As long as we make the new size
938 * permanent before actually freeing any blocks it doesn't matter if
939 * they get written to.
940 */
941 ip->i_disk_size = newsize;
942 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
943
944 if (newsize <= oldsize) {
945 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
946 if (error)
947 goto out_trans_cancel;
948
949 /*
950 * Truncated "down", so we're removing references to old data
951 * here - if we delay flushing for a long time, we expose
952 * ourselves unduly to the notorious NULL files problem. So,
953 * we mark this inode and flush it when the file is closed,
954 * and do not wait the usual (long) time for writeout.
955 */
956 xfs_iflags_set(ip, XFS_ITRUNCATED);
957
958 /* A truncate down always removes post-EOF blocks. */
959 xfs_inode_clear_eofblocks_tag(ip);
960 }
961
962 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
963 setattr_copy(idmap, inode, iattr);
964 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
965
966 XFS_STATS_INC(mp, xs_ig_attrchg);
967
968 if (xfs_has_wsync(mp))
969 xfs_trans_set_sync(tp);
970
971 error = xfs_trans_commit(tp);
972out_unlock:
973 if (lock_flags)
974 xfs_iunlock(ip, lock_flags);
975 return error;
976
977out_trans_cancel:
978 xfs_trans_cancel(tp);
979 goto out_unlock;
980}
981
982int
983xfs_vn_setattr_size(
984 struct mnt_idmap *idmap,
985 struct dentry *dentry,
986 struct iattr *iattr)
987{
988 struct xfs_inode *ip = XFS_I(d_inode(dentry));
989 int error;
990
991 trace_xfs_setattr(ip);
992
993 error = xfs_vn_change_ok(idmap, dentry, iattr);
994 if (error)
995 return error;
996 return xfs_setattr_size(idmap, dentry, ip, iattr);
997}
998
999STATIC int
1000xfs_vn_setattr(
1001 struct mnt_idmap *idmap,
1002 struct dentry *dentry,
1003 struct iattr *iattr)
1004{
1005 struct inode *inode = d_inode(dentry);
1006 struct xfs_inode *ip = XFS_I(inode);
1007 int error;
1008
1009 if (iattr->ia_valid & ATTR_SIZE) {
1010 uint iolock;
1011
1012 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1013 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1014
1015 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1016 if (error) {
1017 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1018 return error;
1019 }
1020
1021 error = xfs_vn_setattr_size(idmap, dentry, iattr);
1022 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1023 } else {
1024 trace_xfs_setattr(ip);
1025
1026 error = xfs_vn_change_ok(idmap, dentry, iattr);
1027 if (!error)
1028 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr);
1029 }
1030
1031 return error;
1032}
1033
1034STATIC int
1035xfs_vn_update_time(
1036 struct inode *inode,
1037 int flags)
1038{
1039 struct xfs_inode *ip = XFS_I(inode);
1040 struct xfs_mount *mp = ip->i_mount;
1041 int log_flags = XFS_ILOG_TIMESTAMP;
1042 struct xfs_trans *tp;
1043 int error;
1044 struct timespec64 now;
1045
1046 trace_xfs_update_time(ip);
1047
1048 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1049 if (!((flags & S_VERSION) &&
1050 inode_maybe_inc_iversion(inode, false))) {
1051 generic_update_time(inode, flags);
1052 return 0;
1053 }
1054
1055 /* Capture the iversion update that just occurred */
1056 log_flags |= XFS_ILOG_CORE;
1057 }
1058
1059 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1060 if (error)
1061 return error;
1062
1063 xfs_ilock(ip, XFS_ILOCK_EXCL);
1064 if (flags & (S_CTIME|S_MTIME))
1065 now = inode_set_ctime_current(inode);
1066 else
1067 now = current_time(inode);
1068
1069 if (flags & S_MTIME)
1070 inode_set_mtime_to_ts(inode, now);
1071 if (flags & S_ATIME)
1072 inode_set_atime_to_ts(inode, now);
1073
1074 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1075 xfs_trans_log_inode(tp, ip, log_flags);
1076 return xfs_trans_commit(tp);
1077}
1078
1079STATIC int
1080xfs_vn_fiemap(
1081 struct inode *inode,
1082 struct fiemap_extent_info *fieinfo,
1083 u64 start,
1084 u64 length)
1085{
1086 int error;
1087
1088 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1089 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1090 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1091 error = iomap_fiemap(inode, fieinfo, start, length,
1092 &xfs_xattr_iomap_ops);
1093 } else {
1094 error = iomap_fiemap(inode, fieinfo, start, length,
1095 &xfs_read_iomap_ops);
1096 }
1097 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1098
1099 return error;
1100}
1101
1102STATIC int
1103xfs_vn_tmpfile(
1104 struct mnt_idmap *idmap,
1105 struct inode *dir,
1106 struct file *file,
1107 umode_t mode)
1108{
1109 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file);
1110
1111 return finish_open_simple(file, err);
1112}
1113
1114static const struct inode_operations xfs_inode_operations = {
1115 .get_inode_acl = xfs_get_acl,
1116 .set_acl = xfs_set_acl,
1117 .getattr = xfs_vn_getattr,
1118 .setattr = xfs_vn_setattr,
1119 .listxattr = xfs_vn_listxattr,
1120 .fiemap = xfs_vn_fiemap,
1121 .update_time = xfs_vn_update_time,
1122 .fileattr_get = xfs_fileattr_get,
1123 .fileattr_set = xfs_fileattr_set,
1124};
1125
1126static const struct inode_operations xfs_dir_inode_operations = {
1127 .create = xfs_vn_create,
1128 .lookup = xfs_vn_lookup,
1129 .link = xfs_vn_link,
1130 .unlink = xfs_vn_unlink,
1131 .symlink = xfs_vn_symlink,
1132 .mkdir = xfs_vn_mkdir,
1133 /*
1134 * Yes, XFS uses the same method for rmdir and unlink.
1135 *
1136 * There are some subtile differences deeper in the code,
1137 * but we use S_ISDIR to check for those.
1138 */
1139 .rmdir = xfs_vn_unlink,
1140 .mknod = xfs_vn_mknod,
1141 .rename = xfs_vn_rename,
1142 .get_inode_acl = xfs_get_acl,
1143 .set_acl = xfs_set_acl,
1144 .getattr = xfs_vn_getattr,
1145 .setattr = xfs_vn_setattr,
1146 .listxattr = xfs_vn_listxattr,
1147 .update_time = xfs_vn_update_time,
1148 .tmpfile = xfs_vn_tmpfile,
1149 .fileattr_get = xfs_fileattr_get,
1150 .fileattr_set = xfs_fileattr_set,
1151};
1152
1153static const struct inode_operations xfs_dir_ci_inode_operations = {
1154 .create = xfs_vn_create,
1155 .lookup = xfs_vn_ci_lookup,
1156 .link = xfs_vn_link,
1157 .unlink = xfs_vn_unlink,
1158 .symlink = xfs_vn_symlink,
1159 .mkdir = xfs_vn_mkdir,
1160 /*
1161 * Yes, XFS uses the same method for rmdir and unlink.
1162 *
1163 * There are some subtile differences deeper in the code,
1164 * but we use S_ISDIR to check for those.
1165 */
1166 .rmdir = xfs_vn_unlink,
1167 .mknod = xfs_vn_mknod,
1168 .rename = xfs_vn_rename,
1169 .get_inode_acl = xfs_get_acl,
1170 .set_acl = xfs_set_acl,
1171 .getattr = xfs_vn_getattr,
1172 .setattr = xfs_vn_setattr,
1173 .listxattr = xfs_vn_listxattr,
1174 .update_time = xfs_vn_update_time,
1175 .tmpfile = xfs_vn_tmpfile,
1176 .fileattr_get = xfs_fileattr_get,
1177 .fileattr_set = xfs_fileattr_set,
1178};
1179
1180static const struct inode_operations xfs_symlink_inode_operations = {
1181 .get_link = xfs_vn_get_link,
1182 .getattr = xfs_vn_getattr,
1183 .setattr = xfs_vn_setattr,
1184 .listxattr = xfs_vn_listxattr,
1185 .update_time = xfs_vn_update_time,
1186};
1187
1188/* Figure out if this file actually supports DAX. */
1189static bool
1190xfs_inode_supports_dax(
1191 struct xfs_inode *ip)
1192{
1193 struct xfs_mount *mp = ip->i_mount;
1194
1195 /* Only supported on regular files. */
1196 if (!S_ISREG(VFS_I(ip)->i_mode))
1197 return false;
1198
1199 /* Block size must match page size */
1200 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1201 return false;
1202
1203 /* Device has to support DAX too. */
1204 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1205}
1206
1207static bool
1208xfs_inode_should_enable_dax(
1209 struct xfs_inode *ip)
1210{
1211 if (!IS_ENABLED(CONFIG_FS_DAX))
1212 return false;
1213 if (xfs_has_dax_never(ip->i_mount))
1214 return false;
1215 if (!xfs_inode_supports_dax(ip))
1216 return false;
1217 if (xfs_has_dax_always(ip->i_mount))
1218 return true;
1219 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1220 return true;
1221 return false;
1222}
1223
1224void
1225xfs_diflags_to_iflags(
1226 struct xfs_inode *ip,
1227 bool init)
1228{
1229 struct inode *inode = VFS_I(ip);
1230 unsigned int xflags = xfs_ip2xflags(ip);
1231 unsigned int flags = 0;
1232
1233 ASSERT(!(IS_DAX(inode) && init));
1234
1235 if (xflags & FS_XFLAG_IMMUTABLE)
1236 flags |= S_IMMUTABLE;
1237 if (xflags & FS_XFLAG_APPEND)
1238 flags |= S_APPEND;
1239 if (xflags & FS_XFLAG_SYNC)
1240 flags |= S_SYNC;
1241 if (xflags & FS_XFLAG_NOATIME)
1242 flags |= S_NOATIME;
1243 if (init && xfs_inode_should_enable_dax(ip))
1244 flags |= S_DAX;
1245
1246 /*
1247 * S_DAX can only be set during inode initialization and is never set by
1248 * the VFS, so we cannot mask off S_DAX in i_flags.
1249 */
1250 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1251 inode->i_flags |= flags;
1252}
1253
1254/*
1255 * Initialize the Linux inode.
1256 *
1257 * When reading existing inodes from disk this is called directly from xfs_iget,
1258 * when creating a new inode it is called from xfs_init_new_inode after setting
1259 * up the inode. These callers have different criteria for clearing XFS_INEW, so
1260 * leave it up to the caller to deal with unlocking the inode appropriately.
1261 */
1262void
1263xfs_setup_inode(
1264 struct xfs_inode *ip)
1265{
1266 struct inode *inode = &ip->i_vnode;
1267 gfp_t gfp_mask;
1268
1269 inode->i_ino = ip->i_ino;
1270 inode->i_state |= I_NEW;
1271
1272 inode_sb_list_add(inode);
1273 /* make the inode look hashed for the writeback code */
1274 inode_fake_hash(inode);
1275
1276 i_size_write(inode, ip->i_disk_size);
1277 xfs_diflags_to_iflags(ip, true);
1278
1279 if (S_ISDIR(inode->i_mode)) {
1280 /*
1281 * We set the i_rwsem class here to avoid potential races with
1282 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1283 * after a filehandle lookup has already found the inode in
1284 * cache before it has been unlocked via unlock_new_inode().
1285 */
1286 lockdep_set_class(&inode->i_rwsem,
1287 &inode->i_sb->s_type->i_mutex_dir_key);
1288 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1289 } else {
1290 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1291 }
1292
1293 /*
1294 * Ensure all page cache allocations are done from GFP_NOFS context to
1295 * prevent direct reclaim recursion back into the filesystem and blowing
1296 * stacks or deadlocking.
1297 */
1298 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1299 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1300
1301 /*
1302 * For real-time inodes update the stable write flags to that of the RT
1303 * device instead of the data device.
1304 */
1305 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip))
1306 xfs_update_stable_writes(ip);
1307
1308 /*
1309 * If there is no attribute fork no ACL can exist on this inode,
1310 * and it can't have any file capabilities attached to it either.
1311 */
1312 if (!xfs_inode_has_attr_fork(ip)) {
1313 inode_has_no_xattr(inode);
1314 cache_no_acl(inode);
1315 }
1316}
1317
1318void
1319xfs_setup_iops(
1320 struct xfs_inode *ip)
1321{
1322 struct inode *inode = &ip->i_vnode;
1323
1324 switch (inode->i_mode & S_IFMT) {
1325 case S_IFREG:
1326 inode->i_op = &xfs_inode_operations;
1327 inode->i_fop = &xfs_file_operations;
1328 if (IS_DAX(inode))
1329 inode->i_mapping->a_ops = &xfs_dax_aops;
1330 else
1331 inode->i_mapping->a_ops = &xfs_address_space_operations;
1332 break;
1333 case S_IFDIR:
1334 if (xfs_has_asciici(XFS_M(inode->i_sb)))
1335 inode->i_op = &xfs_dir_ci_inode_operations;
1336 else
1337 inode->i_op = &xfs_dir_inode_operations;
1338 inode->i_fop = &xfs_dir_file_operations;
1339 break;
1340 case S_IFLNK:
1341 inode->i_op = &xfs_symlink_inode_operations;
1342 break;
1343 default:
1344 inode->i_op = &xfs_inode_operations;
1345 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1346 break;
1347 }
1348}
1/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_shared.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_mount.h"
25#include "xfs_da_format.h"
26#include "xfs_inode.h"
27#include "xfs_bmap.h"
28#include "xfs_bmap_util.h"
29#include "xfs_acl.h"
30#include "xfs_quota.h"
31#include "xfs_error.h"
32#include "xfs_attr.h"
33#include "xfs_trans.h"
34#include "xfs_trace.h"
35#include "xfs_icache.h"
36#include "xfs_symlink.h"
37#include "xfs_da_btree.h"
38#include "xfs_dir2.h"
39#include "xfs_trans_space.h"
40#include "xfs_pnfs.h"
41
42#include <linux/capability.h>
43#include <linux/xattr.h>
44#include <linux/posix_acl.h>
45#include <linux/security.h>
46#include <linux/fiemap.h>
47#include <linux/slab.h>
48
49/*
50 * Directories have different lock order w.r.t. mmap_sem compared to regular
51 * files. This is due to readdir potentially triggering page faults on a user
52 * buffer inside filldir(), and this happens with the ilock on the directory
53 * held. For regular files, the lock order is the other way around - the
54 * mmap_sem is taken during the page fault, and then we lock the ilock to do
55 * block mapping. Hence we need a different class for the directory ilock so
56 * that lockdep can tell them apart.
57 */
58static struct lock_class_key xfs_nondir_ilock_class;
59static struct lock_class_key xfs_dir_ilock_class;
60
61static int
62xfs_initxattrs(
63 struct inode *inode,
64 const struct xattr *xattr_array,
65 void *fs_info)
66{
67 const struct xattr *xattr;
68 struct xfs_inode *ip = XFS_I(inode);
69 int error = 0;
70
71 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
72 error = xfs_attr_set(ip, xattr->name, xattr->value,
73 xattr->value_len, ATTR_SECURE);
74 if (error < 0)
75 break;
76 }
77 return error;
78}
79
80/*
81 * Hook in SELinux. This is not quite correct yet, what we really need
82 * here (as we do for default ACLs) is a mechanism by which creation of
83 * these attrs can be journalled at inode creation time (along with the
84 * inode, of course, such that log replay can't cause these to be lost).
85 */
86
87STATIC int
88xfs_init_security(
89 struct inode *inode,
90 struct inode *dir,
91 const struct qstr *qstr)
92{
93 return security_inode_init_security(inode, dir, qstr,
94 &xfs_initxattrs, NULL);
95}
96
97static void
98xfs_dentry_to_name(
99 struct xfs_name *namep,
100 struct dentry *dentry,
101 int mode)
102{
103 namep->name = dentry->d_name.name;
104 namep->len = dentry->d_name.len;
105 namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
106}
107
108STATIC void
109xfs_cleanup_inode(
110 struct inode *dir,
111 struct inode *inode,
112 struct dentry *dentry)
113{
114 struct xfs_name teardown;
115
116 /* Oh, the horror.
117 * If we can't add the ACL or we fail in
118 * xfs_init_security we must back out.
119 * ENOSPC can hit here, among other things.
120 */
121 xfs_dentry_to_name(&teardown, dentry, 0);
122
123 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
124}
125
126STATIC int
127xfs_generic_create(
128 struct inode *dir,
129 struct dentry *dentry,
130 umode_t mode,
131 dev_t rdev,
132 bool tmpfile) /* unnamed file */
133{
134 struct inode *inode;
135 struct xfs_inode *ip = NULL;
136 struct posix_acl *default_acl, *acl;
137 struct xfs_name name;
138 int error;
139
140 /*
141 * Irix uses Missed'em'V split, but doesn't want to see
142 * the upper 5 bits of (14bit) major.
143 */
144 if (S_ISCHR(mode) || S_ISBLK(mode)) {
145 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
146 return -EINVAL;
147 rdev = sysv_encode_dev(rdev);
148 } else {
149 rdev = 0;
150 }
151
152 error = posix_acl_create(dir, &mode, &default_acl, &acl);
153 if (error)
154 return error;
155
156 if (!tmpfile) {
157 xfs_dentry_to_name(&name, dentry, mode);
158 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
159 } else {
160 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
161 }
162 if (unlikely(error))
163 goto out_free_acl;
164
165 inode = VFS_I(ip);
166
167 error = xfs_init_security(inode, dir, &dentry->d_name);
168 if (unlikely(error))
169 goto out_cleanup_inode;
170
171#ifdef CONFIG_XFS_POSIX_ACL
172 if (default_acl) {
173 error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
174 if (error)
175 goto out_cleanup_inode;
176 }
177 if (acl) {
178 error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
179 if (error)
180 goto out_cleanup_inode;
181 }
182#endif
183
184 if (tmpfile)
185 d_tmpfile(dentry, inode);
186 else
187 d_instantiate(dentry, inode);
188
189 xfs_finish_inode_setup(ip);
190
191 out_free_acl:
192 if (default_acl)
193 posix_acl_release(default_acl);
194 if (acl)
195 posix_acl_release(acl);
196 return error;
197
198 out_cleanup_inode:
199 xfs_finish_inode_setup(ip);
200 if (!tmpfile)
201 xfs_cleanup_inode(dir, inode, dentry);
202 iput(inode);
203 goto out_free_acl;
204}
205
206STATIC int
207xfs_vn_mknod(
208 struct inode *dir,
209 struct dentry *dentry,
210 umode_t mode,
211 dev_t rdev)
212{
213 return xfs_generic_create(dir, dentry, mode, rdev, false);
214}
215
216STATIC int
217xfs_vn_create(
218 struct inode *dir,
219 struct dentry *dentry,
220 umode_t mode,
221 bool flags)
222{
223 return xfs_vn_mknod(dir, dentry, mode, 0);
224}
225
226STATIC int
227xfs_vn_mkdir(
228 struct inode *dir,
229 struct dentry *dentry,
230 umode_t mode)
231{
232 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
233}
234
235STATIC struct dentry *
236xfs_vn_lookup(
237 struct inode *dir,
238 struct dentry *dentry,
239 unsigned int flags)
240{
241 struct xfs_inode *cip;
242 struct xfs_name name;
243 int error;
244
245 if (dentry->d_name.len >= MAXNAMELEN)
246 return ERR_PTR(-ENAMETOOLONG);
247
248 xfs_dentry_to_name(&name, dentry, 0);
249 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
250 if (unlikely(error)) {
251 if (unlikely(error != -ENOENT))
252 return ERR_PTR(error);
253 d_add(dentry, NULL);
254 return NULL;
255 }
256
257 return d_splice_alias(VFS_I(cip), dentry);
258}
259
260STATIC struct dentry *
261xfs_vn_ci_lookup(
262 struct inode *dir,
263 struct dentry *dentry,
264 unsigned int flags)
265{
266 struct xfs_inode *ip;
267 struct xfs_name xname;
268 struct xfs_name ci_name;
269 struct qstr dname;
270 int error;
271
272 if (dentry->d_name.len >= MAXNAMELEN)
273 return ERR_PTR(-ENAMETOOLONG);
274
275 xfs_dentry_to_name(&xname, dentry, 0);
276 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
277 if (unlikely(error)) {
278 if (unlikely(error != -ENOENT))
279 return ERR_PTR(error);
280 /*
281 * call d_add(dentry, NULL) here when d_drop_negative_children
282 * is called in xfs_vn_mknod (ie. allow negative dentries
283 * with CI filesystems).
284 */
285 return NULL;
286 }
287
288 /* if exact match, just splice and exit */
289 if (!ci_name.name)
290 return d_splice_alias(VFS_I(ip), dentry);
291
292 /* else case-insensitive match... */
293 dname.name = ci_name.name;
294 dname.len = ci_name.len;
295 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
296 kmem_free(ci_name.name);
297 return dentry;
298}
299
300STATIC int
301xfs_vn_link(
302 struct dentry *old_dentry,
303 struct inode *dir,
304 struct dentry *dentry)
305{
306 struct inode *inode = d_inode(old_dentry);
307 struct xfs_name name;
308 int error;
309
310 xfs_dentry_to_name(&name, dentry, inode->i_mode);
311
312 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
313 if (unlikely(error))
314 return error;
315
316 ihold(inode);
317 d_instantiate(dentry, inode);
318 return 0;
319}
320
321STATIC int
322xfs_vn_unlink(
323 struct inode *dir,
324 struct dentry *dentry)
325{
326 struct xfs_name name;
327 int error;
328
329 xfs_dentry_to_name(&name, dentry, 0);
330
331 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
332 if (error)
333 return error;
334
335 /*
336 * With unlink, the VFS makes the dentry "negative": no inode,
337 * but still hashed. This is incompatible with case-insensitive
338 * mode, so invalidate (unhash) the dentry in CI-mode.
339 */
340 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
341 d_invalidate(dentry);
342 return 0;
343}
344
345STATIC int
346xfs_vn_symlink(
347 struct inode *dir,
348 struct dentry *dentry,
349 const char *symname)
350{
351 struct inode *inode;
352 struct xfs_inode *cip = NULL;
353 struct xfs_name name;
354 int error;
355 umode_t mode;
356
357 mode = S_IFLNK |
358 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
359 xfs_dentry_to_name(&name, dentry, mode);
360
361 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
362 if (unlikely(error))
363 goto out;
364
365 inode = VFS_I(cip);
366
367 error = xfs_init_security(inode, dir, &dentry->d_name);
368 if (unlikely(error))
369 goto out_cleanup_inode;
370
371 d_instantiate(dentry, inode);
372 xfs_finish_inode_setup(cip);
373 return 0;
374
375 out_cleanup_inode:
376 xfs_finish_inode_setup(cip);
377 xfs_cleanup_inode(dir, inode, dentry);
378 iput(inode);
379 out:
380 return error;
381}
382
383STATIC int
384xfs_vn_rename(
385 struct inode *odir,
386 struct dentry *odentry,
387 struct inode *ndir,
388 struct dentry *ndentry,
389 unsigned int flags)
390{
391 struct inode *new_inode = d_inode(ndentry);
392 int omode = 0;
393 struct xfs_name oname;
394 struct xfs_name nname;
395
396 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
397 return -EINVAL;
398
399 /* if we are exchanging files, we need to set i_mode of both files */
400 if (flags & RENAME_EXCHANGE)
401 omode = d_inode(ndentry)->i_mode;
402
403 xfs_dentry_to_name(&oname, odentry, omode);
404 xfs_dentry_to_name(&nname, ndentry, d_inode(odentry)->i_mode);
405
406 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
407 XFS_I(ndir), &nname,
408 new_inode ? XFS_I(new_inode) : NULL, flags);
409}
410
411/*
412 * careful here - this function can get called recursively, so
413 * we need to be very careful about how much stack we use.
414 * uio is kmalloced for this reason...
415 */
416STATIC const char *
417xfs_vn_get_link(
418 struct dentry *dentry,
419 struct inode *inode,
420 struct delayed_call *done)
421{
422 char *link;
423 int error = -ENOMEM;
424
425 if (!dentry)
426 return ERR_PTR(-ECHILD);
427
428 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
429 if (!link)
430 goto out_err;
431
432 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
433 if (unlikely(error))
434 goto out_kfree;
435
436 set_delayed_call(done, kfree_link, link);
437 return link;
438
439 out_kfree:
440 kfree(link);
441 out_err:
442 return ERR_PTR(error);
443}
444
445STATIC int
446xfs_vn_getattr(
447 struct vfsmount *mnt,
448 struct dentry *dentry,
449 struct kstat *stat)
450{
451 struct inode *inode = d_inode(dentry);
452 struct xfs_inode *ip = XFS_I(inode);
453 struct xfs_mount *mp = ip->i_mount;
454
455 trace_xfs_getattr(ip);
456
457 if (XFS_FORCED_SHUTDOWN(mp))
458 return -EIO;
459
460 stat->size = XFS_ISIZE(ip);
461 stat->dev = inode->i_sb->s_dev;
462 stat->mode = inode->i_mode;
463 stat->nlink = inode->i_nlink;
464 stat->uid = inode->i_uid;
465 stat->gid = inode->i_gid;
466 stat->ino = ip->i_ino;
467 stat->atime = inode->i_atime;
468 stat->mtime = inode->i_mtime;
469 stat->ctime = inode->i_ctime;
470 stat->blocks =
471 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
472
473
474 switch (inode->i_mode & S_IFMT) {
475 case S_IFBLK:
476 case S_IFCHR:
477 stat->blksize = BLKDEV_IOSIZE;
478 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
479 sysv_minor(ip->i_df.if_u2.if_rdev));
480 break;
481 default:
482 if (XFS_IS_REALTIME_INODE(ip)) {
483 /*
484 * If the file blocks are being allocated from a
485 * realtime volume, then return the inode's realtime
486 * extent size or the realtime volume's extent size.
487 */
488 stat->blksize =
489 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
490 } else
491 stat->blksize = xfs_preferred_iosize(mp);
492 stat->rdev = 0;
493 break;
494 }
495
496 return 0;
497}
498
499static void
500xfs_setattr_mode(
501 struct xfs_inode *ip,
502 struct iattr *iattr)
503{
504 struct inode *inode = VFS_I(ip);
505 umode_t mode = iattr->ia_mode;
506
507 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
508
509 inode->i_mode &= S_IFMT;
510 inode->i_mode |= mode & ~S_IFMT;
511}
512
513void
514xfs_setattr_time(
515 struct xfs_inode *ip,
516 struct iattr *iattr)
517{
518 struct inode *inode = VFS_I(ip);
519
520 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
521
522 if (iattr->ia_valid & ATTR_ATIME)
523 inode->i_atime = iattr->ia_atime;
524 if (iattr->ia_valid & ATTR_CTIME)
525 inode->i_ctime = iattr->ia_ctime;
526 if (iattr->ia_valid & ATTR_MTIME)
527 inode->i_mtime = iattr->ia_mtime;
528}
529
530int
531xfs_setattr_nonsize(
532 struct xfs_inode *ip,
533 struct iattr *iattr,
534 int flags)
535{
536 xfs_mount_t *mp = ip->i_mount;
537 struct inode *inode = VFS_I(ip);
538 int mask = iattr->ia_valid;
539 xfs_trans_t *tp;
540 int error;
541 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
542 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
543 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
544 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
545
546 trace_xfs_setattr(ip);
547
548 /* If acls are being inherited, we already have this checked */
549 if (!(flags & XFS_ATTR_NOACL)) {
550 if (mp->m_flags & XFS_MOUNT_RDONLY)
551 return -EROFS;
552
553 if (XFS_FORCED_SHUTDOWN(mp))
554 return -EIO;
555
556 error = inode_change_ok(inode, iattr);
557 if (error)
558 return error;
559 }
560
561 ASSERT((mask & ATTR_SIZE) == 0);
562
563 /*
564 * If disk quotas is on, we make sure that the dquots do exist on disk,
565 * before we start any other transactions. Trying to do this later
566 * is messy. We don't care to take a readlock to look at the ids
567 * in inode here, because we can't hold it across the trans_reserve.
568 * If the IDs do change before we take the ilock, we're covered
569 * because the i_*dquot fields will get updated anyway.
570 */
571 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
572 uint qflags = 0;
573
574 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
575 uid = iattr->ia_uid;
576 qflags |= XFS_QMOPT_UQUOTA;
577 } else {
578 uid = inode->i_uid;
579 }
580 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
581 gid = iattr->ia_gid;
582 qflags |= XFS_QMOPT_GQUOTA;
583 } else {
584 gid = inode->i_gid;
585 }
586
587 /*
588 * We take a reference when we initialize udqp and gdqp,
589 * so it is important that we never blindly double trip on
590 * the same variable. See xfs_create() for an example.
591 */
592 ASSERT(udqp == NULL);
593 ASSERT(gdqp == NULL);
594 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
595 xfs_kgid_to_gid(gid),
596 xfs_get_projid(ip),
597 qflags, &udqp, &gdqp, NULL);
598 if (error)
599 return error;
600 }
601
602 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
603 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
604 if (error)
605 goto out_trans_cancel;
606
607 xfs_ilock(ip, XFS_ILOCK_EXCL);
608
609 /*
610 * Change file ownership. Must be the owner or privileged.
611 */
612 if (mask & (ATTR_UID|ATTR_GID)) {
613 /*
614 * These IDs could have changed since we last looked at them.
615 * But, we're assured that if the ownership did change
616 * while we didn't have the inode locked, inode's dquot(s)
617 * would have changed also.
618 */
619 iuid = inode->i_uid;
620 igid = inode->i_gid;
621 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
622 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
623
624 /*
625 * Do a quota reservation only if uid/gid is actually
626 * going to change.
627 */
628 if (XFS_IS_QUOTA_RUNNING(mp) &&
629 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
630 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
631 ASSERT(tp);
632 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
633 NULL, capable(CAP_FOWNER) ?
634 XFS_QMOPT_FORCE_RES : 0);
635 if (error) /* out of quota */
636 goto out_unlock;
637 }
638 }
639
640 xfs_trans_ijoin(tp, ip, 0);
641
642 /*
643 * Change file ownership. Must be the owner or privileged.
644 */
645 if (mask & (ATTR_UID|ATTR_GID)) {
646 /*
647 * CAP_FSETID overrides the following restrictions:
648 *
649 * The set-user-ID and set-group-ID bits of a file will be
650 * cleared upon successful return from chown()
651 */
652 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
653 !capable(CAP_FSETID))
654 inode->i_mode &= ~(S_ISUID|S_ISGID);
655
656 /*
657 * Change the ownerships and register quota modifications
658 * in the transaction.
659 */
660 if (!uid_eq(iuid, uid)) {
661 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
662 ASSERT(mask & ATTR_UID);
663 ASSERT(udqp);
664 olddquot1 = xfs_qm_vop_chown(tp, ip,
665 &ip->i_udquot, udqp);
666 }
667 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
668 inode->i_uid = uid;
669 }
670 if (!gid_eq(igid, gid)) {
671 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
672 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
673 !XFS_IS_PQUOTA_ON(mp));
674 ASSERT(mask & ATTR_GID);
675 ASSERT(gdqp);
676 olddquot2 = xfs_qm_vop_chown(tp, ip,
677 &ip->i_gdquot, gdqp);
678 }
679 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
680 inode->i_gid = gid;
681 }
682 }
683
684 if (mask & ATTR_MODE)
685 xfs_setattr_mode(ip, iattr);
686 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
687 xfs_setattr_time(ip, iattr);
688
689 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
690
691 XFS_STATS_INC(mp, xs_ig_attrchg);
692
693 if (mp->m_flags & XFS_MOUNT_WSYNC)
694 xfs_trans_set_sync(tp);
695 error = xfs_trans_commit(tp);
696
697 xfs_iunlock(ip, XFS_ILOCK_EXCL);
698
699 /*
700 * Release any dquot(s) the inode had kept before chown.
701 */
702 xfs_qm_dqrele(olddquot1);
703 xfs_qm_dqrele(olddquot2);
704 xfs_qm_dqrele(udqp);
705 xfs_qm_dqrele(gdqp);
706
707 if (error)
708 return error;
709
710 /*
711 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
712 * update. We could avoid this with linked transactions
713 * and passing down the transaction pointer all the way
714 * to attr_set. No previous user of the generic
715 * Posix ACL code seems to care about this issue either.
716 */
717 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
718 error = posix_acl_chmod(inode, inode->i_mode);
719 if (error)
720 return error;
721 }
722
723 return 0;
724
725out_unlock:
726 xfs_iunlock(ip, XFS_ILOCK_EXCL);
727out_trans_cancel:
728 xfs_trans_cancel(tp);
729 xfs_qm_dqrele(udqp);
730 xfs_qm_dqrele(gdqp);
731 return error;
732}
733
734/*
735 * Truncate file. Must have write permission and not be a directory.
736 */
737int
738xfs_setattr_size(
739 struct xfs_inode *ip,
740 struct iattr *iattr)
741{
742 struct xfs_mount *mp = ip->i_mount;
743 struct inode *inode = VFS_I(ip);
744 xfs_off_t oldsize, newsize;
745 struct xfs_trans *tp;
746 int error;
747 uint lock_flags = 0;
748 bool did_zeroing = false;
749
750 trace_xfs_setattr(ip);
751
752 if (mp->m_flags & XFS_MOUNT_RDONLY)
753 return -EROFS;
754
755 if (XFS_FORCED_SHUTDOWN(mp))
756 return -EIO;
757
758 error = inode_change_ok(inode, iattr);
759 if (error)
760 return error;
761
762 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
763 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
764 ASSERT(S_ISREG(inode->i_mode));
765 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
766 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
767
768 oldsize = inode->i_size;
769 newsize = iattr->ia_size;
770
771 /*
772 * Short circuit the truncate case for zero length files.
773 */
774 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
775 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
776 return 0;
777
778 /*
779 * Use the regular setattr path to update the timestamps.
780 */
781 iattr->ia_valid &= ~ATTR_SIZE;
782 return xfs_setattr_nonsize(ip, iattr, 0);
783 }
784
785 /*
786 * Make sure that the dquots are attached to the inode.
787 */
788 error = xfs_qm_dqattach(ip, 0);
789 if (error)
790 return error;
791
792 /*
793 * File data changes must be complete before we start the transaction to
794 * modify the inode. This needs to be done before joining the inode to
795 * the transaction because the inode cannot be unlocked once it is a
796 * part of the transaction.
797 *
798 * Start with zeroing any data block beyond EOF that we may expose on
799 * file extension.
800 */
801 if (newsize > oldsize) {
802 error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
803 if (error)
804 return error;
805 }
806
807 /*
808 * We are going to log the inode size change in this transaction so
809 * any previous writes that are beyond the on disk EOF and the new
810 * EOF that have not been written out need to be written here. If we
811 * do not write the data out, we expose ourselves to the null files
812 * problem. Note that this includes any block zeroing we did above;
813 * otherwise those blocks may not be zeroed after a crash.
814 */
815 if (newsize > ip->i_d.di_size &&
816 (oldsize != ip->i_d.di_size || did_zeroing)) {
817 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
818 ip->i_d.di_size, newsize);
819 if (error)
820 return error;
821 }
822
823 /* Now wait for all direct I/O to complete. */
824 inode_dio_wait(inode);
825
826 /*
827 * We've already locked out new page faults, so now we can safely remove
828 * pages from the page cache knowing they won't get refaulted until we
829 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
830 * complete. The truncate_setsize() call also cleans partial EOF page
831 * PTEs on extending truncates and hence ensures sub-page block size
832 * filesystems are correctly handled, too.
833 *
834 * We have to do all the page cache truncate work outside the
835 * transaction context as the "lock" order is page lock->log space
836 * reservation as defined by extent allocation in the writeback path.
837 * Hence a truncate can fail with ENOMEM from xfs_trans_reserve(), but
838 * having already truncated the in-memory version of the file (i.e. made
839 * user visible changes). There's not much we can do about this, except
840 * to hope that the caller sees ENOMEM and retries the truncate
841 * operation.
842 */
843 if (IS_DAX(inode))
844 error = dax_truncate_page(inode, newsize, xfs_get_blocks_direct);
845 else
846 error = block_truncate_page(inode->i_mapping, newsize,
847 xfs_get_blocks);
848 if (error)
849 return error;
850 truncate_setsize(inode, newsize);
851
852 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
853 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
854 if (error)
855 goto out_trans_cancel;
856
857 lock_flags |= XFS_ILOCK_EXCL;
858 xfs_ilock(ip, XFS_ILOCK_EXCL);
859 xfs_trans_ijoin(tp, ip, 0);
860
861 /*
862 * Only change the c/mtime if we are changing the size or we are
863 * explicitly asked to change it. This handles the semantic difference
864 * between truncate() and ftruncate() as implemented in the VFS.
865 *
866 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
867 * special case where we need to update the times despite not having
868 * these flags set. For all other operations the VFS set these flags
869 * explicitly if it wants a timestamp update.
870 */
871 if (newsize != oldsize &&
872 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
873 iattr->ia_ctime = iattr->ia_mtime =
874 current_fs_time(inode->i_sb);
875 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
876 }
877
878 /*
879 * The first thing we do is set the size to new_size permanently on
880 * disk. This way we don't have to worry about anyone ever being able
881 * to look at the data being freed even in the face of a crash.
882 * What we're getting around here is the case where we free a block, it
883 * is allocated to another file, it is written to, and then we crash.
884 * If the new data gets written to the file but the log buffers
885 * containing the free and reallocation don't, then we'd end up with
886 * garbage in the blocks being freed. As long as we make the new size
887 * permanent before actually freeing any blocks it doesn't matter if
888 * they get written to.
889 */
890 ip->i_d.di_size = newsize;
891 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
892
893 if (newsize <= oldsize) {
894 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
895 if (error)
896 goto out_trans_cancel;
897
898 /*
899 * Truncated "down", so we're removing references to old data
900 * here - if we delay flushing for a long time, we expose
901 * ourselves unduly to the notorious NULL files problem. So,
902 * we mark this inode and flush it when the file is closed,
903 * and do not wait the usual (long) time for writeout.
904 */
905 xfs_iflags_set(ip, XFS_ITRUNCATED);
906
907 /* A truncate down always removes post-EOF blocks. */
908 xfs_inode_clear_eofblocks_tag(ip);
909 }
910
911 if (iattr->ia_valid & ATTR_MODE)
912 xfs_setattr_mode(ip, iattr);
913 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
914 xfs_setattr_time(ip, iattr);
915
916 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
917
918 XFS_STATS_INC(mp, xs_ig_attrchg);
919
920 if (mp->m_flags & XFS_MOUNT_WSYNC)
921 xfs_trans_set_sync(tp);
922
923 error = xfs_trans_commit(tp);
924out_unlock:
925 if (lock_flags)
926 xfs_iunlock(ip, lock_flags);
927 return error;
928
929out_trans_cancel:
930 xfs_trans_cancel(tp);
931 goto out_unlock;
932}
933
934STATIC int
935xfs_vn_setattr(
936 struct dentry *dentry,
937 struct iattr *iattr)
938{
939 struct xfs_inode *ip = XFS_I(d_inode(dentry));
940 int error;
941
942 if (iattr->ia_valid & ATTR_SIZE) {
943 uint iolock = XFS_IOLOCK_EXCL;
944
945 xfs_ilock(ip, iolock);
946 error = xfs_break_layouts(d_inode(dentry), &iolock, true);
947 if (!error) {
948 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
949 iolock |= XFS_MMAPLOCK_EXCL;
950
951 error = xfs_setattr_size(ip, iattr);
952 }
953 xfs_iunlock(ip, iolock);
954 } else {
955 error = xfs_setattr_nonsize(ip, iattr, 0);
956 }
957
958 return error;
959}
960
961STATIC int
962xfs_vn_update_time(
963 struct inode *inode,
964 struct timespec *now,
965 int flags)
966{
967 struct xfs_inode *ip = XFS_I(inode);
968 struct xfs_mount *mp = ip->i_mount;
969 struct xfs_trans *tp;
970 int error;
971
972 trace_xfs_update_time(ip);
973
974 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
975 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
976 if (error) {
977 xfs_trans_cancel(tp);
978 return error;
979 }
980
981 xfs_ilock(ip, XFS_ILOCK_EXCL);
982 if (flags & S_CTIME)
983 inode->i_ctime = *now;
984 if (flags & S_MTIME)
985 inode->i_mtime = *now;
986 if (flags & S_ATIME)
987 inode->i_atime = *now;
988
989 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
990 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
991 return xfs_trans_commit(tp);
992}
993
994#define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
995
996/*
997 * Call fiemap helper to fill in user data.
998 * Returns positive errors to xfs_getbmap.
999 */
1000STATIC int
1001xfs_fiemap_format(
1002 void **arg,
1003 struct getbmapx *bmv,
1004 int *full)
1005{
1006 int error;
1007 struct fiemap_extent_info *fieinfo = *arg;
1008 u32 fiemap_flags = 0;
1009 u64 logical, physical, length;
1010
1011 /* Do nothing for a hole */
1012 if (bmv->bmv_block == -1LL)
1013 return 0;
1014
1015 logical = BBTOB(bmv->bmv_offset);
1016 physical = BBTOB(bmv->bmv_block);
1017 length = BBTOB(bmv->bmv_length);
1018
1019 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1020 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1021 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1022 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1023 FIEMAP_EXTENT_UNKNOWN);
1024 physical = 0; /* no block yet */
1025 }
1026 if (bmv->bmv_oflags & BMV_OF_LAST)
1027 fiemap_flags |= FIEMAP_EXTENT_LAST;
1028
1029 error = fiemap_fill_next_extent(fieinfo, logical, physical,
1030 length, fiemap_flags);
1031 if (error > 0) {
1032 error = 0;
1033 *full = 1; /* user array now full */
1034 }
1035
1036 return error;
1037}
1038
1039STATIC int
1040xfs_vn_fiemap(
1041 struct inode *inode,
1042 struct fiemap_extent_info *fieinfo,
1043 u64 start,
1044 u64 length)
1045{
1046 xfs_inode_t *ip = XFS_I(inode);
1047 struct getbmapx bm;
1048 int error;
1049
1050 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1051 if (error)
1052 return error;
1053
1054 /* Set up bmap header for xfs internal routine */
1055 bm.bmv_offset = BTOBBT(start);
1056 /* Special case for whole file */
1057 if (length == FIEMAP_MAX_OFFSET)
1058 bm.bmv_length = -1LL;
1059 else
1060 bm.bmv_length = BTOBB(start + length) - bm.bmv_offset;
1061
1062 /* We add one because in getbmap world count includes the header */
1063 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1064 fieinfo->fi_extents_max + 1;
1065 bm.bmv_count = min_t(__s32, bm.bmv_count,
1066 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1067 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1068 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1069 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1070 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1071 bm.bmv_iflags |= BMV_IF_DELALLOC;
1072
1073 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1074 if (error)
1075 return error;
1076
1077 return 0;
1078}
1079
1080STATIC int
1081xfs_vn_tmpfile(
1082 struct inode *dir,
1083 struct dentry *dentry,
1084 umode_t mode)
1085{
1086 return xfs_generic_create(dir, dentry, mode, 0, true);
1087}
1088
1089static const struct inode_operations xfs_inode_operations = {
1090 .get_acl = xfs_get_acl,
1091 .set_acl = xfs_set_acl,
1092 .getattr = xfs_vn_getattr,
1093 .setattr = xfs_vn_setattr,
1094 .setxattr = generic_setxattr,
1095 .getxattr = generic_getxattr,
1096 .removexattr = generic_removexattr,
1097 .listxattr = xfs_vn_listxattr,
1098 .fiemap = xfs_vn_fiemap,
1099 .update_time = xfs_vn_update_time,
1100};
1101
1102static const struct inode_operations xfs_dir_inode_operations = {
1103 .create = xfs_vn_create,
1104 .lookup = xfs_vn_lookup,
1105 .link = xfs_vn_link,
1106 .unlink = xfs_vn_unlink,
1107 .symlink = xfs_vn_symlink,
1108 .mkdir = xfs_vn_mkdir,
1109 /*
1110 * Yes, XFS uses the same method for rmdir and unlink.
1111 *
1112 * There are some subtile differences deeper in the code,
1113 * but we use S_ISDIR to check for those.
1114 */
1115 .rmdir = xfs_vn_unlink,
1116 .mknod = xfs_vn_mknod,
1117 .rename2 = xfs_vn_rename,
1118 .get_acl = xfs_get_acl,
1119 .set_acl = xfs_set_acl,
1120 .getattr = xfs_vn_getattr,
1121 .setattr = xfs_vn_setattr,
1122 .setxattr = generic_setxattr,
1123 .getxattr = generic_getxattr,
1124 .removexattr = generic_removexattr,
1125 .listxattr = xfs_vn_listxattr,
1126 .update_time = xfs_vn_update_time,
1127 .tmpfile = xfs_vn_tmpfile,
1128};
1129
1130static const struct inode_operations xfs_dir_ci_inode_operations = {
1131 .create = xfs_vn_create,
1132 .lookup = xfs_vn_ci_lookup,
1133 .link = xfs_vn_link,
1134 .unlink = xfs_vn_unlink,
1135 .symlink = xfs_vn_symlink,
1136 .mkdir = xfs_vn_mkdir,
1137 /*
1138 * Yes, XFS uses the same method for rmdir and unlink.
1139 *
1140 * There are some subtile differences deeper in the code,
1141 * but we use S_ISDIR to check for those.
1142 */
1143 .rmdir = xfs_vn_unlink,
1144 .mknod = xfs_vn_mknod,
1145 .rename2 = xfs_vn_rename,
1146 .get_acl = xfs_get_acl,
1147 .set_acl = xfs_set_acl,
1148 .getattr = xfs_vn_getattr,
1149 .setattr = xfs_vn_setattr,
1150 .setxattr = generic_setxattr,
1151 .getxattr = generic_getxattr,
1152 .removexattr = generic_removexattr,
1153 .listxattr = xfs_vn_listxattr,
1154 .update_time = xfs_vn_update_time,
1155 .tmpfile = xfs_vn_tmpfile,
1156};
1157
1158static const struct inode_operations xfs_symlink_inode_operations = {
1159 .readlink = generic_readlink,
1160 .get_link = xfs_vn_get_link,
1161 .getattr = xfs_vn_getattr,
1162 .setattr = xfs_vn_setattr,
1163 .setxattr = generic_setxattr,
1164 .getxattr = generic_getxattr,
1165 .removexattr = generic_removexattr,
1166 .listxattr = xfs_vn_listxattr,
1167 .update_time = xfs_vn_update_time,
1168};
1169
1170STATIC void
1171xfs_diflags_to_iflags(
1172 struct inode *inode,
1173 struct xfs_inode *ip)
1174{
1175 uint16_t flags = ip->i_d.di_flags;
1176
1177 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1178 S_NOATIME | S_DAX);
1179
1180 if (flags & XFS_DIFLAG_IMMUTABLE)
1181 inode->i_flags |= S_IMMUTABLE;
1182 if (flags & XFS_DIFLAG_APPEND)
1183 inode->i_flags |= S_APPEND;
1184 if (flags & XFS_DIFLAG_SYNC)
1185 inode->i_flags |= S_SYNC;
1186 if (flags & XFS_DIFLAG_NOATIME)
1187 inode->i_flags |= S_NOATIME;
1188 if (S_ISREG(inode->i_mode) &&
1189 ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
1190 (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
1191 ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1192 inode->i_flags |= S_DAX;
1193}
1194
1195/*
1196 * Initialize the Linux inode and set up the operation vectors.
1197 *
1198 * When reading existing inodes from disk this is called directly from xfs_iget,
1199 * when creating a new inode it is called from xfs_ialloc after setting up the
1200 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1201 * it up to the caller to deal with unlocking the inode appropriately.
1202 */
1203void
1204xfs_setup_inode(
1205 struct xfs_inode *ip)
1206{
1207 struct inode *inode = &ip->i_vnode;
1208 gfp_t gfp_mask;
1209
1210 inode->i_ino = ip->i_ino;
1211 inode->i_state = I_NEW;
1212
1213 inode_sb_list_add(inode);
1214 /* make the inode look hashed for the writeback code */
1215 hlist_add_fake(&inode->i_hash);
1216
1217 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1218 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1219
1220 switch (inode->i_mode & S_IFMT) {
1221 case S_IFBLK:
1222 case S_IFCHR:
1223 inode->i_rdev =
1224 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1225 sysv_minor(ip->i_df.if_u2.if_rdev));
1226 break;
1227 default:
1228 inode->i_rdev = 0;
1229 break;
1230 }
1231
1232 i_size_write(inode, ip->i_d.di_size);
1233 xfs_diflags_to_iflags(inode, ip);
1234
1235 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1236 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1237 switch (inode->i_mode & S_IFMT) {
1238 case S_IFREG:
1239 inode->i_op = &xfs_inode_operations;
1240 inode->i_fop = &xfs_file_operations;
1241 inode->i_mapping->a_ops = &xfs_address_space_operations;
1242 break;
1243 case S_IFDIR:
1244 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1245 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1246 inode->i_op = &xfs_dir_ci_inode_operations;
1247 else
1248 inode->i_op = &xfs_dir_inode_operations;
1249 inode->i_fop = &xfs_dir_file_operations;
1250 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1251 break;
1252 case S_IFLNK:
1253 inode->i_op = &xfs_symlink_inode_operations;
1254 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1255 inode->i_mapping->a_ops = &xfs_address_space_operations;
1256 break;
1257 default:
1258 inode->i_op = &xfs_inode_operations;
1259 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1260 break;
1261 }
1262
1263 /*
1264 * Ensure all page cache allocations are done from GFP_NOFS context to
1265 * prevent direct reclaim recursion back into the filesystem and blowing
1266 * stacks or deadlocking.
1267 */
1268 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1269 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1270
1271 /*
1272 * If there is no attribute fork no ACL can exist on this inode,
1273 * and it can't have any file capabilities attached to it either.
1274 */
1275 if (!XFS_IFORK_Q(ip)) {
1276 inode_has_no_xattr(inode);
1277 cache_no_acl(inode);
1278 }
1279}