Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 LT |
6 | #include "xfs.h" |
7 | #include "xfs_fs.h" | |
70a9883c | 8 | #include "xfs_shared.h" |
239880ef DC |
9 | #include "xfs_format.h" |
10 | #include "xfs_log_format.h" | |
11 | #include "xfs_trans_resv.h" | |
1da177e4 | 12 | #include "xfs_mount.h" |
1da177e4 | 13 | #include "xfs_inode.h" |
239880ef | 14 | #include "xfs_acl.h" |
239880ef | 15 | #include "xfs_quota.h" |
fd920008 AH |
16 | #include "xfs_da_format.h" |
17 | #include "xfs_da_btree.h" | |
1da177e4 | 18 | #include "xfs_attr.h" |
239880ef | 19 | #include "xfs_trans.h" |
0b1b213f | 20 | #include "xfs_trace.h" |
27b52867 | 21 | #include "xfs_icache.h" |
c24b5dfa | 22 | #include "xfs_symlink.h" |
1b767ee3 | 23 | #include "xfs_dir2.h" |
68a9f5e7 | 24 | #include "xfs_iomap.h" |
a5155b87 | 25 | #include "xfs_error.h" |
9fefd5db | 26 | #include "xfs_ioctl.h" |
efc2efeb | 27 | #include "xfs_xattr.h" |
1da177e4 | 28 | |
ef14f0c1 | 29 | #include <linux/posix_acl.h> |
446ada4a | 30 | #include <linux/security.h> |
c3b1b131 | 31 | #include <linux/iversion.h> |
10c5db28 | 32 | #include <linux/fiemap.h> |
1da177e4 | 33 | |
93a8614e | 34 | /* |
c1e8d7c6 | 35 | * Directories have different lock order w.r.t. mmap_lock compared to regular |
93a8614e DC |
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 | |
c1e8d7c6 | 39 | * mmap_lock is taken during the page fault, and then we lock the ilock to do |
93a8614e DC |
40 | * block mapping. Hence we need a different class for the directory ilock so |
41 | * that lockdep can tell them apart. | |
42 | */ | |
43 | static struct lock_class_key xfs_nondir_ilock_class; | |
44 | static struct lock_class_key xfs_dir_ilock_class; | |
45 | ||
8d2a5e6e DC |
46 | static int |
47 | xfs_initxattrs( | |
48 | struct inode *inode, | |
49 | const struct xattr *xattr_array, | |
50 | void *fs_info) | |
9d8f13ba | 51 | { |
8d2a5e6e DC |
52 | const struct xattr *xattr; |
53 | struct xfs_inode *ip = XFS_I(inode); | |
54 | int error = 0; | |
9d8f13ba MZ |
55 | |
56 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
a2544622 CH |
57 | struct xfs_da_args args = { |
58 | .dp = ip, | |
d5f0f49a | 59 | .attr_filter = XFS_ATTR_SECURE, |
a2544622 CH |
60 | .name = xattr->name, |
61 | .namelen = strlen(xattr->name), | |
62 | .value = xattr->value, | |
63 | .valuelen = xattr->value_len, | |
64 | }; | |
efc2efeb | 65 | error = xfs_attr_change(&args); |
9d8f13ba MZ |
66 | if (error < 0) |
67 | break; | |
68 | } | |
69 | return error; | |
70 | } | |
71 | ||
446ada4a NS |
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 | */ | |
70b589a3 ES |
78 | int |
79 | xfs_inode_init_security( | |
af048193 | 80 | struct inode *inode, |
2a7dba39 EP |
81 | struct inode *dir, |
82 | const struct qstr *qstr) | |
446ada4a | 83 | { |
2451337d | 84 | return security_inode_init_security(inode, dir, qstr, |
a5a14de2 | 85 | &xfs_initxattrs, NULL); |
446ada4a NS |
86 | } |
87 | ||
556b8b16 BN |
88 | static void |
89 | xfs_dentry_to_name( | |
fab8eef8 AG |
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 | ||
98 | static int | |
99 | xfs_dentry_mode_to_name( | |
556b8b16 | 100 | struct xfs_name *namep, |
0cb97766 DC |
101 | struct dentry *dentry, |
102 | int mode) | |
556b8b16 BN |
103 | { |
104 | namep->name = dentry->d_name.name; | |
105 | namep->len = dentry->d_name.len; | |
1fc4d33f | 106 | namep->type = xfs_mode_to_ftype(mode); |
fab8eef8 AG |
107 | |
108 | if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN)) | |
109 | return -EFSCORRUPTED; | |
110 | ||
111 | return 0; | |
556b8b16 BN |
112 | } |
113 | ||
7989cb8e | 114 | STATIC void |
416c6d5b | 115 | xfs_cleanup_inode( |
739bfb2a | 116 | struct inode *dir, |
af048193 | 117 | struct inode *inode, |
8f112e3b | 118 | struct dentry *dentry) |
3a69c7dc | 119 | { |
556b8b16 | 120 | struct xfs_name teardown; |
3a69c7dc YL |
121 | |
122 | /* Oh, the horror. | |
220b5284 | 123 | * If we can't add the ACL or we fail in |
70b589a3 | 124 | * xfs_inode_init_security we must back out. |
3a69c7dc YL |
125 | * ENOSPC can hit here, among other things. |
126 | */ | |
fab8eef8 | 127 | xfs_dentry_to_name(&teardown, dentry); |
3a69c7dc | 128 | |
8f112e3b | 129 | xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); |
3a69c7dc YL |
130 | } |
131 | ||
e6a688c3 DC |
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 | */ | |
145 | static inline bool | |
146 | xfs_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 | ||
1da177e4 | 163 | STATIC int |
d540e43b | 164 | xfs_generic_create( |
f2d40141 CB |
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 */ | |
1da177e4 | 171 | { |
db0bb7ba | 172 | struct inode *inode; |
979ebab1 | 173 | struct xfs_inode *ip = NULL; |
2401dc29 | 174 | struct posix_acl *default_acl, *acl; |
556b8b16 | 175 | struct xfs_name name; |
1da177e4 LT |
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 | */ | |
517b5e8c CH |
182 | if (S_ISCHR(mode) || S_ISBLK(mode)) { |
183 | if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff)) | |
184 | return -EINVAL; | |
517b5e8c CH |
185 | } else { |
186 | rdev = 0; | |
187 | } | |
1da177e4 | 188 | |
2401dc29 CH |
189 | error = posix_acl_create(dir, &mode, &default_acl, &acl); |
190 | if (error) | |
191 | return error; | |
1da177e4 | 192 | |
fab8eef8 AG |
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 | ||
d540e43b | 198 | if (!tmpfile) { |
f2d40141 | 199 | error = xfs_create(idmap, XFS_I(dir), &name, mode, rdev, |
e6a688c3 DC |
200 | xfs_create_need_xattr(dir, default_acl, acl), |
201 | &ip); | |
d540e43b | 202 | } else { |
f2d40141 | 203 | error = xfs_create_tmpfile(idmap, XFS_I(dir), mode, &ip); |
d540e43b | 204 | } |
db0bb7ba CH |
205 | if (unlikely(error)) |
206 | goto out_free_acl; | |
446ada4a | 207 | |
01651646 | 208 | inode = VFS_I(ip); |
979ebab1 | 209 | |
70b589a3 | 210 | error = xfs_inode_init_security(inode, dir, &dentry->d_name); |
db0bb7ba CH |
211 | if (unlikely(error)) |
212 | goto out_cleanup_inode; | |
213 | ||
214 | if (default_acl) { | |
8ba35875 | 215 | error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); |
2401dc29 | 216 | if (error) |
db0bb7ba | 217 | goto out_cleanup_inode; |
1da177e4 | 218 | } |
2401dc29 | 219 | if (acl) { |
8ba35875 | 220 | error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); |
2401dc29 CH |
221 | if (error) |
222 | goto out_cleanup_inode; | |
223 | } | |
1da177e4 | 224 | |
2b3d1d41 CH |
225 | xfs_setup_iops(ip); |
226 | ||
c4a6bf7f DW |
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); | |
863f144f | 237 | d_tmpfile(tmpfile, inode); |
c4a6bf7f | 238 | } else |
d540e43b BF |
239 | d_instantiate(dentry, inode); |
240 | ||
58c90473 DC |
241 | xfs_finish_inode_setup(ip); |
242 | ||
2401dc29 | 243 | out_free_acl: |
88269b88 KX |
244 | posix_acl_release(default_acl); |
245 | posix_acl_release(acl); | |
2451337d | 246 | return error; |
db0bb7ba CH |
247 | |
248 | out_cleanup_inode: | |
58c90473 | 249 | xfs_finish_inode_setup(ip); |
d540e43b BF |
250 | if (!tmpfile) |
251 | xfs_cleanup_inode(dir, inode, dentry); | |
44a8736b | 252 | xfs_irele(ip); |
2401dc29 | 253 | goto out_free_acl; |
1da177e4 LT |
254 | } |
255 | ||
d540e43b BF |
256 | STATIC int |
257 | xfs_vn_mknod( | |
5ebb29be | 258 | struct mnt_idmap *idmap, |
549c7297 CB |
259 | struct inode *dir, |
260 | struct dentry *dentry, | |
261 | umode_t mode, | |
262 | dev_t rdev) | |
d540e43b | 263 | { |
f2d40141 | 264 | return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL); |
d540e43b BF |
265 | } |
266 | ||
1da177e4 | 267 | STATIC int |
416c6d5b | 268 | xfs_vn_create( |
6c960e68 | 269 | struct mnt_idmap *idmap, |
549c7297 CB |
270 | struct inode *dir, |
271 | struct dentry *dentry, | |
272 | umode_t mode, | |
273 | bool flags) | |
1da177e4 | 274 | { |
f2d40141 | 275 | return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL); |
1da177e4 LT |
276 | } |
277 | ||
278 | STATIC int | |
416c6d5b | 279 | xfs_vn_mkdir( |
c54bd91e | 280 | struct mnt_idmap *idmap, |
549c7297 CB |
281 | struct inode *dir, |
282 | struct dentry *dentry, | |
283 | umode_t mode) | |
1da177e4 | 284 | { |
f2d40141 | 285 | return xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL); |
1da177e4 LT |
286 | } |
287 | ||
288 | STATIC struct dentry * | |
416c6d5b | 289 | xfs_vn_lookup( |
1da177e4 LT |
290 | struct inode *dir, |
291 | struct dentry *dentry, | |
00cd8dd3 | 292 | unsigned int flags) |
1da177e4 | 293 | { |
b113a6d3 | 294 | struct inode *inode; |
ef1f5e7a | 295 | struct xfs_inode *cip; |
556b8b16 | 296 | struct xfs_name name; |
1da177e4 LT |
297 | int error; |
298 | ||
299 | if (dentry->d_name.len >= MAXNAMELEN) | |
300 | return ERR_PTR(-ENAMETOOLONG); | |
301 | ||
fab8eef8 | 302 | xfs_dentry_to_name(&name, dentry); |
384f3ced | 303 | error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); |
b113a6d3 AV |
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); | |
1da177e4 LT |
311 | } |
312 | ||
384f3ced BN |
313 | STATIC struct dentry * |
314 | xfs_vn_ci_lookup( | |
315 | struct inode *dir, | |
316 | struct dentry *dentry, | |
00cd8dd3 | 317 | unsigned int flags) |
384f3ced BN |
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 | ||
fab8eef8 | 328 | xfs_dentry_to_name(&xname, dentry); |
384f3ced BN |
329 | error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); |
330 | if (unlikely(error)) { | |
2451337d DC |
331 | if (unlikely(error != -ENOENT)) |
332 | return ERR_PTR(error); | |
866d5dc9 BN |
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 | */ | |
384f3ced BN |
338 | return NULL; |
339 | } | |
340 | ||
341 | /* if exact match, just splice and exit */ | |
342 | if (!ci_name.name) | |
01651646 | 343 | return d_splice_alias(VFS_I(ip), dentry); |
384f3ced BN |
344 | |
345 | /* else case-insensitive match... */ | |
346 | dname.name = ci_name.name; | |
347 | dname.len = ci_name.len; | |
e45b590b | 348 | dentry = d_add_ci(dentry, VFS_I(ip), &dname); |
384f3ced BN |
349 | kmem_free(ci_name.name); |
350 | return dentry; | |
351 | } | |
352 | ||
1da177e4 | 353 | STATIC int |
416c6d5b | 354 | xfs_vn_link( |
1da177e4 LT |
355 | struct dentry *old_dentry, |
356 | struct inode *dir, | |
357 | struct dentry *dentry) | |
358 | { | |
2b0143b5 | 359 | struct inode *inode = d_inode(old_dentry); |
556b8b16 | 360 | struct xfs_name name; |
1da177e4 LT |
361 | int error; |
362 | ||
fab8eef8 AG |
363 | error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode); |
364 | if (unlikely(error)) | |
365 | return error; | |
1da177e4 | 366 | |
556b8b16 | 367 | error = xfs_link(XFS_I(dir), XFS_I(inode), &name); |
d9424b3c | 368 | if (unlikely(error)) |
2451337d | 369 | return error; |
a3da7896 | 370 | |
7de9c6ee | 371 | ihold(inode); |
a3da7896 CH |
372 | d_instantiate(dentry, inode); |
373 | return 0; | |
1da177e4 LT |
374 | } |
375 | ||
376 | STATIC int | |
416c6d5b | 377 | xfs_vn_unlink( |
1da177e4 LT |
378 | struct inode *dir, |
379 | struct dentry *dentry) | |
380 | { | |
556b8b16 | 381 | struct xfs_name name; |
1da177e4 LT |
382 | int error; |
383 | ||
fab8eef8 | 384 | xfs_dentry_to_name(&name, dentry); |
1da177e4 | 385 | |
2b0143b5 | 386 | error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry))); |
e5700704 CH |
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 | */ | |
38c26bfd | 395 | if (xfs_has_asciici(XFS_M(dir->i_sb))) |
e5700704 CH |
396 | d_invalidate(dentry); |
397 | return 0; | |
1da177e4 LT |
398 | } |
399 | ||
400 | STATIC int | |
416c6d5b | 401 | xfs_vn_symlink( |
7a77db95 | 402 | struct mnt_idmap *idmap, |
549c7297 CB |
403 | struct inode *dir, |
404 | struct dentry *dentry, | |
405 | const char *symname) | |
1da177e4 | 406 | { |
3937be5b CH |
407 | struct inode *inode; |
408 | struct xfs_inode *cip = NULL; | |
556b8b16 | 409 | struct xfs_name name; |
1da177e4 | 410 | int error; |
576b1d67 | 411 | umode_t mode; |
1da177e4 | 412 | |
3e5daf05 | 413 | mode = S_IFLNK | |
ce3b0f8d | 414 | (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); |
fab8eef8 AG |
415 | error = xfs_dentry_mode_to_name(&name, dentry, mode); |
416 | if (unlikely(error)) | |
417 | goto out; | |
1da177e4 | 418 | |
f2d40141 | 419 | error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip); |
3937be5b CH |
420 | if (unlikely(error)) |
421 | goto out; | |
422 | ||
01651646 | 423 | inode = VFS_I(cip); |
3937be5b | 424 | |
70b589a3 | 425 | error = xfs_inode_init_security(inode, dir, &dentry->d_name); |
3937be5b CH |
426 | if (unlikely(error)) |
427 | goto out_cleanup_inode; | |
428 | ||
2b3d1d41 CH |
429 | xfs_setup_iops(cip); |
430 | ||
3937be5b | 431 | d_instantiate(dentry, inode); |
58c90473 | 432 | xfs_finish_inode_setup(cip); |
3937be5b CH |
433 | return 0; |
434 | ||
435 | out_cleanup_inode: | |
58c90473 | 436 | xfs_finish_inode_setup(cip); |
8f112e3b | 437 | xfs_cleanup_inode(dir, inode, dentry); |
44a8736b | 438 | xfs_irele(cip); |
3937be5b | 439 | out: |
2451337d | 440 | return error; |
1da177e4 LT |
441 | } |
442 | ||
1da177e4 | 443 | STATIC int |
416c6d5b | 444 | xfs_vn_rename( |
e18275ae | 445 | struct mnt_idmap *idmap, |
549c7297 CB |
446 | struct inode *odir, |
447 | struct dentry *odentry, | |
448 | struct inode *ndir, | |
449 | struct dentry *ndentry, | |
450 | unsigned int flags) | |
1da177e4 | 451 | { |
2b0143b5 | 452 | struct inode *new_inode = d_inode(ndentry); |
d31a1825 | 453 | int omode = 0; |
fab8eef8 | 454 | int error; |
556b8b16 BN |
455 | struct xfs_name oname; |
456 | struct xfs_name nname; | |
1da177e4 | 457 | |
7dcf5c3e | 458 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
dbe1b5ca CM |
459 | return -EINVAL; |
460 | ||
d31a1825 CM |
461 | /* if we are exchanging files, we need to set i_mode of both files */ |
462 | if (flags & RENAME_EXCHANGE) | |
2b0143b5 | 463 | omode = d_inode(ndentry)->i_mode; |
d31a1825 | 464 | |
fab8eef8 AG |
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; | |
556b8b16 | 473 | |
f2d40141 | 474 | return xfs_rename(idmap, XFS_I(odir), &oname, |
f736d93d | 475 | XFS_I(d_inode(odentry)), XFS_I(ndir), &nname, |
d31a1825 | 476 | new_inode ? XFS_I(new_inode) : NULL, flags); |
1da177e4 LT |
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 | */ | |
680baacb | 484 | STATIC const char * |
6b255391 | 485 | xfs_vn_get_link( |
1da177e4 | 486 | struct dentry *dentry, |
6b255391 | 487 | struct inode *inode, |
fceef393 | 488 | struct delayed_call *done) |
1da177e4 | 489 | { |
1da177e4 | 490 | char *link; |
804c83c3 | 491 | int error = -ENOMEM; |
1da177e4 | 492 | |
6b255391 AV |
493 | if (!dentry) |
494 | return ERR_PTR(-ECHILD); | |
495 | ||
6eb0b8df | 496 | link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); |
804c83c3 CH |
497 | if (!link) |
498 | goto out_err; | |
1da177e4 | 499 | |
2b0143b5 | 500 | error = xfs_readlink(XFS_I(d_inode(dentry)), link); |
804c83c3 CH |
501 | if (unlikely(error)) |
502 | goto out_kfree; | |
1da177e4 | 503 | |
fceef393 AV |
504 | set_delayed_call(done, kfree_link, link); |
505 | return link; | |
804c83c3 CH |
506 | |
507 | out_kfree: | |
508 | kfree(link); | |
509 | out_err: | |
680baacb | 510 | return ERR_PTR(error); |
1da177e4 LT |
511 | } |
512 | ||
dd2d535e CH |
513 | static uint32_t |
514 | xfs_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)) | |
a7bcb147 | 524 | return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip)); |
dd2d535e CH |
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 | */ | |
0560f31a | 539 | if (xfs_has_large_iosize(mp)) { |
dd2d535e | 540 | if (mp->m_swidth) |
a7bcb147 | 541 | return XFS_FSB_TO_B(mp, mp->m_swidth); |
0560f31a | 542 | if (xfs_has_allocsize(mp)) |
5da8a07c | 543 | return 1U << mp->m_allocsize_log; |
dd2d535e CH |
544 | } |
545 | ||
546 | return PAGE_SIZE; | |
547 | } | |
548 | ||
1da177e4 | 549 | STATIC int |
416c6d5b | 550 | xfs_vn_getattr( |
b74d24f7 | 551 | struct mnt_idmap *idmap, |
a528d35e DH |
552 | const struct path *path, |
553 | struct kstat *stat, | |
554 | u32 request_mask, | |
555 | unsigned int query_flags) | |
1da177e4 | 556 | { |
a528d35e | 557 | struct inode *inode = d_inode(path->dentry); |
c43f4087 CH |
558 | struct xfs_inode *ip = XFS_I(inode); |
559 | struct xfs_mount *mp = ip->i_mount; | |
e67fe633 CB |
560 | vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode); |
561 | vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode); | |
c43f4087 | 562 | |
cca28fb8 | 563 | trace_xfs_getattr(ip); |
c43f4087 | 564 | |
75c8c50f | 565 | if (xfs_is_shutdown(mp)) |
b474c7ae | 566 | return -EIO; |
c43f4087 CH |
567 | |
568 | stat->size = XFS_ISIZE(ip); | |
569 | stat->dev = inode->i_sb->s_dev; | |
c19b3b05 | 570 | stat->mode = inode->i_mode; |
54d7b5c1 | 571 | stat->nlink = inode->i_nlink; |
42b7cc11 CB |
572 | stat->uid = vfsuid_into_kuid(vfsuid); |
573 | stat->gid = vfsgid_into_kgid(vfsgid); | |
c43f4087 | 574 | stat->ino = ip->i_ino; |
c43f4087 | 575 | stat->atime = inode->i_atime; |
f798accd CB |
576 | stat->mtime = inode->i_mtime; |
577 | stat->ctime = inode_get_ctime(inode); | |
6e73a545 | 578 | stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); |
c43f4087 | 579 | |
38c26bfd | 580 | if (xfs_has_v3inodes(mp)) { |
5f955f26 DW |
581 | if (request_mask & STATX_BTIME) { |
582 | stat->result_mask |= STATX_BTIME; | |
e98d5e88 | 583 | stat->btime = ip->i_crtime; |
5f955f26 DW |
584 | } |
585 | } | |
586 | ||
cbc06310 JL |
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 | ||
1b9598c8 LR |
592 | /* |
593 | * Note: If you add another clause to set an attribute flag, please | |
594 | * update attributes_mask below. | |
595 | */ | |
db07349d | 596 | if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) |
5f955f26 | 597 | stat->attributes |= STATX_ATTR_IMMUTABLE; |
db07349d | 598 | if (ip->i_diflags & XFS_DIFLAG_APPEND) |
5f955f26 | 599 | stat->attributes |= STATX_ATTR_APPEND; |
db07349d | 600 | if (ip->i_diflags & XFS_DIFLAG_NODUMP) |
5f955f26 | 601 | stat->attributes |= STATX_ATTR_NODUMP; |
c43f4087 | 602 | |
1b9598c8 LR |
603 | stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | |
604 | STATX_ATTR_APPEND | | |
605 | STATX_ATTR_NODUMP); | |
606 | ||
c43f4087 CH |
607 | switch (inode->i_mode & S_IFMT) { |
608 | case S_IFBLK: | |
609 | case S_IFCHR: | |
610 | stat->blksize = BLKDEV_IOSIZE; | |
66f36464 | 611 | stat->rdev = inode->i_rdev; |
c43f4087 | 612 | break; |
61a223df EB |
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; | |
c43f4087 | 623 | default: |
dd2d535e | 624 | stat->blksize = xfs_stat_blksize(ip); |
c43f4087 CH |
625 | stat->rdev = 0; |
626 | break; | |
69e23b9a | 627 | } |
c43f4087 CH |
628 | |
629 | return 0; | |
1da177e4 LT |
630 | } |
631 | ||
69bca807 JK |
632 | static int |
633 | xfs_vn_change_ok( | |
c1632a0f | 634 | struct mnt_idmap *idmap, |
f736d93d CH |
635 | struct dentry *dentry, |
636 | struct iattr *iattr) | |
69bca807 | 637 | { |
31051c85 | 638 | struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; |
69bca807 | 639 | |
2e973b2c | 640 | if (xfs_is_readonly(mp)) |
69bca807 JK |
641 | return -EROFS; |
642 | ||
75c8c50f | 643 | if (xfs_is_shutdown(mp)) |
69bca807 JK |
644 | return -EIO; |
645 | ||
c1632a0f | 646 | return setattr_prepare(idmap, dentry, iattr); |
69bca807 JK |
647 | } |
648 | ||
649 | /* | |
650 | * Set non-size attributes of an inode. | |
651 | * | |
652 | * Caution: The caller of this function is responsible for calling | |
31051c85 | 653 | * setattr_prepare() or otherwise verifying the change is fine. |
69bca807 | 654 | */ |
5d24ec4c | 655 | static int |
c4ed4243 | 656 | xfs_setattr_nonsize( |
c1632a0f | 657 | struct mnt_idmap *idmap, |
138060ba | 658 | struct dentry *dentry, |
c4ed4243 | 659 | struct xfs_inode *ip, |
5d24ec4c | 660 | struct iattr *iattr) |
c4ed4243 CH |
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; | |
dd3b015d DW |
667 | kuid_t uid = GLOBAL_ROOT_UID; |
668 | kgid_t gid = GLOBAL_ROOT_GID; | |
c4ed4243 | 669 | struct xfs_dquot *udqp = NULL, *gdqp = NULL; |
dd3b015d | 670 | struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL; |
c4ed4243 | 671 | |
c4ed4243 CH |
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)) { | |
4d7ca409 | 686 | uid = from_vfsuid(idmap, i_user_ns(inode), |
b27c82e1 | 687 | iattr->ia_vfsuid); |
c4ed4243 CH |
688 | qflags |= XFS_QMOPT_UQUOTA; |
689 | } else { | |
7aab1b28 | 690 | uid = inode->i_uid; |
c4ed4243 CH |
691 | } |
692 | if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { | |
4d7ca409 | 693 | gid = from_vfsgid(idmap, i_user_ns(inode), |
b27c82e1 | 694 | iattr->ia_vfsgid); |
c4ed4243 CH |
695 | qflags |= XFS_QMOPT_GQUOTA; |
696 | } else { | |
7aab1b28 | 697 | gid = inode->i_gid; |
c4ed4243 CH |
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); | |
ceaf603c | 707 | error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, |
7aab1b28 | 708 | qflags, &udqp, &gdqp, NULL); |
c4ed4243 CH |
709 | if (error) |
710 | return error; | |
711 | } | |
712 | ||
7317a03d | 713 | error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, |
eba0549b | 714 | has_capability_noaudit(current, CAP_FOWNER), &tp); |
c4ed4243 | 715 | if (error) |
253f4911 | 716 | goto out_dqrele; |
c4ed4243 | 717 | |
c4ed4243 | 718 | /* |
dd3b015d DW |
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. | |
c4ed4243 | 724 | */ |
35faf310 | 725 | if (XFS_IS_UQUOTA_ON(mp) && |
0dbe12f2 | 726 | i_uid_needs_update(idmap, iattr, inode)) { |
dd3b015d DW |
727 | ASSERT(udqp); |
728 | old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp); | |
729 | } | |
35faf310 | 730 | if (XFS_IS_GQUOTA_ON(mp) && |
0dbe12f2 | 731 | i_gid_needs_update(idmap, iattr, inode)) { |
dd3b015d DW |
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); | |
c4ed4243 CH |
735 | } |
736 | ||
c1632a0f | 737 | setattr_copy(idmap, inode, iattr); |
c4ed4243 CH |
738 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
739 | ||
ff6d6af2 | 740 | XFS_STATS_INC(mp, xs_ig_attrchg); |
c4ed4243 | 741 | |
0560f31a | 742 | if (xfs_has_wsync(mp)) |
c4ed4243 | 743 | xfs_trans_set_sync(tp); |
70393313 | 744 | error = xfs_trans_commit(tp); |
c4ed4243 | 745 | |
c4ed4243 CH |
746 | /* |
747 | * Release any dquot(s) the inode had kept before chown. | |
748 | */ | |
dd3b015d DW |
749 | xfs_qm_dqrele(old_udqp); |
750 | xfs_qm_dqrele(old_gdqp); | |
c4ed4243 CH |
751 | xfs_qm_dqrele(udqp); |
752 | xfs_qm_dqrele(gdqp); | |
753 | ||
754 | if (error) | |
b474c7ae | 755 | return error; |
c4ed4243 CH |
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 | */ | |
5d24ec4c | 764 | if (mask & ATTR_MODE) { |
13e83a49 | 765 | error = posix_acl_chmod(idmap, dentry, inode->i_mode); |
c4ed4243 | 766 | if (error) |
b474c7ae | 767 | return error; |
c4ed4243 CH |
768 | } |
769 | ||
770 | return 0; | |
771 | ||
253f4911 | 772 | out_dqrele: |
c4ed4243 CH |
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. | |
69bca807 JK |
780 | * |
781 | * Caution: The caller of this function is responsible for calling | |
31051c85 | 782 | * setattr_prepare() or otherwise verifying the change is fine. |
c4ed4243 | 783 | */ |
7bf7a193 | 784 | STATIC int |
c4ed4243 | 785 | xfs_setattr_size( |
c1632a0f | 786 | struct mnt_idmap *idmap, |
138060ba | 787 | struct dentry *dentry, |
c4ed4243 | 788 | struct xfs_inode *ip, |
76ca4c23 | 789 | struct iattr *iattr) |
c4ed4243 CH |
790 | { |
791 | struct xfs_mount *mp = ip->i_mount; | |
792 | struct inode *inode = VFS_I(ip); | |
673e8e59 | 793 | xfs_off_t oldsize, newsize; |
c4ed4243 CH |
794 | struct xfs_trans *tp; |
795 | int error; | |
f38996f5 | 796 | uint lock_flags = 0; |
5885ebda | 797 | bool did_zeroing = false; |
c4ed4243 | 798 | |
76ca4c23 | 799 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
e8e9ad42 | 800 | ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); |
c19b3b05 | 801 | ASSERT(S_ISREG(inode->i_mode)); |
fe60a8a0 | 802 | ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| |
88a9e03b | 803 | ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); |
c4ed4243 | 804 | |
ce7ae151 | 805 | oldsize = inode->i_size; |
673e8e59 CH |
806 | newsize = iattr->ia_size; |
807 | ||
c4ed4243 CH |
808 | /* |
809 | * Short circuit the truncate case for zero length files. | |
810 | */ | |
daf83964 | 811 | if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { |
fe60a8a0 | 812 | if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) |
76ca4c23 | 813 | return 0; |
681b1200 CH |
814 | |
815 | /* | |
816 | * Use the regular setattr path to update the timestamps. | |
817 | */ | |
681b1200 | 818 | iattr->ia_valid &= ~ATTR_SIZE; |
c1632a0f | 819 | return xfs_setattr_nonsize(idmap, dentry, ip, iattr); |
c4ed4243 CH |
820 | } |
821 | ||
822 | /* | |
823 | * Make sure that the dquots are attached to the inode. | |
824 | */ | |
c14cfcca | 825 | error = xfs_qm_dqattach(ip); |
c4ed4243 | 826 | if (error) |
76ca4c23 | 827 | return error; |
c4ed4243 | 828 | |
f0c6bcba CH |
829 | /* |
830 | * Wait for all direct I/O to complete. | |
831 | */ | |
832 | inode_dio_wait(inode); | |
833 | ||
c4ed4243 | 834 | /* |
5885ebda DC |
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 | * | |
f0c6bcba CH |
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. | |
c4ed4243 | 843 | */ |
673e8e59 | 844 | if (newsize > oldsize) { |
f5c54717 | 845 | trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); |
f1ba5faf SR |
846 | error = xfs_zero_range(ip, oldsize, newsize - oldsize, |
847 | &did_zeroing); | |
f0c6bcba | 848 | } else { |
869ae85d BF |
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; | |
f1ba5faf | 859 | error = xfs_truncate_page(ip, newsize, &did_zeroing); |
c4ed4243 | 860 | } |
c4ed4243 | 861 | |
f0c6bcba CH |
862 | if (error) |
863 | return error; | |
864 | ||
49abc3a8 | 865 | /* |
0f9160b4 DC |
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. | |
49abc3a8 | 872 | * |
0f9160b4 DC |
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. | |
253f4911 | 876 | * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but |
0f9160b4 DC |
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. | |
350976ae EG |
881 | * |
882 | * And we update in-core i_size and truncate page cache beyond newsize | |
13d2c10b CH |
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. | |
49abc3a8 | 885 | */ |
49abc3a8 | 886 | truncate_setsize(inode, newsize); |
c4ed4243 | 887 | |
350976ae EG |
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 || | |
13d2c10b | 897 | (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { |
350976ae | 898 | error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
13d2c10b | 899 | ip->i_disk_size, newsize - 1); |
350976ae EG |
900 | if (error) |
901 | return error; | |
902 | } | |
903 | ||
253f4911 | 904 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); |
c4ed4243 | 905 | if (error) |
253f4911 | 906 | return error; |
c4ed4243 | 907 | |
c4ed4243 | 908 | lock_flags |= XFS_ILOCK_EXCL; |
c4ed4243 | 909 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
ddc3415a | 910 | xfs_trans_ijoin(tp, ip, 0); |
c4ed4243 CH |
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 | */ | |
fe60a8a0 CH |
922 | if (newsize != oldsize && |
923 | !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { | |
c4ed4243 | 924 | iattr->ia_ctime = iattr->ia_mtime = |
f798accd | 925 | current_time(inode); |
fe60a8a0 | 926 | iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; |
c4ed4243 CH |
927 | } |
928 | ||
673e8e59 CH |
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 | */ | |
13d2c10b | 941 | ip->i_disk_size = newsize; |
673e8e59 CH |
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); | |
c4ed4243 | 946 | if (error) |
4906e215 | 947 | goto out_trans_cancel; |
c4ed4243 CH |
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); | |
27b52867 BF |
957 | |
958 | /* A truncate down always removes post-EOF blocks. */ | |
959 | xfs_inode_clear_eofblocks_tag(ip); | |
c4ed4243 CH |
960 | } |
961 | ||
e014f37d | 962 | ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID))); |
c1632a0f | 963 | setattr_copy(idmap, inode, iattr); |
c4ed4243 CH |
964 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
965 | ||
ff6d6af2 | 966 | XFS_STATS_INC(mp, xs_ig_attrchg); |
c4ed4243 | 967 | |
0560f31a | 968 | if (xfs_has_wsync(mp)) |
c4ed4243 CH |
969 | xfs_trans_set_sync(tp); |
970 | ||
70393313 | 971 | error = xfs_trans_commit(tp); |
c4ed4243 CH |
972 | out_unlock: |
973 | if (lock_flags) | |
974 | xfs_iunlock(ip, lock_flags); | |
975 | return error; | |
976 | ||
c4ed4243 | 977 | out_trans_cancel: |
4906e215 | 978 | xfs_trans_cancel(tp); |
c4ed4243 CH |
979 | goto out_unlock; |
980 | } | |
981 | ||
69bca807 JK |
982 | int |
983 | xfs_vn_setattr_size( | |
c1632a0f | 984 | struct mnt_idmap *idmap, |
69bca807 JK |
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 | ||
c1632a0f | 993 | error = xfs_vn_change_ok(idmap, dentry, iattr); |
69bca807 JK |
994 | if (error) |
995 | return error; | |
c1632a0f | 996 | return xfs_setattr_size(idmap, dentry, ip, iattr); |
69bca807 JK |
997 | } |
998 | ||
1da177e4 | 999 | STATIC int |
416c6d5b | 1000 | xfs_vn_setattr( |
c1632a0f | 1001 | struct mnt_idmap *idmap, |
76ca4c23 CH |
1002 | struct dentry *dentry, |
1003 | struct iattr *iattr) | |
1da177e4 | 1004 | { |
26f88363 CH |
1005 | struct inode *inode = d_inode(dentry); |
1006 | struct xfs_inode *ip = XFS_I(inode); | |
76ca4c23 CH |
1007 | int error; |
1008 | ||
1009 | if (iattr->ia_valid & ATTR_SIZE) { | |
c63a8eae | 1010 | uint iolock; |
781355c6 | 1011 | |
c63a8eae DW |
1012 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
1013 | iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; | |
781355c6 | 1014 | |
69eb5fa1 | 1015 | error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); |
c63a8eae DW |
1016 | if (error) { |
1017 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); | |
65523218 | 1018 | return error; |
c63a8eae | 1019 | } |
e8e9ad42 | 1020 | |
c1632a0f | 1021 | error = xfs_vn_setattr_size(idmap, dentry, iattr); |
65523218 | 1022 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
76ca4c23 | 1023 | } else { |
26f88363 CH |
1024 | trace_xfs_setattr(ip); |
1025 | ||
c1632a0f | 1026 | error = xfs_vn_change_ok(idmap, dentry, iattr); |
26f88363 | 1027 | if (!error) |
c1632a0f | 1028 | error = xfs_setattr_nonsize(idmap, dentry, ip, iattr); |
76ca4c23 CH |
1029 | } |
1030 | ||
2451337d | 1031 | return error; |
1da177e4 LT |
1032 | } |
1033 | ||
69ff2826 CH |
1034 | STATIC int |
1035 | xfs_vn_update_time( | |
1036 | struct inode *inode, | |
69ff2826 CH |
1037 | int flags) |
1038 | { | |
1039 | struct xfs_inode *ip = XFS_I(inode); | |
1040 | struct xfs_mount *mp = ip->i_mount; | |
c3b1b131 | 1041 | int log_flags = XFS_ILOG_TIMESTAMP; |
69ff2826 CH |
1042 | struct xfs_trans *tp; |
1043 | int error; | |
51b0f3eb | 1044 | struct timespec64 now; |
69ff2826 CH |
1045 | |
1046 | trace_xfs_update_time(ip); | |
1047 | ||
c3b1b131 CH |
1048 | if (inode->i_sb->s_flags & SB_LAZYTIME) { |
1049 | if (!((flags & S_VERSION) && | |
541d4c79 JL |
1050 | inode_maybe_inc_iversion(inode, false))) { |
1051 | generic_update_time(inode, flags); | |
1052 | return 0; | |
1053 | } | |
c3b1b131 CH |
1054 | |
1055 | /* Capture the iversion update that just occurred */ | |
1056 | log_flags |= XFS_ILOG_CORE; | |
1057 | } | |
1058 | ||
253f4911 CH |
1059 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
1060 | if (error) | |
2451337d | 1061 | return error; |
69ff2826 CH |
1062 | |
1063 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
51b0f3eb JL |
1064 | if (flags & (S_CTIME|S_MTIME)) |
1065 | now = inode_set_ctime_current(inode); | |
1066 | else | |
1067 | now = current_time(inode); | |
1068 | ||
3987848c | 1069 | if (flags & S_MTIME) |
51b0f3eb | 1070 | inode->i_mtime = now; |
3987848c | 1071 | if (flags & S_ATIME) |
51b0f3eb | 1072 | inode->i_atime = now; |
3987848c | 1073 | |
69ff2826 | 1074 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
c3b1b131 | 1075 | xfs_trans_log_inode(tp, ip, log_flags); |
70393313 | 1076 | return xfs_trans_commit(tp); |
69ff2826 CH |
1077 | } |
1078 | ||
f35642e2 ES |
1079 | STATIC int |
1080 | xfs_vn_fiemap( | |
1081 | struct inode *inode, | |
1082 | struct fiemap_extent_info *fieinfo, | |
1083 | u64 start, | |
1084 | u64 length) | |
1085 | { | |
f35642e2 ES |
1086 | int error; |
1087 | ||
d2bb140e | 1088 | xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); |
1d4795e7 CH |
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, | |
690c2a38 | 1095 | &xfs_read_iomap_ops); |
1d4795e7 | 1096 | } |
d2bb140e | 1097 | xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); |
f35642e2 | 1098 | |
d2bb140e | 1099 | return error; |
f35642e2 ES |
1100 | } |
1101 | ||
99b6436b ZYW |
1102 | STATIC int |
1103 | xfs_vn_tmpfile( | |
011e2b71 | 1104 | struct mnt_idmap *idmap, |
549c7297 | 1105 | struct inode *dir, |
863f144f | 1106 | struct file *file, |
549c7297 | 1107 | umode_t mode) |
99b6436b | 1108 | { |
f2d40141 | 1109 | int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file); |
863f144f MS |
1110 | |
1111 | return finish_open_simple(file, err); | |
99b6436b ZYW |
1112 | } |
1113 | ||
41be8bed | 1114 | static const struct inode_operations xfs_inode_operations = { |
cac2f8b8 | 1115 | .get_inode_acl = xfs_get_acl, |
2401dc29 | 1116 | .set_acl = xfs_set_acl, |
416c6d5b NS |
1117 | .getattr = xfs_vn_getattr, |
1118 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1119 | .listxattr = xfs_vn_listxattr, |
f35642e2 | 1120 | .fiemap = xfs_vn_fiemap, |
69ff2826 | 1121 | .update_time = xfs_vn_update_time, |
9fefd5db MS |
1122 | .fileattr_get = xfs_fileattr_get, |
1123 | .fileattr_set = xfs_fileattr_set, | |
1da177e4 LT |
1124 | }; |
1125 | ||
41be8bed | 1126 | static const struct inode_operations xfs_dir_inode_operations = { |
416c6d5b NS |
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, | |
8f112e3b CH |
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, | |
416c6d5b | 1140 | .mknod = xfs_vn_mknod, |
2773bf00 | 1141 | .rename = xfs_vn_rename, |
cac2f8b8 | 1142 | .get_inode_acl = xfs_get_acl, |
2401dc29 | 1143 | .set_acl = xfs_set_acl, |
416c6d5b NS |
1144 | .getattr = xfs_vn_getattr, |
1145 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1146 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1147 | .update_time = xfs_vn_update_time, |
99b6436b | 1148 | .tmpfile = xfs_vn_tmpfile, |
9fefd5db MS |
1149 | .fileattr_get = xfs_fileattr_get, |
1150 | .fileattr_set = xfs_fileattr_set, | |
1da177e4 LT |
1151 | }; |
1152 | ||
41be8bed | 1153 | static const struct inode_operations xfs_dir_ci_inode_operations = { |
384f3ced BN |
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, | |
8f112e3b CH |
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, | |
384f3ced | 1167 | .mknod = xfs_vn_mknod, |
2773bf00 | 1168 | .rename = xfs_vn_rename, |
cac2f8b8 | 1169 | .get_inode_acl = xfs_get_acl, |
2401dc29 | 1170 | .set_acl = xfs_set_acl, |
384f3ced BN |
1171 | .getattr = xfs_vn_getattr, |
1172 | .setattr = xfs_vn_setattr, | |
384f3ced | 1173 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1174 | .update_time = xfs_vn_update_time, |
99b6436b | 1175 | .tmpfile = xfs_vn_tmpfile, |
9fefd5db MS |
1176 | .fileattr_get = xfs_fileattr_get, |
1177 | .fileattr_set = xfs_fileattr_set, | |
384f3ced BN |
1178 | }; |
1179 | ||
41be8bed | 1180 | static const struct inode_operations xfs_symlink_inode_operations = { |
6b255391 | 1181 | .get_link = xfs_vn_get_link, |
416c6d5b NS |
1182 | .getattr = xfs_vn_getattr, |
1183 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1184 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1185 | .update_time = xfs_vn_update_time, |
1da177e4 | 1186 | }; |
41be8bed | 1187 | |
ba23cba9 DW |
1188 | /* Figure out if this file actually supports DAX. */ |
1189 | static bool | |
1190 | xfs_inode_supports_dax( | |
1191 | struct xfs_inode *ip) | |
1192 | { | |
1193 | struct xfs_mount *mp = ip->i_mount; | |
1194 | ||
32dbc565 IW |
1195 | /* Only supported on regular files. */ |
1196 | if (!S_ISREG(VFS_I(ip)->i_mode)) | |
ba23cba9 DW |
1197 | return false; |
1198 | ||
ba23cba9 DW |
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. */ | |
30fa529e | 1204 | return xfs_inode_buftarg(ip)->bt_daxdev != NULL; |
ba23cba9 DW |
1205 | } |
1206 | ||
32dbc565 IW |
1207 | static bool |
1208 | xfs_inode_should_enable_dax( | |
1209 | struct xfs_inode *ip) | |
1210 | { | |
1211 | if (!IS_ENABLED(CONFIG_FS_DAX)) | |
1212 | return false; | |
0560f31a | 1213 | if (xfs_has_dax_never(ip->i_mount)) |
32dbc565 IW |
1214 | return false; |
1215 | if (!xfs_inode_supports_dax(ip)) | |
1216 | return false; | |
0560f31a | 1217 | if (xfs_has_dax_always(ip->i_mount)) |
32dbc565 | 1218 | return true; |
3e09ab8f | 1219 | if (ip->i_diflags2 & XFS_DIFLAG2_DAX) |
32dbc565 IW |
1220 | return true; |
1221 | return false; | |
1222 | } | |
1223 | ||
840d493d | 1224 | void |
41be8bed | 1225 | xfs_diflags_to_iflags( |
840d493d IW |
1226 | struct xfs_inode *ip, |
1227 | bool init) | |
41be8bed | 1228 | { |
840d493d IW |
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; | |
41be8bed CH |
1252 | } |
1253 | ||
1254 | /* | |
2b3d1d41 | 1255 | * Initialize the Linux inode. |
bf904248 | 1256 | * |
58c90473 | 1257 | * When reading existing inodes from disk this is called directly from xfs_iget, |
132c460e YX |
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. | |
41be8bed CH |
1261 | */ |
1262 | void | |
1263 | xfs_setup_inode( | |
1264 | struct xfs_inode *ip) | |
1265 | { | |
bf904248 | 1266 | struct inode *inode = &ip->i_vnode; |
ad22c7a0 | 1267 | gfp_t gfp_mask; |
bf904248 DC |
1268 | |
1269 | inode->i_ino = ip->i_ino; | |
f38a032b | 1270 | inode->i_state |= I_NEW; |
646ec461 CH |
1271 | |
1272 | inode_sb_list_add(inode); | |
c6f6cd06 | 1273 | /* make the inode look hashed for the writeback code */ |
5bef9151 | 1274 | inode_fake_hash(inode); |
41be8bed | 1275 | |
13d2c10b | 1276 | i_size_write(inode, ip->i_disk_size); |
840d493d | 1277 | xfs_diflags_to_iflags(ip, true); |
41be8bed | 1278 | |
2b3d1d41 | 1279 | if (S_ISDIR(inode->i_mode)) { |
ef215e39 DC |
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); | |
93a8614e | 1288 | lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class); |
2b3d1d41 | 1289 | } else { |
2b3d1d41 | 1290 | lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class); |
41be8bed CH |
1291 | } |
1292 | ||
ad22c7a0 DC |
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 | ||
510792ee CH |
1301 | /* |
1302 | * If there is no attribute fork no ACL can exist on this inode, | |
1303 | * and it can't have any file capabilities attached to it either. | |
1304 | */ | |
932b42c6 | 1305 | if (!xfs_inode_has_attr_fork(ip)) { |
510792ee | 1306 | inode_has_no_xattr(inode); |
6311b108 | 1307 | cache_no_acl(inode); |
510792ee | 1308 | } |
41be8bed | 1309 | } |
2b3d1d41 CH |
1310 | |
1311 | void | |
1312 | xfs_setup_iops( | |
1313 | struct xfs_inode *ip) | |
1314 | { | |
1315 | struct inode *inode = &ip->i_vnode; | |
1316 | ||
41be8bed CH |
1317 | switch (inode->i_mode & S_IFMT) { |
1318 | case S_IFREG: | |
1319 | inode->i_op = &xfs_inode_operations; | |
1320 | inode->i_fop = &xfs_file_operations; | |
6e2608df DW |
1321 | if (IS_DAX(inode)) |
1322 | inode->i_mapping->a_ops = &xfs_dax_aops; | |
1323 | else | |
1324 | inode->i_mapping->a_ops = &xfs_address_space_operations; | |
41be8bed CH |
1325 | break; |
1326 | case S_IFDIR: | |
38c26bfd | 1327 | if (xfs_has_asciici(XFS_M(inode->i_sb))) |
41be8bed CH |
1328 | inode->i_op = &xfs_dir_ci_inode_operations; |
1329 | else | |
1330 | inode->i_op = &xfs_dir_inode_operations; | |
1331 | inode->i_fop = &xfs_dir_file_operations; | |
1332 | break; | |
1333 | case S_IFLNK: | |
7b7820b8 | 1334 | inode->i_op = &xfs_symlink_inode_operations; |
41be8bed CH |
1335 | break; |
1336 | default: | |
1337 | inode->i_op = &xfs_inode_operations; | |
1338 | init_special_inode(inode, inode->i_mode, inode->i_rdev); | |
1339 | break; | |
1340 | } | |
41be8bed | 1341 | } |