]>
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" |
1da177e4 | 16 | #include "xfs_attr.h" |
239880ef | 17 | #include "xfs_trans.h" |
0b1b213f | 18 | #include "xfs_trace.h" |
27b52867 | 19 | #include "xfs_icache.h" |
c24b5dfa | 20 | #include "xfs_symlink.h" |
1b767ee3 | 21 | #include "xfs_dir2.h" |
68a9f5e7 | 22 | #include "xfs_iomap.h" |
a5155b87 | 23 | #include "xfs_error.h" |
1da177e4 | 24 | |
ef14f0c1 | 25 | #include <linux/posix_acl.h> |
446ada4a | 26 | #include <linux/security.h> |
c3b1b131 | 27 | #include <linux/iversion.h> |
10c5db28 | 28 | #include <linux/fiemap.h> |
1da177e4 | 29 | |
93a8614e | 30 | /* |
c1e8d7c6 | 31 | * Directories have different lock order w.r.t. mmap_lock compared to regular |
93a8614e DC |
32 | * files. This is due to readdir potentially triggering page faults on a user |
33 | * buffer inside filldir(), and this happens with the ilock on the directory | |
34 | * held. For regular files, the lock order is the other way around - the | |
c1e8d7c6 | 35 | * mmap_lock is taken during the page fault, and then we lock the ilock to do |
93a8614e DC |
36 | * block mapping. Hence we need a different class for the directory ilock so |
37 | * that lockdep can tell them apart. | |
38 | */ | |
39 | static struct lock_class_key xfs_nondir_ilock_class; | |
40 | static struct lock_class_key xfs_dir_ilock_class; | |
41 | ||
8d2a5e6e DC |
42 | static int |
43 | xfs_initxattrs( | |
44 | struct inode *inode, | |
45 | const struct xattr *xattr_array, | |
46 | void *fs_info) | |
9d8f13ba | 47 | { |
8d2a5e6e DC |
48 | const struct xattr *xattr; |
49 | struct xfs_inode *ip = XFS_I(inode); | |
50 | int error = 0; | |
9d8f13ba MZ |
51 | |
52 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
a2544622 CH |
53 | struct xfs_da_args args = { |
54 | .dp = ip, | |
d5f0f49a | 55 | .attr_filter = XFS_ATTR_SECURE, |
a2544622 CH |
56 | .name = xattr->name, |
57 | .namelen = strlen(xattr->name), | |
58 | .value = xattr->value, | |
59 | .valuelen = xattr->value_len, | |
60 | }; | |
61 | error = xfs_attr_set(&args); | |
9d8f13ba MZ |
62 | if (error < 0) |
63 | break; | |
64 | } | |
65 | return error; | |
66 | } | |
67 | ||
446ada4a NS |
68 | /* |
69 | * Hook in SELinux. This is not quite correct yet, what we really need | |
70 | * here (as we do for default ACLs) is a mechanism by which creation of | |
71 | * these attrs can be journalled at inode creation time (along with the | |
72 | * inode, of course, such that log replay can't cause these to be lost). | |
73 | */ | |
9d8f13ba | 74 | |
446ada4a | 75 | STATIC int |
416c6d5b | 76 | xfs_init_security( |
af048193 | 77 | struct inode *inode, |
2a7dba39 EP |
78 | struct inode *dir, |
79 | const struct qstr *qstr) | |
446ada4a | 80 | { |
2451337d | 81 | return security_inode_init_security(inode, dir, qstr, |
a5a14de2 | 82 | &xfs_initxattrs, NULL); |
446ada4a NS |
83 | } |
84 | ||
556b8b16 BN |
85 | static void |
86 | xfs_dentry_to_name( | |
fab8eef8 AG |
87 | struct xfs_name *namep, |
88 | struct dentry *dentry) | |
89 | { | |
90 | namep->name = dentry->d_name.name; | |
91 | namep->len = dentry->d_name.len; | |
92 | namep->type = XFS_DIR3_FT_UNKNOWN; | |
93 | } | |
94 | ||
95 | static int | |
96 | xfs_dentry_mode_to_name( | |
556b8b16 | 97 | struct xfs_name *namep, |
0cb97766 DC |
98 | struct dentry *dentry, |
99 | int mode) | |
556b8b16 BN |
100 | { |
101 | namep->name = dentry->d_name.name; | |
102 | namep->len = dentry->d_name.len; | |
1fc4d33f | 103 | namep->type = xfs_mode_to_ftype(mode); |
fab8eef8 AG |
104 | |
105 | if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN)) | |
106 | return -EFSCORRUPTED; | |
107 | ||
108 | return 0; | |
556b8b16 BN |
109 | } |
110 | ||
7989cb8e | 111 | STATIC void |
416c6d5b | 112 | xfs_cleanup_inode( |
739bfb2a | 113 | struct inode *dir, |
af048193 | 114 | struct inode *inode, |
8f112e3b | 115 | struct dentry *dentry) |
3a69c7dc | 116 | { |
556b8b16 | 117 | struct xfs_name teardown; |
3a69c7dc YL |
118 | |
119 | /* Oh, the horror. | |
220b5284 | 120 | * If we can't add the ACL or we fail in |
416c6d5b | 121 | * xfs_init_security we must back out. |
3a69c7dc YL |
122 | * ENOSPC can hit here, among other things. |
123 | */ | |
fab8eef8 | 124 | xfs_dentry_to_name(&teardown, dentry); |
3a69c7dc | 125 | |
8f112e3b | 126 | xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); |
3a69c7dc YL |
127 | } |
128 | ||
e6a688c3 DC |
129 | /* |
130 | * Check to see if we are likely to need an extended attribute to be added to | |
131 | * the inode we are about to allocate. This allows the attribute fork to be | |
132 | * created during the inode allocation, reducing the number of transactions we | |
133 | * need to do in this fast path. | |
134 | * | |
135 | * The security checks are optimistic, but not guaranteed. The two LSMs that | |
136 | * require xattrs to be added here (selinux and smack) are also the only two | |
137 | * LSMs that add a sb->s_security structure to the superblock. Hence if security | |
138 | * is enabled and sb->s_security is set, we have a pretty good idea that we are | |
139 | * going to be asked to add a security xattr immediately after allocating the | |
140 | * xfs inode and instantiating the VFS inode. | |
141 | */ | |
142 | static inline bool | |
143 | xfs_create_need_xattr( | |
144 | struct inode *dir, | |
145 | struct posix_acl *default_acl, | |
146 | struct posix_acl *acl) | |
147 | { | |
148 | if (acl) | |
149 | return true; | |
150 | if (default_acl) | |
151 | return true; | |
152 | #if IS_ENABLED(CONFIG_SECURITY) | |
153 | if (dir->i_sb->s_security) | |
154 | return true; | |
155 | #endif | |
156 | return false; | |
157 | } | |
158 | ||
159 | ||
1da177e4 | 160 | STATIC int |
d540e43b | 161 | xfs_generic_create( |
f736d93d | 162 | struct user_namespace *mnt_userns, |
1da177e4 LT |
163 | struct inode *dir, |
164 | struct dentry *dentry, | |
1a67aafb | 165 | umode_t mode, |
d540e43b BF |
166 | dev_t rdev, |
167 | bool tmpfile) /* unnamed file */ | |
1da177e4 | 168 | { |
db0bb7ba | 169 | struct inode *inode; |
979ebab1 | 170 | struct xfs_inode *ip = NULL; |
2401dc29 | 171 | struct posix_acl *default_acl, *acl; |
556b8b16 | 172 | struct xfs_name name; |
1da177e4 LT |
173 | int error; |
174 | ||
175 | /* | |
176 | * Irix uses Missed'em'V split, but doesn't want to see | |
177 | * the upper 5 bits of (14bit) major. | |
178 | */ | |
517b5e8c CH |
179 | if (S_ISCHR(mode) || S_ISBLK(mode)) { |
180 | if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff)) | |
181 | return -EINVAL; | |
517b5e8c CH |
182 | } else { |
183 | rdev = 0; | |
184 | } | |
1da177e4 | 185 | |
2401dc29 CH |
186 | error = posix_acl_create(dir, &mode, &default_acl, &acl); |
187 | if (error) | |
188 | return error; | |
1da177e4 | 189 | |
fab8eef8 AG |
190 | /* Verify mode is valid also for tmpfile case */ |
191 | error = xfs_dentry_mode_to_name(&name, dentry, mode); | |
192 | if (unlikely(error)) | |
193 | goto out_free_acl; | |
194 | ||
d540e43b | 195 | if (!tmpfile) { |
f736d93d | 196 | error = xfs_create(mnt_userns, XFS_I(dir), &name, mode, rdev, |
e6a688c3 DC |
197 | xfs_create_need_xattr(dir, default_acl, acl), |
198 | &ip); | |
d540e43b | 199 | } else { |
f736d93d | 200 | error = xfs_create_tmpfile(mnt_userns, XFS_I(dir), mode, &ip); |
d540e43b | 201 | } |
db0bb7ba CH |
202 | if (unlikely(error)) |
203 | goto out_free_acl; | |
446ada4a | 204 | |
01651646 | 205 | inode = VFS_I(ip); |
979ebab1 | 206 | |
2a7dba39 | 207 | error = xfs_init_security(inode, dir, &dentry->d_name); |
db0bb7ba CH |
208 | if (unlikely(error)) |
209 | goto out_cleanup_inode; | |
210 | ||
2401dc29 | 211 | #ifdef CONFIG_XFS_POSIX_ACL |
db0bb7ba | 212 | if (default_acl) { |
8ba35875 | 213 | error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); |
2401dc29 | 214 | if (error) |
db0bb7ba | 215 | goto out_cleanup_inode; |
1da177e4 | 216 | } |
2401dc29 | 217 | if (acl) { |
8ba35875 | 218 | error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); |
2401dc29 CH |
219 | if (error) |
220 | goto out_cleanup_inode; | |
221 | } | |
222 | #endif | |
1da177e4 | 223 | |
2b3d1d41 CH |
224 | xfs_setup_iops(ip); |
225 | ||
c4a6bf7f DW |
226 | if (tmpfile) { |
227 | /* | |
228 | * The VFS requires that any inode fed to d_tmpfile must have | |
229 | * nlink == 1 so that it can decrement the nlink in d_tmpfile. | |
230 | * However, we created the temp file with nlink == 0 because | |
231 | * we're not allowed to put an inode with nlink > 0 on the | |
232 | * unlinked list. Therefore we have to set nlink to 1 so that | |
233 | * d_tmpfile can immediately set it back to zero. | |
234 | */ | |
235 | set_nlink(inode, 1); | |
d540e43b | 236 | d_tmpfile(dentry, inode); |
c4a6bf7f | 237 | } else |
d540e43b BF |
238 | d_instantiate(dentry, inode); |
239 | ||
58c90473 DC |
240 | xfs_finish_inode_setup(ip); |
241 | ||
2401dc29 | 242 | out_free_acl: |
88269b88 KX |
243 | posix_acl_release(default_acl); |
244 | posix_acl_release(acl); | |
2451337d | 245 | return error; |
db0bb7ba CH |
246 | |
247 | out_cleanup_inode: | |
58c90473 | 248 | xfs_finish_inode_setup(ip); |
d540e43b BF |
249 | if (!tmpfile) |
250 | xfs_cleanup_inode(dir, inode, dentry); | |
44a8736b | 251 | xfs_irele(ip); |
2401dc29 | 252 | goto out_free_acl; |
1da177e4 LT |
253 | } |
254 | ||
d540e43b BF |
255 | STATIC int |
256 | xfs_vn_mknod( | |
549c7297 CB |
257 | struct user_namespace *mnt_userns, |
258 | struct inode *dir, | |
259 | struct dentry *dentry, | |
260 | umode_t mode, | |
261 | dev_t rdev) | |
d540e43b | 262 | { |
f736d93d | 263 | return xfs_generic_create(mnt_userns, dir, dentry, mode, rdev, false); |
d540e43b BF |
264 | } |
265 | ||
1da177e4 | 266 | STATIC int |
416c6d5b | 267 | xfs_vn_create( |
549c7297 CB |
268 | struct user_namespace *mnt_userns, |
269 | struct inode *dir, | |
270 | struct dentry *dentry, | |
271 | umode_t mode, | |
272 | bool flags) | |
1da177e4 | 273 | { |
f736d93d | 274 | return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, false); |
1da177e4 LT |
275 | } |
276 | ||
277 | STATIC int | |
416c6d5b | 278 | xfs_vn_mkdir( |
549c7297 CB |
279 | struct user_namespace *mnt_userns, |
280 | struct inode *dir, | |
281 | struct dentry *dentry, | |
282 | umode_t mode) | |
1da177e4 | 283 | { |
f736d93d CH |
284 | return xfs_generic_create(mnt_userns, dir, dentry, mode | S_IFDIR, 0, |
285 | false); | |
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 | */ | |
395 | if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb)) | |
396 | d_invalidate(dentry); | |
397 | return 0; | |
1da177e4 LT |
398 | } |
399 | ||
400 | STATIC int | |
416c6d5b | 401 | xfs_vn_symlink( |
549c7297 CB |
402 | struct user_namespace *mnt_userns, |
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 | |
f736d93d | 419 | error = xfs_symlink(mnt_userns, 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 | |
2a7dba39 | 425 | error = xfs_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( |
549c7297 CB |
445 | struct user_namespace *mnt_userns, |
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 | |
f736d93d CH |
474 | return xfs_rename(mnt_userns, XFS_I(odir), &oname, |
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 | ||
30ee052e CH |
513 | STATIC const char * |
514 | xfs_vn_get_link_inline( | |
515 | struct dentry *dentry, | |
516 | struct inode *inode, | |
517 | struct delayed_call *done) | |
518 | { | |
a5155b87 | 519 | struct xfs_inode *ip = XFS_I(inode); |
ae294787 DW |
520 | char *link; |
521 | ||
a5155b87 | 522 | ASSERT(ip->i_df.if_flags & XFS_IFINLINE); |
ae294787 DW |
523 | |
524 | /* | |
525 | * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if | |
526 | * if_data is junk. | |
527 | */ | |
a5155b87 | 528 | link = ip->i_df.if_u1.if_data; |
a71895c5 | 529 | if (XFS_IS_CORRUPT(ip->i_mount, !link)) |
ae294787 DW |
530 | return ERR_PTR(-EFSCORRUPTED); |
531 | return link; | |
30ee052e CH |
532 | } |
533 | ||
dd2d535e CH |
534 | static uint32_t |
535 | xfs_stat_blksize( | |
536 | struct xfs_inode *ip) | |
537 | { | |
538 | struct xfs_mount *mp = ip->i_mount; | |
539 | ||
540 | /* | |
541 | * If the file blocks are being allocated from a realtime volume, then | |
542 | * always return the realtime extent size. | |
543 | */ | |
544 | if (XFS_IS_REALTIME_INODE(ip)) | |
545 | return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog; | |
546 | ||
547 | /* | |
548 | * Allow large block sizes to be reported to userspace programs if the | |
549 | * "largeio" mount option is used. | |
550 | * | |
551 | * If compatibility mode is specified, simply return the basic unit of | |
552 | * caching so that we don't get inefficient read/modify/write I/O from | |
553 | * user apps. Otherwise.... | |
554 | * | |
555 | * If the underlying volume is a stripe, then return the stripe width in | |
556 | * bytes as the recommended I/O size. It is not a stripe and we've set a | |
557 | * default buffered I/O size, return that, otherwise return the compat | |
558 | * default. | |
559 | */ | |
7c6b94b1 | 560 | if (mp->m_flags & XFS_MOUNT_LARGEIO) { |
dd2d535e CH |
561 | if (mp->m_swidth) |
562 | return mp->m_swidth << mp->m_sb.sb_blocklog; | |
3274d008 | 563 | if (mp->m_flags & XFS_MOUNT_ALLOCSIZE) |
5da8a07c | 564 | return 1U << mp->m_allocsize_log; |
dd2d535e CH |
565 | } |
566 | ||
567 | return PAGE_SIZE; | |
568 | } | |
569 | ||
1da177e4 | 570 | STATIC int |
416c6d5b | 571 | xfs_vn_getattr( |
549c7297 | 572 | struct user_namespace *mnt_userns, |
a528d35e DH |
573 | const struct path *path, |
574 | struct kstat *stat, | |
575 | u32 request_mask, | |
576 | unsigned int query_flags) | |
1da177e4 | 577 | { |
a528d35e | 578 | struct inode *inode = d_inode(path->dentry); |
c43f4087 CH |
579 | struct xfs_inode *ip = XFS_I(inode); |
580 | struct xfs_mount *mp = ip->i_mount; | |
581 | ||
cca28fb8 | 582 | trace_xfs_getattr(ip); |
c43f4087 CH |
583 | |
584 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 585 | return -EIO; |
c43f4087 CH |
586 | |
587 | stat->size = XFS_ISIZE(ip); | |
588 | stat->dev = inode->i_sb->s_dev; | |
c19b3b05 | 589 | stat->mode = inode->i_mode; |
54d7b5c1 | 590 | stat->nlink = inode->i_nlink; |
f736d93d CH |
591 | stat->uid = i_uid_into_mnt(mnt_userns, inode); |
592 | stat->gid = i_gid_into_mnt(mnt_userns, inode); | |
c43f4087 | 593 | stat->ino = ip->i_ino; |
c43f4087 | 594 | stat->atime = inode->i_atime; |
f9581b14 CH |
595 | stat->mtime = inode->i_mtime; |
596 | stat->ctime = inode->i_ctime; | |
6e73a545 | 597 | stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); |
c43f4087 | 598 | |
6471e9c5 | 599 | if (xfs_sb_version_has_v3inode(&mp->m_sb)) { |
5f955f26 DW |
600 | if (request_mask & STATX_BTIME) { |
601 | stat->result_mask |= STATX_BTIME; | |
8d2d878d | 602 | stat->btime = ip->i_d.di_crtime; |
5f955f26 DW |
603 | } |
604 | } | |
605 | ||
1b9598c8 LR |
606 | /* |
607 | * Note: If you add another clause to set an attribute flag, please | |
608 | * update attributes_mask below. | |
609 | */ | |
db07349d | 610 | if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) |
5f955f26 | 611 | stat->attributes |= STATX_ATTR_IMMUTABLE; |
db07349d | 612 | if (ip->i_diflags & XFS_DIFLAG_APPEND) |
5f955f26 | 613 | stat->attributes |= STATX_ATTR_APPEND; |
db07349d | 614 | if (ip->i_diflags & XFS_DIFLAG_NODUMP) |
5f955f26 | 615 | stat->attributes |= STATX_ATTR_NODUMP; |
c43f4087 | 616 | |
1b9598c8 LR |
617 | stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | |
618 | STATX_ATTR_APPEND | | |
619 | STATX_ATTR_NODUMP); | |
620 | ||
c43f4087 CH |
621 | switch (inode->i_mode & S_IFMT) { |
622 | case S_IFBLK: | |
623 | case S_IFCHR: | |
624 | stat->blksize = BLKDEV_IOSIZE; | |
66f36464 | 625 | stat->rdev = inode->i_rdev; |
c43f4087 CH |
626 | break; |
627 | default: | |
dd2d535e | 628 | stat->blksize = xfs_stat_blksize(ip); |
c43f4087 CH |
629 | stat->rdev = 0; |
630 | break; | |
69e23b9a | 631 | } |
c43f4087 CH |
632 | |
633 | return 0; | |
1da177e4 LT |
634 | } |
635 | ||
56c19e89 DC |
636 | static void |
637 | xfs_setattr_mode( | |
56c19e89 DC |
638 | struct xfs_inode *ip, |
639 | struct iattr *iattr) | |
640 | { | |
0c3d88df CH |
641 | struct inode *inode = VFS_I(ip); |
642 | umode_t mode = iattr->ia_mode; | |
56c19e89 | 643 | |
56c19e89 DC |
644 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
645 | ||
56c19e89 DC |
646 | inode->i_mode &= S_IFMT; |
647 | inode->i_mode |= mode & ~S_IFMT; | |
648 | } | |
649 | ||
52785112 | 650 | void |
c91c46c1 CH |
651 | xfs_setattr_time( |
652 | struct xfs_inode *ip, | |
653 | struct iattr *iattr) | |
654 | { | |
655 | struct inode *inode = VFS_I(ip); | |
656 | ||
657 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
658 | ||
3987848c | 659 | if (iattr->ia_valid & ATTR_ATIME) |
c91c46c1 | 660 | inode->i_atime = iattr->ia_atime; |
3987848c | 661 | if (iattr->ia_valid & ATTR_CTIME) |
c91c46c1 | 662 | inode->i_ctime = iattr->ia_ctime; |
3987848c | 663 | if (iattr->ia_valid & ATTR_MTIME) |
c91c46c1 | 664 | inode->i_mtime = iattr->ia_mtime; |
c91c46c1 CH |
665 | } |
666 | ||
69bca807 JK |
667 | static int |
668 | xfs_vn_change_ok( | |
f736d93d CH |
669 | struct user_namespace *mnt_userns, |
670 | struct dentry *dentry, | |
671 | struct iattr *iattr) | |
69bca807 | 672 | { |
31051c85 | 673 | struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; |
69bca807 JK |
674 | |
675 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
676 | return -EROFS; | |
677 | ||
678 | if (XFS_FORCED_SHUTDOWN(mp)) | |
679 | return -EIO; | |
680 | ||
f736d93d | 681 | return setattr_prepare(mnt_userns, dentry, iattr); |
69bca807 JK |
682 | } |
683 | ||
684 | /* | |
685 | * Set non-size attributes of an inode. | |
686 | * | |
687 | * Caution: The caller of this function is responsible for calling | |
31051c85 | 688 | * setattr_prepare() or otherwise verifying the change is fine. |
69bca807 | 689 | */ |
5d24ec4c | 690 | static int |
c4ed4243 | 691 | xfs_setattr_nonsize( |
f736d93d | 692 | struct user_namespace *mnt_userns, |
c4ed4243 | 693 | struct xfs_inode *ip, |
5d24ec4c | 694 | struct iattr *iattr) |
c4ed4243 CH |
695 | { |
696 | xfs_mount_t *mp = ip->i_mount; | |
697 | struct inode *inode = VFS_I(ip); | |
698 | int mask = iattr->ia_valid; | |
699 | xfs_trans_t *tp; | |
700 | int error; | |
7aab1b28 DE |
701 | kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID; |
702 | kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID; | |
c4ed4243 CH |
703 | struct xfs_dquot *udqp = NULL, *gdqp = NULL; |
704 | struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL; | |
705 | ||
c4ed4243 CH |
706 | ASSERT((mask & ATTR_SIZE) == 0); |
707 | ||
708 | /* | |
709 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
710 | * before we start any other transactions. Trying to do this later | |
711 | * is messy. We don't care to take a readlock to look at the ids | |
712 | * in inode here, because we can't hold it across the trans_reserve. | |
713 | * If the IDs do change before we take the ilock, we're covered | |
714 | * because the i_*dquot fields will get updated anyway. | |
715 | */ | |
716 | if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { | |
717 | uint qflags = 0; | |
718 | ||
719 | if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { | |
720 | uid = iattr->ia_uid; | |
721 | qflags |= XFS_QMOPT_UQUOTA; | |
722 | } else { | |
7aab1b28 | 723 | uid = inode->i_uid; |
c4ed4243 CH |
724 | } |
725 | if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { | |
726 | gid = iattr->ia_gid; | |
727 | qflags |= XFS_QMOPT_GQUOTA; | |
728 | } else { | |
7aab1b28 | 729 | gid = inode->i_gid; |
c4ed4243 CH |
730 | } |
731 | ||
732 | /* | |
733 | * We take a reference when we initialize udqp and gdqp, | |
734 | * so it is important that we never blindly double trip on | |
735 | * the same variable. See xfs_create() for an example. | |
736 | */ | |
737 | ASSERT(udqp == NULL); | |
738 | ASSERT(gdqp == NULL); | |
ceaf603c | 739 | error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, |
7aab1b28 | 740 | qflags, &udqp, &gdqp, NULL); |
c4ed4243 CH |
741 | if (error) |
742 | return error; | |
743 | } | |
744 | ||
7317a03d DW |
745 | error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, |
746 | capable(CAP_FOWNER), &tp); | |
c4ed4243 | 747 | if (error) |
253f4911 | 748 | goto out_dqrele; |
c4ed4243 | 749 | |
c4ed4243 CH |
750 | /* |
751 | * Change file ownership. Must be the owner or privileged. | |
752 | */ | |
753 | if (mask & (ATTR_UID|ATTR_GID)) { | |
754 | /* | |
755 | * These IDs could have changed since we last looked at them. | |
756 | * But, we're assured that if the ownership did change | |
757 | * while we didn't have the inode locked, inode's dquot(s) | |
758 | * would have changed also. | |
759 | */ | |
7aab1b28 DE |
760 | iuid = inode->i_uid; |
761 | igid = inode->i_gid; | |
c4ed4243 CH |
762 | gid = (mask & ATTR_GID) ? iattr->ia_gid : igid; |
763 | uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid; | |
764 | ||
c4ed4243 CH |
765 | /* |
766 | * CAP_FSETID overrides the following restrictions: | |
767 | * | |
768 | * The set-user-ID and set-group-ID bits of a file will be | |
769 | * cleared upon successful return from chown() | |
770 | */ | |
c19b3b05 | 771 | if ((inode->i_mode & (S_ISUID|S_ISGID)) && |
c4ed4243 | 772 | !capable(CAP_FSETID)) |
c19b3b05 | 773 | inode->i_mode &= ~(S_ISUID|S_ISGID); |
c4ed4243 CH |
774 | |
775 | /* | |
776 | * Change the ownerships and register quota modifications | |
777 | * in the transaction. | |
778 | */ | |
7aab1b28 | 779 | if (!uid_eq(iuid, uid)) { |
c4ed4243 CH |
780 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) { |
781 | ASSERT(mask & ATTR_UID); | |
782 | ASSERT(udqp); | |
783 | olddquot1 = xfs_qm_vop_chown(tp, ip, | |
784 | &ip->i_udquot, udqp); | |
785 | } | |
c4ed4243 CH |
786 | inode->i_uid = uid; |
787 | } | |
7aab1b28 | 788 | if (!gid_eq(igid, gid)) { |
c4ed4243 | 789 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) { |
5a01dd54 JL |
790 | ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) || |
791 | !XFS_IS_PQUOTA_ON(mp)); | |
c4ed4243 CH |
792 | ASSERT(mask & ATTR_GID); |
793 | ASSERT(gdqp); | |
794 | olddquot2 = xfs_qm_vop_chown(tp, ip, | |
795 | &ip->i_gdquot, gdqp); | |
796 | } | |
c4ed4243 CH |
797 | inode->i_gid = gid; |
798 | } | |
799 | } | |
800 | ||
56c19e89 | 801 | if (mask & ATTR_MODE) |
0c3d88df | 802 | xfs_setattr_mode(ip, iattr); |
c91c46c1 CH |
803 | if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) |
804 | xfs_setattr_time(ip, iattr); | |
c4ed4243 CH |
805 | |
806 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
807 | ||
ff6d6af2 | 808 | XFS_STATS_INC(mp, xs_ig_attrchg); |
c4ed4243 CH |
809 | |
810 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
811 | xfs_trans_set_sync(tp); | |
70393313 | 812 | error = xfs_trans_commit(tp); |
c4ed4243 | 813 | |
c4ed4243 CH |
814 | /* |
815 | * Release any dquot(s) the inode had kept before chown. | |
816 | */ | |
817 | xfs_qm_dqrele(olddquot1); | |
818 | xfs_qm_dqrele(olddquot2); | |
819 | xfs_qm_dqrele(udqp); | |
820 | xfs_qm_dqrele(gdqp); | |
821 | ||
822 | if (error) | |
b474c7ae | 823 | return error; |
c4ed4243 CH |
824 | |
825 | /* | |
826 | * XXX(hch): Updating the ACL entries is not atomic vs the i_mode | |
827 | * update. We could avoid this with linked transactions | |
828 | * and passing down the transaction pointer all the way | |
829 | * to attr_set. No previous user of the generic | |
830 | * Posix ACL code seems to care about this issue either. | |
831 | */ | |
5d24ec4c | 832 | if (mask & ATTR_MODE) { |
f736d93d | 833 | error = posix_acl_chmod(mnt_userns, inode, inode->i_mode); |
c4ed4243 | 834 | if (error) |
b474c7ae | 835 | return error; |
c4ed4243 CH |
836 | } |
837 | ||
838 | return 0; | |
839 | ||
253f4911 | 840 | out_dqrele: |
c4ed4243 CH |
841 | xfs_qm_dqrele(udqp); |
842 | xfs_qm_dqrele(gdqp); | |
843 | return error; | |
844 | } | |
845 | ||
846 | /* | |
847 | * Truncate file. Must have write permission and not be a directory. | |
69bca807 JK |
848 | * |
849 | * Caution: The caller of this function is responsible for calling | |
31051c85 | 850 | * setattr_prepare() or otherwise verifying the change is fine. |
c4ed4243 | 851 | */ |
7bf7a193 | 852 | STATIC int |
c4ed4243 | 853 | xfs_setattr_size( |
f736d93d | 854 | struct user_namespace *mnt_userns, |
c4ed4243 | 855 | struct xfs_inode *ip, |
76ca4c23 | 856 | struct iattr *iattr) |
c4ed4243 CH |
857 | { |
858 | struct xfs_mount *mp = ip->i_mount; | |
859 | struct inode *inode = VFS_I(ip); | |
673e8e59 | 860 | xfs_off_t oldsize, newsize; |
c4ed4243 CH |
861 | struct xfs_trans *tp; |
862 | int error; | |
f38996f5 | 863 | uint lock_flags = 0; |
5885ebda | 864 | bool did_zeroing = false; |
c4ed4243 | 865 | |
76ca4c23 | 866 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
e8e9ad42 | 867 | ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); |
c19b3b05 | 868 | ASSERT(S_ISREG(inode->i_mode)); |
fe60a8a0 | 869 | ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| |
88a9e03b | 870 | ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); |
c4ed4243 | 871 | |
ce7ae151 | 872 | oldsize = inode->i_size; |
673e8e59 CH |
873 | newsize = iattr->ia_size; |
874 | ||
c4ed4243 CH |
875 | /* |
876 | * Short circuit the truncate case for zero length files. | |
877 | */ | |
daf83964 | 878 | if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { |
fe60a8a0 | 879 | if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) |
76ca4c23 | 880 | return 0; |
681b1200 CH |
881 | |
882 | /* | |
883 | * Use the regular setattr path to update the timestamps. | |
884 | */ | |
681b1200 | 885 | iattr->ia_valid &= ~ATTR_SIZE; |
f736d93d | 886 | return xfs_setattr_nonsize(mnt_userns, ip, iattr); |
c4ed4243 CH |
887 | } |
888 | ||
889 | /* | |
890 | * Make sure that the dquots are attached to the inode. | |
891 | */ | |
c14cfcca | 892 | error = xfs_qm_dqattach(ip); |
c4ed4243 | 893 | if (error) |
76ca4c23 | 894 | return error; |
c4ed4243 | 895 | |
f0c6bcba CH |
896 | /* |
897 | * Wait for all direct I/O to complete. | |
898 | */ | |
899 | inode_dio_wait(inode); | |
900 | ||
c4ed4243 | 901 | /* |
5885ebda DC |
902 | * File data changes must be complete before we start the transaction to |
903 | * modify the inode. This needs to be done before joining the inode to | |
904 | * the transaction because the inode cannot be unlocked once it is a | |
905 | * part of the transaction. | |
906 | * | |
f0c6bcba CH |
907 | * Start with zeroing any data beyond EOF that we may expose on file |
908 | * extension, or zeroing out the rest of the block on a downward | |
909 | * truncate. | |
c4ed4243 | 910 | */ |
673e8e59 | 911 | if (newsize > oldsize) { |
f5c54717 CH |
912 | trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); |
913 | error = iomap_zero_range(inode, oldsize, newsize - oldsize, | |
f150b423 | 914 | &did_zeroing, &xfs_buffered_write_iomap_ops); |
f0c6bcba | 915 | } else { |
869ae85d BF |
916 | /* |
917 | * iomap won't detect a dirty page over an unwritten block (or a | |
918 | * cow block over a hole) and subsequently skips zeroing the | |
919 | * newly post-EOF portion of the page. Flush the new EOF to | |
920 | * convert the block before the pagecache truncate. | |
921 | */ | |
922 | error = filemap_write_and_wait_range(inode->i_mapping, newsize, | |
923 | newsize); | |
924 | if (error) | |
925 | return error; | |
459f0fbc | 926 | error = iomap_truncate_page(inode, newsize, &did_zeroing, |
f150b423 | 927 | &xfs_buffered_write_iomap_ops); |
c4ed4243 | 928 | } |
c4ed4243 | 929 | |
f0c6bcba CH |
930 | if (error) |
931 | return error; | |
932 | ||
49abc3a8 | 933 | /* |
0f9160b4 DC |
934 | * We've already locked out new page faults, so now we can safely remove |
935 | * pages from the page cache knowing they won't get refaulted until we | |
936 | * drop the XFS_MMAP_EXCL lock after the extent manipulations are | |
937 | * complete. The truncate_setsize() call also cleans partial EOF page | |
938 | * PTEs on extending truncates and hence ensures sub-page block size | |
939 | * filesystems are correctly handled, too. | |
49abc3a8 | 940 | * |
0f9160b4 DC |
941 | * We have to do all the page cache truncate work outside the |
942 | * transaction context as the "lock" order is page lock->log space | |
943 | * reservation as defined by extent allocation in the writeback path. | |
253f4911 | 944 | * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but |
0f9160b4 DC |
945 | * having already truncated the in-memory version of the file (i.e. made |
946 | * user visible changes). There's not much we can do about this, except | |
947 | * to hope that the caller sees ENOMEM and retries the truncate | |
948 | * operation. | |
350976ae EG |
949 | * |
950 | * And we update in-core i_size and truncate page cache beyond newsize | |
13d2c10b CH |
951 | * before writeback the [i_disk_size, newsize] range, so we're |
952 | * guaranteed not to write stale data past the new EOF on truncate down. | |
49abc3a8 | 953 | */ |
49abc3a8 | 954 | truncate_setsize(inode, newsize); |
c4ed4243 | 955 | |
350976ae EG |
956 | /* |
957 | * We are going to log the inode size change in this transaction so | |
958 | * any previous writes that are beyond the on disk EOF and the new | |
959 | * EOF that have not been written out need to be written here. If we | |
960 | * do not write the data out, we expose ourselves to the null files | |
961 | * problem. Note that this includes any block zeroing we did above; | |
962 | * otherwise those blocks may not be zeroed after a crash. | |
963 | */ | |
964 | if (did_zeroing || | |
13d2c10b | 965 | (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { |
350976ae | 966 | error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
13d2c10b | 967 | ip->i_disk_size, newsize - 1); |
350976ae EG |
968 | if (error) |
969 | return error; | |
970 | } | |
971 | ||
253f4911 | 972 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); |
c4ed4243 | 973 | if (error) |
253f4911 | 974 | return error; |
c4ed4243 | 975 | |
c4ed4243 | 976 | lock_flags |= XFS_ILOCK_EXCL; |
c4ed4243 | 977 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
ddc3415a | 978 | xfs_trans_ijoin(tp, ip, 0); |
c4ed4243 CH |
979 | |
980 | /* | |
981 | * Only change the c/mtime if we are changing the size or we are | |
982 | * explicitly asked to change it. This handles the semantic difference | |
983 | * between truncate() and ftruncate() as implemented in the VFS. | |
984 | * | |
985 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
986 | * special case where we need to update the times despite not having | |
987 | * these flags set. For all other operations the VFS set these flags | |
988 | * explicitly if it wants a timestamp update. | |
989 | */ | |
fe60a8a0 CH |
990 | if (newsize != oldsize && |
991 | !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { | |
c4ed4243 | 992 | iattr->ia_ctime = iattr->ia_mtime = |
c2050a45 | 993 | current_time(inode); |
fe60a8a0 | 994 | iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; |
c4ed4243 CH |
995 | } |
996 | ||
673e8e59 CH |
997 | /* |
998 | * The first thing we do is set the size to new_size permanently on | |
999 | * disk. This way we don't have to worry about anyone ever being able | |
1000 | * to look at the data being freed even in the face of a crash. | |
1001 | * What we're getting around here is the case where we free a block, it | |
1002 | * is allocated to another file, it is written to, and then we crash. | |
1003 | * If the new data gets written to the file but the log buffers | |
1004 | * containing the free and reallocation don't, then we'd end up with | |
1005 | * garbage in the blocks being freed. As long as we make the new size | |
1006 | * permanent before actually freeing any blocks it doesn't matter if | |
1007 | * they get written to. | |
1008 | */ | |
13d2c10b | 1009 | ip->i_disk_size = newsize; |
673e8e59 CH |
1010 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1011 | ||
1012 | if (newsize <= oldsize) { | |
1013 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); | |
c4ed4243 | 1014 | if (error) |
4906e215 | 1015 | goto out_trans_cancel; |
c4ed4243 CH |
1016 | |
1017 | /* | |
1018 | * Truncated "down", so we're removing references to old data | |
1019 | * here - if we delay flushing for a long time, we expose | |
1020 | * ourselves unduly to the notorious NULL files problem. So, | |
1021 | * we mark this inode and flush it when the file is closed, | |
1022 | * and do not wait the usual (long) time for writeout. | |
1023 | */ | |
1024 | xfs_iflags_set(ip, XFS_ITRUNCATED); | |
27b52867 BF |
1025 | |
1026 | /* A truncate down always removes post-EOF blocks. */ | |
1027 | xfs_inode_clear_eofblocks_tag(ip); | |
c4ed4243 CH |
1028 | } |
1029 | ||
fe60a8a0 | 1030 | if (iattr->ia_valid & ATTR_MODE) |
0c3d88df | 1031 | xfs_setattr_mode(ip, iattr); |
fe60a8a0 | 1032 | if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) |
c91c46c1 | 1033 | xfs_setattr_time(ip, iattr); |
c4ed4243 CH |
1034 | |
1035 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1036 | ||
ff6d6af2 | 1037 | XFS_STATS_INC(mp, xs_ig_attrchg); |
c4ed4243 CH |
1038 | |
1039 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1040 | xfs_trans_set_sync(tp); | |
1041 | ||
70393313 | 1042 | error = xfs_trans_commit(tp); |
c4ed4243 CH |
1043 | out_unlock: |
1044 | if (lock_flags) | |
1045 | xfs_iunlock(ip, lock_flags); | |
1046 | return error; | |
1047 | ||
c4ed4243 | 1048 | out_trans_cancel: |
4906e215 | 1049 | xfs_trans_cancel(tp); |
c4ed4243 CH |
1050 | goto out_unlock; |
1051 | } | |
1052 | ||
69bca807 JK |
1053 | int |
1054 | xfs_vn_setattr_size( | |
f736d93d | 1055 | struct user_namespace *mnt_userns, |
69bca807 JK |
1056 | struct dentry *dentry, |
1057 | struct iattr *iattr) | |
1058 | { | |
1059 | struct xfs_inode *ip = XFS_I(d_inode(dentry)); | |
1060 | int error; | |
1061 | ||
1062 | trace_xfs_setattr(ip); | |
1063 | ||
f736d93d | 1064 | error = xfs_vn_change_ok(mnt_userns, dentry, iattr); |
69bca807 JK |
1065 | if (error) |
1066 | return error; | |
f736d93d | 1067 | return xfs_setattr_size(mnt_userns, ip, iattr); |
69bca807 JK |
1068 | } |
1069 | ||
1da177e4 | 1070 | STATIC int |
416c6d5b | 1071 | xfs_vn_setattr( |
549c7297 | 1072 | struct user_namespace *mnt_userns, |
76ca4c23 CH |
1073 | struct dentry *dentry, |
1074 | struct iattr *iattr) | |
1da177e4 | 1075 | { |
26f88363 CH |
1076 | struct inode *inode = d_inode(dentry); |
1077 | struct xfs_inode *ip = XFS_I(inode); | |
76ca4c23 CH |
1078 | int error; |
1079 | ||
1080 | if (iattr->ia_valid & ATTR_SIZE) { | |
c63a8eae | 1081 | uint iolock; |
781355c6 | 1082 | |
c63a8eae DW |
1083 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
1084 | iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; | |
781355c6 | 1085 | |
69eb5fa1 | 1086 | error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); |
c63a8eae DW |
1087 | if (error) { |
1088 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); | |
65523218 | 1089 | return error; |
c63a8eae | 1090 | } |
e8e9ad42 | 1091 | |
f736d93d | 1092 | error = xfs_vn_setattr_size(mnt_userns, dentry, iattr); |
65523218 | 1093 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
76ca4c23 | 1094 | } else { |
26f88363 CH |
1095 | trace_xfs_setattr(ip); |
1096 | ||
f736d93d | 1097 | error = xfs_vn_change_ok(mnt_userns, dentry, iattr); |
26f88363 | 1098 | if (!error) |
f736d93d | 1099 | error = xfs_setattr_nonsize(mnt_userns, ip, iattr); |
76ca4c23 CH |
1100 | } |
1101 | ||
2451337d | 1102 | return error; |
1da177e4 LT |
1103 | } |
1104 | ||
69ff2826 CH |
1105 | STATIC int |
1106 | xfs_vn_update_time( | |
1107 | struct inode *inode, | |
95582b00 | 1108 | struct timespec64 *now, |
69ff2826 CH |
1109 | int flags) |
1110 | { | |
1111 | struct xfs_inode *ip = XFS_I(inode); | |
1112 | struct xfs_mount *mp = ip->i_mount; | |
c3b1b131 | 1113 | int log_flags = XFS_ILOG_TIMESTAMP; |
69ff2826 CH |
1114 | struct xfs_trans *tp; |
1115 | int error; | |
1116 | ||
1117 | trace_xfs_update_time(ip); | |
1118 | ||
c3b1b131 CH |
1119 | if (inode->i_sb->s_flags & SB_LAZYTIME) { |
1120 | if (!((flags & S_VERSION) && | |
1121 | inode_maybe_inc_iversion(inode, false))) | |
1122 | return generic_update_time(inode, now, flags); | |
1123 | ||
1124 | /* Capture the iversion update that just occurred */ | |
1125 | log_flags |= XFS_ILOG_CORE; | |
1126 | } | |
1127 | ||
253f4911 CH |
1128 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
1129 | if (error) | |
2451337d | 1130 | return error; |
69ff2826 CH |
1131 | |
1132 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3987848c | 1133 | if (flags & S_CTIME) |
69ff2826 | 1134 | inode->i_ctime = *now; |
3987848c | 1135 | if (flags & S_MTIME) |
69ff2826 | 1136 | inode->i_mtime = *now; |
3987848c | 1137 | if (flags & S_ATIME) |
69ff2826 | 1138 | inode->i_atime = *now; |
3987848c | 1139 | |
69ff2826 | 1140 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
c3b1b131 | 1141 | xfs_trans_log_inode(tp, ip, log_flags); |
70393313 | 1142 | return xfs_trans_commit(tp); |
69ff2826 CH |
1143 | } |
1144 | ||
f35642e2 ES |
1145 | STATIC int |
1146 | xfs_vn_fiemap( | |
1147 | struct inode *inode, | |
1148 | struct fiemap_extent_info *fieinfo, | |
1149 | u64 start, | |
1150 | u64 length) | |
1151 | { | |
f35642e2 ES |
1152 | int error; |
1153 | ||
d2bb140e | 1154 | xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); |
1d4795e7 CH |
1155 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { |
1156 | fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; | |
1157 | error = iomap_fiemap(inode, fieinfo, start, length, | |
1158 | &xfs_xattr_iomap_ops); | |
1159 | } else { | |
1160 | error = iomap_fiemap(inode, fieinfo, start, length, | |
690c2a38 | 1161 | &xfs_read_iomap_ops); |
1d4795e7 | 1162 | } |
d2bb140e | 1163 | xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); |
f35642e2 | 1164 | |
d2bb140e | 1165 | return error; |
f35642e2 ES |
1166 | } |
1167 | ||
99b6436b ZYW |
1168 | STATIC int |
1169 | xfs_vn_tmpfile( | |
549c7297 CB |
1170 | struct user_namespace *mnt_userns, |
1171 | struct inode *dir, | |
1172 | struct dentry *dentry, | |
1173 | umode_t mode) | |
99b6436b | 1174 | { |
f736d93d | 1175 | return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, true); |
99b6436b ZYW |
1176 | } |
1177 | ||
41be8bed | 1178 | static const struct inode_operations xfs_inode_operations = { |
4e34e719 | 1179 | .get_acl = xfs_get_acl, |
2401dc29 | 1180 | .set_acl = xfs_set_acl, |
416c6d5b NS |
1181 | .getattr = xfs_vn_getattr, |
1182 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1183 | .listxattr = xfs_vn_listxattr, |
f35642e2 | 1184 | .fiemap = xfs_vn_fiemap, |
69ff2826 | 1185 | .update_time = xfs_vn_update_time, |
1da177e4 LT |
1186 | }; |
1187 | ||
41be8bed | 1188 | static const struct inode_operations xfs_dir_inode_operations = { |
416c6d5b NS |
1189 | .create = xfs_vn_create, |
1190 | .lookup = xfs_vn_lookup, | |
1191 | .link = xfs_vn_link, | |
1192 | .unlink = xfs_vn_unlink, | |
1193 | .symlink = xfs_vn_symlink, | |
1194 | .mkdir = xfs_vn_mkdir, | |
8f112e3b CH |
1195 | /* |
1196 | * Yes, XFS uses the same method for rmdir and unlink. | |
1197 | * | |
1198 | * There are some subtile differences deeper in the code, | |
1199 | * but we use S_ISDIR to check for those. | |
1200 | */ | |
1201 | .rmdir = xfs_vn_unlink, | |
416c6d5b | 1202 | .mknod = xfs_vn_mknod, |
2773bf00 | 1203 | .rename = xfs_vn_rename, |
4e34e719 | 1204 | .get_acl = xfs_get_acl, |
2401dc29 | 1205 | .set_acl = xfs_set_acl, |
416c6d5b NS |
1206 | .getattr = xfs_vn_getattr, |
1207 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1208 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1209 | .update_time = xfs_vn_update_time, |
99b6436b | 1210 | .tmpfile = xfs_vn_tmpfile, |
1da177e4 LT |
1211 | }; |
1212 | ||
41be8bed | 1213 | static const struct inode_operations xfs_dir_ci_inode_operations = { |
384f3ced BN |
1214 | .create = xfs_vn_create, |
1215 | .lookup = xfs_vn_ci_lookup, | |
1216 | .link = xfs_vn_link, | |
1217 | .unlink = xfs_vn_unlink, | |
1218 | .symlink = xfs_vn_symlink, | |
1219 | .mkdir = xfs_vn_mkdir, | |
8f112e3b CH |
1220 | /* |
1221 | * Yes, XFS uses the same method for rmdir and unlink. | |
1222 | * | |
1223 | * There are some subtile differences deeper in the code, | |
1224 | * but we use S_ISDIR to check for those. | |
1225 | */ | |
1226 | .rmdir = xfs_vn_unlink, | |
384f3ced | 1227 | .mknod = xfs_vn_mknod, |
2773bf00 | 1228 | .rename = xfs_vn_rename, |
4e34e719 | 1229 | .get_acl = xfs_get_acl, |
2401dc29 | 1230 | .set_acl = xfs_set_acl, |
384f3ced BN |
1231 | .getattr = xfs_vn_getattr, |
1232 | .setattr = xfs_vn_setattr, | |
384f3ced | 1233 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1234 | .update_time = xfs_vn_update_time, |
99b6436b | 1235 | .tmpfile = xfs_vn_tmpfile, |
384f3ced BN |
1236 | }; |
1237 | ||
41be8bed | 1238 | static const struct inode_operations xfs_symlink_inode_operations = { |
6b255391 | 1239 | .get_link = xfs_vn_get_link, |
416c6d5b NS |
1240 | .getattr = xfs_vn_getattr, |
1241 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1242 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1243 | .update_time = xfs_vn_update_time, |
1da177e4 | 1244 | }; |
41be8bed | 1245 | |
30ee052e | 1246 | static const struct inode_operations xfs_inline_symlink_inode_operations = { |
30ee052e CH |
1247 | .get_link = xfs_vn_get_link_inline, |
1248 | .getattr = xfs_vn_getattr, | |
1249 | .setattr = xfs_vn_setattr, | |
30ee052e CH |
1250 | .listxattr = xfs_vn_listxattr, |
1251 | .update_time = xfs_vn_update_time, | |
1252 | }; | |
1253 | ||
ba23cba9 DW |
1254 | /* Figure out if this file actually supports DAX. */ |
1255 | static bool | |
1256 | xfs_inode_supports_dax( | |
1257 | struct xfs_inode *ip) | |
1258 | { | |
1259 | struct xfs_mount *mp = ip->i_mount; | |
1260 | ||
32dbc565 IW |
1261 | /* Only supported on regular files. */ |
1262 | if (!S_ISREG(VFS_I(ip)->i_mode)) | |
ba23cba9 DW |
1263 | return false; |
1264 | ||
32dbc565 IW |
1265 | /* Only supported on non-reflinked files. */ |
1266 | if (xfs_is_reflink_inode(ip)) | |
ba23cba9 DW |
1267 | return false; |
1268 | ||
1269 | /* Block size must match page size */ | |
1270 | if (mp->m_sb.sb_blocksize != PAGE_SIZE) | |
1271 | return false; | |
1272 | ||
1273 | /* Device has to support DAX too. */ | |
30fa529e | 1274 | return xfs_inode_buftarg(ip)->bt_daxdev != NULL; |
ba23cba9 DW |
1275 | } |
1276 | ||
32dbc565 IW |
1277 | static bool |
1278 | xfs_inode_should_enable_dax( | |
1279 | struct xfs_inode *ip) | |
1280 | { | |
1281 | if (!IS_ENABLED(CONFIG_FS_DAX)) | |
1282 | return false; | |
1283 | if (ip->i_mount->m_flags & XFS_MOUNT_DAX_NEVER) | |
1284 | return false; | |
1285 | if (!xfs_inode_supports_dax(ip)) | |
1286 | return false; | |
1287 | if (ip->i_mount->m_flags & XFS_MOUNT_DAX_ALWAYS) | |
1288 | return true; | |
1289 | if (ip->i_d.di_flags2 & XFS_DIFLAG2_DAX) | |
1290 | return true; | |
1291 | return false; | |
1292 | } | |
1293 | ||
840d493d | 1294 | void |
41be8bed | 1295 | xfs_diflags_to_iflags( |
840d493d IW |
1296 | struct xfs_inode *ip, |
1297 | bool init) | |
41be8bed | 1298 | { |
840d493d IW |
1299 | struct inode *inode = VFS_I(ip); |
1300 | unsigned int xflags = xfs_ip2xflags(ip); | |
1301 | unsigned int flags = 0; | |
1302 | ||
1303 | ASSERT(!(IS_DAX(inode) && init)); | |
1304 | ||
1305 | if (xflags & FS_XFLAG_IMMUTABLE) | |
1306 | flags |= S_IMMUTABLE; | |
1307 | if (xflags & FS_XFLAG_APPEND) | |
1308 | flags |= S_APPEND; | |
1309 | if (xflags & FS_XFLAG_SYNC) | |
1310 | flags |= S_SYNC; | |
1311 | if (xflags & FS_XFLAG_NOATIME) | |
1312 | flags |= S_NOATIME; | |
1313 | if (init && xfs_inode_should_enable_dax(ip)) | |
1314 | flags |= S_DAX; | |
1315 | ||
1316 | /* | |
1317 | * S_DAX can only be set during inode initialization and is never set by | |
1318 | * the VFS, so we cannot mask off S_DAX in i_flags. | |
1319 | */ | |
1320 | inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME); | |
1321 | inode->i_flags |= flags; | |
41be8bed CH |
1322 | } |
1323 | ||
1324 | /* | |
2b3d1d41 | 1325 | * Initialize the Linux inode. |
bf904248 | 1326 | * |
58c90473 DC |
1327 | * When reading existing inodes from disk this is called directly from xfs_iget, |
1328 | * when creating a new inode it is called from xfs_ialloc after setting up the | |
1329 | * inode. These callers have different criteria for clearing XFS_INEW, so leave | |
1330 | * it up to the caller to deal with unlocking the inode appropriately. | |
41be8bed CH |
1331 | */ |
1332 | void | |
1333 | xfs_setup_inode( | |
1334 | struct xfs_inode *ip) | |
1335 | { | |
bf904248 | 1336 | struct inode *inode = &ip->i_vnode; |
ad22c7a0 | 1337 | gfp_t gfp_mask; |
bf904248 DC |
1338 | |
1339 | inode->i_ino = ip->i_ino; | |
eaff8079 | 1340 | inode->i_state = I_NEW; |
646ec461 CH |
1341 | |
1342 | inode_sb_list_add(inode); | |
c6f6cd06 | 1343 | /* make the inode look hashed for the writeback code */ |
5bef9151 | 1344 | inode_fake_hash(inode); |
41be8bed | 1345 | |
13d2c10b | 1346 | i_size_write(inode, ip->i_disk_size); |
840d493d | 1347 | xfs_diflags_to_iflags(ip, true); |
41be8bed | 1348 | |
2b3d1d41 | 1349 | if (S_ISDIR(inode->i_mode)) { |
ef215e39 DC |
1350 | /* |
1351 | * We set the i_rwsem class here to avoid potential races with | |
1352 | * lockdep_annotate_inode_mutex_key() reinitialising the lock | |
1353 | * after a filehandle lookup has already found the inode in | |
1354 | * cache before it has been unlocked via unlock_new_inode(). | |
1355 | */ | |
1356 | lockdep_set_class(&inode->i_rwsem, | |
1357 | &inode->i_sb->s_type->i_mutex_dir_key); | |
93a8614e | 1358 | lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class); |
2b3d1d41 | 1359 | } else { |
2b3d1d41 | 1360 | lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class); |
41be8bed CH |
1361 | } |
1362 | ||
ad22c7a0 DC |
1363 | /* |
1364 | * Ensure all page cache allocations are done from GFP_NOFS context to | |
1365 | * prevent direct reclaim recursion back into the filesystem and blowing | |
1366 | * stacks or deadlocking. | |
1367 | */ | |
1368 | gfp_mask = mapping_gfp_mask(inode->i_mapping); | |
1369 | mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); | |
1370 | ||
510792ee CH |
1371 | /* |
1372 | * If there is no attribute fork no ACL can exist on this inode, | |
1373 | * and it can't have any file capabilities attached to it either. | |
1374 | */ | |
1375 | if (!XFS_IFORK_Q(ip)) { | |
1376 | inode_has_no_xattr(inode); | |
6311b108 | 1377 | cache_no_acl(inode); |
510792ee | 1378 | } |
41be8bed | 1379 | } |
2b3d1d41 CH |
1380 | |
1381 | void | |
1382 | xfs_setup_iops( | |
1383 | struct xfs_inode *ip) | |
1384 | { | |
1385 | struct inode *inode = &ip->i_vnode; | |
1386 | ||
41be8bed CH |
1387 | switch (inode->i_mode & S_IFMT) { |
1388 | case S_IFREG: | |
1389 | inode->i_op = &xfs_inode_operations; | |
1390 | inode->i_fop = &xfs_file_operations; | |
6e2608df DW |
1391 | if (IS_DAX(inode)) |
1392 | inode->i_mapping->a_ops = &xfs_dax_aops; | |
1393 | else | |
1394 | inode->i_mapping->a_ops = &xfs_address_space_operations; | |
41be8bed CH |
1395 | break; |
1396 | case S_IFDIR: | |
1397 | if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb)) | |
1398 | inode->i_op = &xfs_dir_ci_inode_operations; | |
1399 | else | |
1400 | inode->i_op = &xfs_dir_inode_operations; | |
1401 | inode->i_fop = &xfs_dir_file_operations; | |
1402 | break; | |
1403 | case S_IFLNK: | |
30ee052e CH |
1404 | if (ip->i_df.if_flags & XFS_IFINLINE) |
1405 | inode->i_op = &xfs_inline_symlink_inode_operations; | |
1406 | else | |
1407 | inode->i_op = &xfs_symlink_inode_operations; | |
41be8bed CH |
1408 | break; |
1409 | default: | |
1410 | inode->i_op = &xfs_inode_operations; | |
1411 | init_special_inode(inode, inode->i_mode, inode->i_rdev); | |
1412 | break; | |
1413 | } | |
41be8bed | 1414 | } |