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( |
f736d93d | 165 | struct user_namespace *mnt_userns, |
1da177e4 LT |
166 | struct inode *dir, |
167 | struct dentry *dentry, | |
1a67aafb | 168 | umode_t mode, |
d540e43b | 169 | dev_t rdev, |
863f144f | 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) { |
f736d93d | 199 | error = xfs_create(mnt_userns, XFS_I(dir), &name, mode, rdev, |
e6a688c3 DC |
200 | xfs_create_need_xattr(dir, default_acl, acl), |
201 | &ip); | |
d540e43b | 202 | } else { |
f736d93d | 203 | error = xfs_create_tmpfile(mnt_userns, 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( | |
549c7297 CB |
258 | struct user_namespace *mnt_userns, |
259 | struct inode *dir, | |
260 | struct dentry *dentry, | |
261 | umode_t mode, | |
262 | dev_t rdev) | |
d540e43b | 263 | { |
863f144f | 264 | return xfs_generic_create(mnt_userns, dir, dentry, mode, rdev, NULL); |
d540e43b BF |
265 | } |
266 | ||
1da177e4 | 267 | STATIC int |
416c6d5b | 268 | xfs_vn_create( |
549c7297 CB |
269 | struct user_namespace *mnt_userns, |
270 | struct inode *dir, | |
271 | struct dentry *dentry, | |
272 | umode_t mode, | |
273 | bool flags) | |
1da177e4 | 274 | { |
863f144f | 275 | return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, NULL); |
1da177e4 LT |
276 | } |
277 | ||
278 | STATIC int | |
416c6d5b | 279 | xfs_vn_mkdir( |
549c7297 CB |
280 | struct user_namespace *mnt_userns, |
281 | struct inode *dir, | |
282 | struct dentry *dentry, | |
283 | umode_t mode) | |
1da177e4 | 284 | { |
f736d93d | 285 | return xfs_generic_create(mnt_userns, dir, dentry, mode | S_IFDIR, 0, |
863f144f | 286 | NULL); |
1da177e4 LT |
287 | } |
288 | ||
289 | STATIC struct dentry * | |
416c6d5b | 290 | xfs_vn_lookup( |
1da177e4 LT |
291 | struct inode *dir, |
292 | struct dentry *dentry, | |
00cd8dd3 | 293 | unsigned int flags) |
1da177e4 | 294 | { |
b113a6d3 | 295 | struct inode *inode; |
ef1f5e7a | 296 | struct xfs_inode *cip; |
556b8b16 | 297 | struct xfs_name name; |
1da177e4 LT |
298 | int error; |
299 | ||
300 | if (dentry->d_name.len >= MAXNAMELEN) | |
301 | return ERR_PTR(-ENAMETOOLONG); | |
302 | ||
fab8eef8 | 303 | xfs_dentry_to_name(&name, dentry); |
384f3ced | 304 | error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); |
b113a6d3 AV |
305 | if (likely(!error)) |
306 | inode = VFS_I(cip); | |
307 | else if (likely(error == -ENOENT)) | |
308 | inode = NULL; | |
309 | else | |
310 | inode = ERR_PTR(error); | |
311 | return d_splice_alias(inode, dentry); | |
1da177e4 LT |
312 | } |
313 | ||
384f3ced BN |
314 | STATIC struct dentry * |
315 | xfs_vn_ci_lookup( | |
316 | struct inode *dir, | |
317 | struct dentry *dentry, | |
00cd8dd3 | 318 | unsigned int flags) |
384f3ced BN |
319 | { |
320 | struct xfs_inode *ip; | |
321 | struct xfs_name xname; | |
322 | struct xfs_name ci_name; | |
323 | struct qstr dname; | |
324 | int error; | |
325 | ||
326 | if (dentry->d_name.len >= MAXNAMELEN) | |
327 | return ERR_PTR(-ENAMETOOLONG); | |
328 | ||
fab8eef8 | 329 | xfs_dentry_to_name(&xname, dentry); |
384f3ced BN |
330 | error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); |
331 | if (unlikely(error)) { | |
2451337d DC |
332 | if (unlikely(error != -ENOENT)) |
333 | return ERR_PTR(error); | |
866d5dc9 BN |
334 | /* |
335 | * call d_add(dentry, NULL) here when d_drop_negative_children | |
336 | * is called in xfs_vn_mknod (ie. allow negative dentries | |
337 | * with CI filesystems). | |
338 | */ | |
384f3ced BN |
339 | return NULL; |
340 | } | |
341 | ||
342 | /* if exact match, just splice and exit */ | |
343 | if (!ci_name.name) | |
01651646 | 344 | return d_splice_alias(VFS_I(ip), dentry); |
384f3ced BN |
345 | |
346 | /* else case-insensitive match... */ | |
347 | dname.name = ci_name.name; | |
348 | dname.len = ci_name.len; | |
e45b590b | 349 | dentry = d_add_ci(dentry, VFS_I(ip), &dname); |
384f3ced BN |
350 | kmem_free(ci_name.name); |
351 | return dentry; | |
352 | } | |
353 | ||
1da177e4 | 354 | STATIC int |
416c6d5b | 355 | xfs_vn_link( |
1da177e4 LT |
356 | struct dentry *old_dentry, |
357 | struct inode *dir, | |
358 | struct dentry *dentry) | |
359 | { | |
2b0143b5 | 360 | struct inode *inode = d_inode(old_dentry); |
556b8b16 | 361 | struct xfs_name name; |
1da177e4 LT |
362 | int error; |
363 | ||
fab8eef8 AG |
364 | error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode); |
365 | if (unlikely(error)) | |
366 | return error; | |
1da177e4 | 367 | |
556b8b16 | 368 | error = xfs_link(XFS_I(dir), XFS_I(inode), &name); |
d9424b3c | 369 | if (unlikely(error)) |
2451337d | 370 | return error; |
a3da7896 | 371 | |
7de9c6ee | 372 | ihold(inode); |
a3da7896 CH |
373 | d_instantiate(dentry, inode); |
374 | return 0; | |
1da177e4 LT |
375 | } |
376 | ||
377 | STATIC int | |
416c6d5b | 378 | xfs_vn_unlink( |
1da177e4 LT |
379 | struct inode *dir, |
380 | struct dentry *dentry) | |
381 | { | |
556b8b16 | 382 | struct xfs_name name; |
1da177e4 LT |
383 | int error; |
384 | ||
fab8eef8 | 385 | xfs_dentry_to_name(&name, dentry); |
1da177e4 | 386 | |
2b0143b5 | 387 | error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry))); |
e5700704 CH |
388 | if (error) |
389 | return error; | |
390 | ||
391 | /* | |
392 | * With unlink, the VFS makes the dentry "negative": no inode, | |
393 | * but still hashed. This is incompatible with case-insensitive | |
394 | * mode, so invalidate (unhash) the dentry in CI-mode. | |
395 | */ | |
38c26bfd | 396 | if (xfs_has_asciici(XFS_M(dir->i_sb))) |
e5700704 CH |
397 | d_invalidate(dentry); |
398 | return 0; | |
1da177e4 LT |
399 | } |
400 | ||
401 | STATIC int | |
416c6d5b | 402 | xfs_vn_symlink( |
549c7297 CB |
403 | struct user_namespace *mnt_userns, |
404 | struct inode *dir, | |
405 | struct dentry *dentry, | |
406 | const char *symname) | |
1da177e4 | 407 | { |
3937be5b CH |
408 | struct inode *inode; |
409 | struct xfs_inode *cip = NULL; | |
556b8b16 | 410 | struct xfs_name name; |
1da177e4 | 411 | int error; |
576b1d67 | 412 | umode_t mode; |
1da177e4 | 413 | |
3e5daf05 | 414 | mode = S_IFLNK | |
ce3b0f8d | 415 | (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); |
fab8eef8 AG |
416 | error = xfs_dentry_mode_to_name(&name, dentry, mode); |
417 | if (unlikely(error)) | |
418 | goto out; | |
1da177e4 | 419 | |
f736d93d | 420 | error = xfs_symlink(mnt_userns, XFS_I(dir), &name, symname, mode, &cip); |
3937be5b CH |
421 | if (unlikely(error)) |
422 | goto out; | |
423 | ||
01651646 | 424 | inode = VFS_I(cip); |
3937be5b | 425 | |
70b589a3 | 426 | error = xfs_inode_init_security(inode, dir, &dentry->d_name); |
3937be5b CH |
427 | if (unlikely(error)) |
428 | goto out_cleanup_inode; | |
429 | ||
2b3d1d41 CH |
430 | xfs_setup_iops(cip); |
431 | ||
3937be5b | 432 | d_instantiate(dentry, inode); |
58c90473 | 433 | xfs_finish_inode_setup(cip); |
3937be5b CH |
434 | return 0; |
435 | ||
436 | out_cleanup_inode: | |
58c90473 | 437 | xfs_finish_inode_setup(cip); |
8f112e3b | 438 | xfs_cleanup_inode(dir, inode, dentry); |
44a8736b | 439 | xfs_irele(cip); |
3937be5b | 440 | out: |
2451337d | 441 | return error; |
1da177e4 LT |
442 | } |
443 | ||
1da177e4 | 444 | STATIC int |
416c6d5b | 445 | xfs_vn_rename( |
549c7297 CB |
446 | struct user_namespace *mnt_userns, |
447 | struct inode *odir, | |
448 | struct dentry *odentry, | |
449 | struct inode *ndir, | |
450 | struct dentry *ndentry, | |
451 | unsigned int flags) | |
1da177e4 | 452 | { |
2b0143b5 | 453 | struct inode *new_inode = d_inode(ndentry); |
d31a1825 | 454 | int omode = 0; |
fab8eef8 | 455 | int error; |
556b8b16 BN |
456 | struct xfs_name oname; |
457 | struct xfs_name nname; | |
1da177e4 | 458 | |
7dcf5c3e | 459 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
dbe1b5ca CM |
460 | return -EINVAL; |
461 | ||
d31a1825 CM |
462 | /* if we are exchanging files, we need to set i_mode of both files */ |
463 | if (flags & RENAME_EXCHANGE) | |
2b0143b5 | 464 | omode = d_inode(ndentry)->i_mode; |
d31a1825 | 465 | |
fab8eef8 AG |
466 | error = xfs_dentry_mode_to_name(&oname, odentry, omode); |
467 | if (omode && unlikely(error)) | |
468 | return error; | |
469 | ||
470 | error = xfs_dentry_mode_to_name(&nname, ndentry, | |
471 | d_inode(odentry)->i_mode); | |
472 | if (unlikely(error)) | |
473 | return error; | |
556b8b16 | 474 | |
f736d93d CH |
475 | return xfs_rename(mnt_userns, XFS_I(odir), &oname, |
476 | XFS_I(d_inode(odentry)), XFS_I(ndir), &nname, | |
d31a1825 | 477 | new_inode ? XFS_I(new_inode) : NULL, flags); |
1da177e4 LT |
478 | } |
479 | ||
480 | /* | |
481 | * careful here - this function can get called recursively, so | |
482 | * we need to be very careful about how much stack we use. | |
483 | * uio is kmalloced for this reason... | |
484 | */ | |
680baacb | 485 | STATIC const char * |
6b255391 | 486 | xfs_vn_get_link( |
1da177e4 | 487 | struct dentry *dentry, |
6b255391 | 488 | struct inode *inode, |
fceef393 | 489 | struct delayed_call *done) |
1da177e4 | 490 | { |
1da177e4 | 491 | char *link; |
804c83c3 | 492 | int error = -ENOMEM; |
1da177e4 | 493 | |
6b255391 AV |
494 | if (!dentry) |
495 | return ERR_PTR(-ECHILD); | |
496 | ||
6eb0b8df | 497 | link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); |
804c83c3 CH |
498 | if (!link) |
499 | goto out_err; | |
1da177e4 | 500 | |
2b0143b5 | 501 | error = xfs_readlink(XFS_I(d_inode(dentry)), link); |
804c83c3 CH |
502 | if (unlikely(error)) |
503 | goto out_kfree; | |
1da177e4 | 504 | |
fceef393 AV |
505 | set_delayed_call(done, kfree_link, link); |
506 | return link; | |
804c83c3 CH |
507 | |
508 | out_kfree: | |
509 | kfree(link); | |
510 | out_err: | |
680baacb | 511 | return ERR_PTR(error); |
1da177e4 LT |
512 | } |
513 | ||
dd2d535e CH |
514 | static uint32_t |
515 | xfs_stat_blksize( | |
516 | struct xfs_inode *ip) | |
517 | { | |
518 | struct xfs_mount *mp = ip->i_mount; | |
519 | ||
520 | /* | |
521 | * If the file blocks are being allocated from a realtime volume, then | |
522 | * always return the realtime extent size. | |
523 | */ | |
524 | if (XFS_IS_REALTIME_INODE(ip)) | |
a7bcb147 | 525 | return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip)); |
dd2d535e CH |
526 | |
527 | /* | |
528 | * Allow large block sizes to be reported to userspace programs if the | |
529 | * "largeio" mount option is used. | |
530 | * | |
531 | * If compatibility mode is specified, simply return the basic unit of | |
532 | * caching so that we don't get inefficient read/modify/write I/O from | |
533 | * user apps. Otherwise.... | |
534 | * | |
535 | * If the underlying volume is a stripe, then return the stripe width in | |
536 | * bytes as the recommended I/O size. It is not a stripe and we've set a | |
537 | * default buffered I/O size, return that, otherwise return the compat | |
538 | * default. | |
539 | */ | |
0560f31a | 540 | if (xfs_has_large_iosize(mp)) { |
dd2d535e | 541 | if (mp->m_swidth) |
a7bcb147 | 542 | return XFS_FSB_TO_B(mp, mp->m_swidth); |
0560f31a | 543 | if (xfs_has_allocsize(mp)) |
5da8a07c | 544 | return 1U << mp->m_allocsize_log; |
dd2d535e CH |
545 | } |
546 | ||
547 | return PAGE_SIZE; | |
548 | } | |
549 | ||
1da177e4 | 550 | STATIC int |
416c6d5b | 551 | xfs_vn_getattr( |
549c7297 | 552 | struct user_namespace *mnt_userns, |
a528d35e DH |
553 | const struct path *path, |
554 | struct kstat *stat, | |
555 | u32 request_mask, | |
556 | unsigned int query_flags) | |
1da177e4 | 557 | { |
a528d35e | 558 | struct inode *inode = d_inode(path->dentry); |
c43f4087 CH |
559 | struct xfs_inode *ip = XFS_I(inode); |
560 | struct xfs_mount *mp = ip->i_mount; | |
42b7cc11 CB |
561 | vfsuid_t vfsuid = i_uid_into_vfsuid(mnt_userns, inode); |
562 | vfsgid_t vfsgid = i_gid_into_vfsgid(mnt_userns, inode); | |
c43f4087 | 563 | |
cca28fb8 | 564 | trace_xfs_getattr(ip); |
c43f4087 | 565 | |
75c8c50f | 566 | if (xfs_is_shutdown(mp)) |
b474c7ae | 567 | return -EIO; |
c43f4087 CH |
568 | |
569 | stat->size = XFS_ISIZE(ip); | |
570 | stat->dev = inode->i_sb->s_dev; | |
c19b3b05 | 571 | stat->mode = inode->i_mode; |
54d7b5c1 | 572 | stat->nlink = inode->i_nlink; |
42b7cc11 CB |
573 | stat->uid = vfsuid_into_kuid(vfsuid); |
574 | stat->gid = vfsgid_into_kgid(vfsgid); | |
c43f4087 | 575 | stat->ino = ip->i_ino; |
c43f4087 | 576 | stat->atime = inode->i_atime; |
f9581b14 CH |
577 | stat->mtime = inode->i_mtime; |
578 | stat->ctime = inode->i_ctime; | |
6e73a545 | 579 | stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); |
c43f4087 | 580 | |
38c26bfd | 581 | if (xfs_has_v3inodes(mp)) { |
5f955f26 DW |
582 | if (request_mask & STATX_BTIME) { |
583 | stat->result_mask |= STATX_BTIME; | |
e98d5e88 | 584 | stat->btime = ip->i_crtime; |
5f955f26 DW |
585 | } |
586 | } | |
587 | ||
1b9598c8 LR |
588 | /* |
589 | * Note: If you add another clause to set an attribute flag, please | |
590 | * update attributes_mask below. | |
591 | */ | |
db07349d | 592 | if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) |
5f955f26 | 593 | stat->attributes |= STATX_ATTR_IMMUTABLE; |
db07349d | 594 | if (ip->i_diflags & XFS_DIFLAG_APPEND) |
5f955f26 | 595 | stat->attributes |= STATX_ATTR_APPEND; |
db07349d | 596 | if (ip->i_diflags & XFS_DIFLAG_NODUMP) |
5f955f26 | 597 | stat->attributes |= STATX_ATTR_NODUMP; |
c43f4087 | 598 | |
1b9598c8 LR |
599 | stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | |
600 | STATX_ATTR_APPEND | | |
601 | STATX_ATTR_NODUMP); | |
602 | ||
c43f4087 CH |
603 | switch (inode->i_mode & S_IFMT) { |
604 | case S_IFBLK: | |
605 | case S_IFCHR: | |
606 | stat->blksize = BLKDEV_IOSIZE; | |
66f36464 | 607 | stat->rdev = inode->i_rdev; |
c43f4087 | 608 | break; |
61a223df EB |
609 | case S_IFREG: |
610 | if (request_mask & STATX_DIOALIGN) { | |
611 | struct xfs_buftarg *target = xfs_inode_buftarg(ip); | |
612 | struct block_device *bdev = target->bt_bdev; | |
613 | ||
614 | stat->result_mask |= STATX_DIOALIGN; | |
615 | stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; | |
616 | stat->dio_offset_align = bdev_logical_block_size(bdev); | |
617 | } | |
618 | fallthrough; | |
c43f4087 | 619 | default: |
dd2d535e | 620 | stat->blksize = xfs_stat_blksize(ip); |
c43f4087 CH |
621 | stat->rdev = 0; |
622 | break; | |
69e23b9a | 623 | } |
c43f4087 CH |
624 | |
625 | return 0; | |
1da177e4 LT |
626 | } |
627 | ||
69bca807 JK |
628 | static int |
629 | xfs_vn_change_ok( | |
f736d93d CH |
630 | struct user_namespace *mnt_userns, |
631 | struct dentry *dentry, | |
632 | struct iattr *iattr) | |
69bca807 | 633 | { |
31051c85 | 634 | struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; |
69bca807 | 635 | |
2e973b2c | 636 | if (xfs_is_readonly(mp)) |
69bca807 JK |
637 | return -EROFS; |
638 | ||
75c8c50f | 639 | if (xfs_is_shutdown(mp)) |
69bca807 JK |
640 | return -EIO; |
641 | ||
f736d93d | 642 | return setattr_prepare(mnt_userns, dentry, iattr); |
69bca807 JK |
643 | } |
644 | ||
645 | /* | |
646 | * Set non-size attributes of an inode. | |
647 | * | |
648 | * Caution: The caller of this function is responsible for calling | |
31051c85 | 649 | * setattr_prepare() or otherwise verifying the change is fine. |
69bca807 | 650 | */ |
5d24ec4c | 651 | static int |
c4ed4243 | 652 | xfs_setattr_nonsize( |
f736d93d | 653 | struct user_namespace *mnt_userns, |
138060ba | 654 | struct dentry *dentry, |
c4ed4243 | 655 | struct xfs_inode *ip, |
5d24ec4c | 656 | struct iattr *iattr) |
c4ed4243 CH |
657 | { |
658 | xfs_mount_t *mp = ip->i_mount; | |
659 | struct inode *inode = VFS_I(ip); | |
660 | int mask = iattr->ia_valid; | |
661 | xfs_trans_t *tp; | |
662 | int error; | |
dd3b015d DW |
663 | kuid_t uid = GLOBAL_ROOT_UID; |
664 | kgid_t gid = GLOBAL_ROOT_GID; | |
c4ed4243 | 665 | struct xfs_dquot *udqp = NULL, *gdqp = NULL; |
dd3b015d | 666 | struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL; |
c4ed4243 | 667 | |
c4ed4243 CH |
668 | ASSERT((mask & ATTR_SIZE) == 0); |
669 | ||
670 | /* | |
671 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
672 | * before we start any other transactions. Trying to do this later | |
673 | * is messy. We don't care to take a readlock to look at the ids | |
674 | * in inode here, because we can't hold it across the trans_reserve. | |
675 | * If the IDs do change before we take the ilock, we're covered | |
676 | * because the i_*dquot fields will get updated anyway. | |
677 | */ | |
678 | if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { | |
679 | uint qflags = 0; | |
680 | ||
681 | if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { | |
b27c82e1 CB |
682 | uid = from_vfsuid(mnt_userns, i_user_ns(inode), |
683 | iattr->ia_vfsuid); | |
c4ed4243 CH |
684 | qflags |= XFS_QMOPT_UQUOTA; |
685 | } else { | |
7aab1b28 | 686 | uid = inode->i_uid; |
c4ed4243 CH |
687 | } |
688 | if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { | |
b27c82e1 CB |
689 | gid = from_vfsgid(mnt_userns, i_user_ns(inode), |
690 | iattr->ia_vfsgid); | |
c4ed4243 CH |
691 | qflags |= XFS_QMOPT_GQUOTA; |
692 | } else { | |
7aab1b28 | 693 | gid = inode->i_gid; |
c4ed4243 CH |
694 | } |
695 | ||
696 | /* | |
697 | * We take a reference when we initialize udqp and gdqp, | |
698 | * so it is important that we never blindly double trip on | |
699 | * the same variable. See xfs_create() for an example. | |
700 | */ | |
701 | ASSERT(udqp == NULL); | |
702 | ASSERT(gdqp == NULL); | |
ceaf603c | 703 | error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, |
7aab1b28 | 704 | qflags, &udqp, &gdqp, NULL); |
c4ed4243 CH |
705 | if (error) |
706 | return error; | |
707 | } | |
708 | ||
7317a03d | 709 | error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, |
eba0549b | 710 | has_capability_noaudit(current, CAP_FOWNER), &tp); |
c4ed4243 | 711 | if (error) |
253f4911 | 712 | goto out_dqrele; |
c4ed4243 | 713 | |
c4ed4243 | 714 | /* |
dd3b015d DW |
715 | * Register quota modifications in the transaction. Must be the owner |
716 | * or privileged. These IDs could have changed since we last looked at | |
717 | * them. But, we're assured that if the ownership did change while we | |
718 | * didn't have the inode locked, inode's dquot(s) would have changed | |
719 | * also. | |
c4ed4243 | 720 | */ |
35faf310 | 721 | if (XFS_IS_UQUOTA_ON(mp) && |
b27c82e1 | 722 | i_uid_needs_update(mnt_userns, iattr, inode)) { |
dd3b015d DW |
723 | ASSERT(udqp); |
724 | old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp); | |
725 | } | |
35faf310 | 726 | if (XFS_IS_GQUOTA_ON(mp) && |
b27c82e1 | 727 | i_gid_needs_update(mnt_userns, iattr, inode)) { |
dd3b015d DW |
728 | ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp)); |
729 | ASSERT(gdqp); | |
730 | old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp); | |
c4ed4243 CH |
731 | } |
732 | ||
e014f37d | 733 | setattr_copy(mnt_userns, inode, iattr); |
c4ed4243 CH |
734 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
735 | ||
ff6d6af2 | 736 | XFS_STATS_INC(mp, xs_ig_attrchg); |
c4ed4243 | 737 | |
0560f31a | 738 | if (xfs_has_wsync(mp)) |
c4ed4243 | 739 | xfs_trans_set_sync(tp); |
70393313 | 740 | error = xfs_trans_commit(tp); |
c4ed4243 | 741 | |
c4ed4243 CH |
742 | /* |
743 | * Release any dquot(s) the inode had kept before chown. | |
744 | */ | |
dd3b015d DW |
745 | xfs_qm_dqrele(old_udqp); |
746 | xfs_qm_dqrele(old_gdqp); | |
c4ed4243 CH |
747 | xfs_qm_dqrele(udqp); |
748 | xfs_qm_dqrele(gdqp); | |
749 | ||
750 | if (error) | |
b474c7ae | 751 | return error; |
c4ed4243 CH |
752 | |
753 | /* | |
754 | * XXX(hch): Updating the ACL entries is not atomic vs the i_mode | |
755 | * update. We could avoid this with linked transactions | |
756 | * and passing down the transaction pointer all the way | |
757 | * to attr_set. No previous user of the generic | |
758 | * Posix ACL code seems to care about this issue either. | |
759 | */ | |
5d24ec4c | 760 | if (mask & ATTR_MODE) { |
138060ba | 761 | error = posix_acl_chmod(mnt_userns, dentry, inode->i_mode); |
c4ed4243 | 762 | if (error) |
b474c7ae | 763 | return error; |
c4ed4243 CH |
764 | } |
765 | ||
766 | return 0; | |
767 | ||
253f4911 | 768 | out_dqrele: |
c4ed4243 CH |
769 | xfs_qm_dqrele(udqp); |
770 | xfs_qm_dqrele(gdqp); | |
771 | return error; | |
772 | } | |
773 | ||
774 | /* | |
775 | * Truncate file. Must have write permission and not be a directory. | |
69bca807 JK |
776 | * |
777 | * Caution: The caller of this function is responsible for calling | |
31051c85 | 778 | * setattr_prepare() or otherwise verifying the change is fine. |
c4ed4243 | 779 | */ |
7bf7a193 | 780 | STATIC int |
c4ed4243 | 781 | xfs_setattr_size( |
f736d93d | 782 | struct user_namespace *mnt_userns, |
138060ba | 783 | struct dentry *dentry, |
c4ed4243 | 784 | struct xfs_inode *ip, |
76ca4c23 | 785 | struct iattr *iattr) |
c4ed4243 CH |
786 | { |
787 | struct xfs_mount *mp = ip->i_mount; | |
788 | struct inode *inode = VFS_I(ip); | |
673e8e59 | 789 | xfs_off_t oldsize, newsize; |
c4ed4243 CH |
790 | struct xfs_trans *tp; |
791 | int error; | |
f38996f5 | 792 | uint lock_flags = 0; |
5885ebda | 793 | bool did_zeroing = false; |
c4ed4243 | 794 | |
76ca4c23 | 795 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
e8e9ad42 | 796 | ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); |
c19b3b05 | 797 | ASSERT(S_ISREG(inode->i_mode)); |
fe60a8a0 | 798 | ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| |
88a9e03b | 799 | ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); |
c4ed4243 | 800 | |
ce7ae151 | 801 | oldsize = inode->i_size; |
673e8e59 CH |
802 | newsize = iattr->ia_size; |
803 | ||
c4ed4243 CH |
804 | /* |
805 | * Short circuit the truncate case for zero length files. | |
806 | */ | |
daf83964 | 807 | if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { |
fe60a8a0 | 808 | if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) |
76ca4c23 | 809 | return 0; |
681b1200 CH |
810 | |
811 | /* | |
812 | * Use the regular setattr path to update the timestamps. | |
813 | */ | |
681b1200 | 814 | iattr->ia_valid &= ~ATTR_SIZE; |
138060ba | 815 | return xfs_setattr_nonsize(mnt_userns, dentry, ip, iattr); |
c4ed4243 CH |
816 | } |
817 | ||
818 | /* | |
819 | * Make sure that the dquots are attached to the inode. | |
820 | */ | |
c14cfcca | 821 | error = xfs_qm_dqattach(ip); |
c4ed4243 | 822 | if (error) |
76ca4c23 | 823 | return error; |
c4ed4243 | 824 | |
f0c6bcba CH |
825 | /* |
826 | * Wait for all direct I/O to complete. | |
827 | */ | |
828 | inode_dio_wait(inode); | |
829 | ||
c4ed4243 | 830 | /* |
5885ebda DC |
831 | * File data changes must be complete before we start the transaction to |
832 | * modify the inode. This needs to be done before joining the inode to | |
833 | * the transaction because the inode cannot be unlocked once it is a | |
834 | * part of the transaction. | |
835 | * | |
f0c6bcba CH |
836 | * Start with zeroing any data beyond EOF that we may expose on file |
837 | * extension, or zeroing out the rest of the block on a downward | |
838 | * truncate. | |
c4ed4243 | 839 | */ |
673e8e59 | 840 | if (newsize > oldsize) { |
f5c54717 | 841 | trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); |
f1ba5faf SR |
842 | error = xfs_zero_range(ip, oldsize, newsize - oldsize, |
843 | &did_zeroing); | |
f0c6bcba | 844 | } else { |
869ae85d BF |
845 | /* |
846 | * iomap won't detect a dirty page over an unwritten block (or a | |
847 | * cow block over a hole) and subsequently skips zeroing the | |
848 | * newly post-EOF portion of the page. Flush the new EOF to | |
849 | * convert the block before the pagecache truncate. | |
850 | */ | |
851 | error = filemap_write_and_wait_range(inode->i_mapping, newsize, | |
852 | newsize); | |
853 | if (error) | |
854 | return error; | |
f1ba5faf | 855 | error = xfs_truncate_page(ip, newsize, &did_zeroing); |
c4ed4243 | 856 | } |
c4ed4243 | 857 | |
f0c6bcba CH |
858 | if (error) |
859 | return error; | |
860 | ||
49abc3a8 | 861 | /* |
0f9160b4 DC |
862 | * We've already locked out new page faults, so now we can safely remove |
863 | * pages from the page cache knowing they won't get refaulted until we | |
864 | * drop the XFS_MMAP_EXCL lock after the extent manipulations are | |
865 | * complete. The truncate_setsize() call also cleans partial EOF page | |
866 | * PTEs on extending truncates and hence ensures sub-page block size | |
867 | * filesystems are correctly handled, too. | |
49abc3a8 | 868 | * |
0f9160b4 DC |
869 | * We have to do all the page cache truncate work outside the |
870 | * transaction context as the "lock" order is page lock->log space | |
871 | * reservation as defined by extent allocation in the writeback path. | |
253f4911 | 872 | * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but |
0f9160b4 DC |
873 | * having already truncated the in-memory version of the file (i.e. made |
874 | * user visible changes). There's not much we can do about this, except | |
875 | * to hope that the caller sees ENOMEM and retries the truncate | |
876 | * operation. | |
350976ae EG |
877 | * |
878 | * And we update in-core i_size and truncate page cache beyond newsize | |
13d2c10b CH |
879 | * before writeback the [i_disk_size, newsize] range, so we're |
880 | * guaranteed not to write stale data past the new EOF on truncate down. | |
49abc3a8 | 881 | */ |
49abc3a8 | 882 | truncate_setsize(inode, newsize); |
c4ed4243 | 883 | |
350976ae EG |
884 | /* |
885 | * We are going to log the inode size change in this transaction so | |
886 | * any previous writes that are beyond the on disk EOF and the new | |
887 | * EOF that have not been written out need to be written here. If we | |
888 | * do not write the data out, we expose ourselves to the null files | |
889 | * problem. Note that this includes any block zeroing we did above; | |
890 | * otherwise those blocks may not be zeroed after a crash. | |
891 | */ | |
892 | if (did_zeroing || | |
13d2c10b | 893 | (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { |
350976ae | 894 | error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
13d2c10b | 895 | ip->i_disk_size, newsize - 1); |
350976ae EG |
896 | if (error) |
897 | return error; | |
898 | } | |
899 | ||
253f4911 | 900 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); |
c4ed4243 | 901 | if (error) |
253f4911 | 902 | return error; |
c4ed4243 | 903 | |
c4ed4243 | 904 | lock_flags |= XFS_ILOCK_EXCL; |
c4ed4243 | 905 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
ddc3415a | 906 | xfs_trans_ijoin(tp, ip, 0); |
c4ed4243 CH |
907 | |
908 | /* | |
909 | * Only change the c/mtime if we are changing the size or we are | |
910 | * explicitly asked to change it. This handles the semantic difference | |
911 | * between truncate() and ftruncate() as implemented in the VFS. | |
912 | * | |
913 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
914 | * special case where we need to update the times despite not having | |
915 | * these flags set. For all other operations the VFS set these flags | |
916 | * explicitly if it wants a timestamp update. | |
917 | */ | |
fe60a8a0 CH |
918 | if (newsize != oldsize && |
919 | !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { | |
c4ed4243 | 920 | iattr->ia_ctime = iattr->ia_mtime = |
c2050a45 | 921 | current_time(inode); |
fe60a8a0 | 922 | iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; |
c4ed4243 CH |
923 | } |
924 | ||
673e8e59 CH |
925 | /* |
926 | * The first thing we do is set the size to new_size permanently on | |
927 | * disk. This way we don't have to worry about anyone ever being able | |
928 | * to look at the data being freed even in the face of a crash. | |
929 | * What we're getting around here is the case where we free a block, it | |
930 | * is allocated to another file, it is written to, and then we crash. | |
931 | * If the new data gets written to the file but the log buffers | |
932 | * containing the free and reallocation don't, then we'd end up with | |
933 | * garbage in the blocks being freed. As long as we make the new size | |
934 | * permanent before actually freeing any blocks it doesn't matter if | |
935 | * they get written to. | |
936 | */ | |
13d2c10b | 937 | ip->i_disk_size = newsize; |
673e8e59 CH |
938 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
939 | ||
940 | if (newsize <= oldsize) { | |
941 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); | |
c4ed4243 | 942 | if (error) |
4906e215 | 943 | goto out_trans_cancel; |
c4ed4243 CH |
944 | |
945 | /* | |
946 | * Truncated "down", so we're removing references to old data | |
947 | * here - if we delay flushing for a long time, we expose | |
948 | * ourselves unduly to the notorious NULL files problem. So, | |
949 | * we mark this inode and flush it when the file is closed, | |
950 | * and do not wait the usual (long) time for writeout. | |
951 | */ | |
952 | xfs_iflags_set(ip, XFS_ITRUNCATED); | |
27b52867 BF |
953 | |
954 | /* A truncate down always removes post-EOF blocks. */ | |
955 | xfs_inode_clear_eofblocks_tag(ip); | |
c4ed4243 CH |
956 | } |
957 | ||
e014f37d DW |
958 | ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID))); |
959 | setattr_copy(mnt_userns, inode, iattr); | |
c4ed4243 CH |
960 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
961 | ||
ff6d6af2 | 962 | XFS_STATS_INC(mp, xs_ig_attrchg); |
c4ed4243 | 963 | |
0560f31a | 964 | if (xfs_has_wsync(mp)) |
c4ed4243 CH |
965 | xfs_trans_set_sync(tp); |
966 | ||
70393313 | 967 | error = xfs_trans_commit(tp); |
c4ed4243 CH |
968 | out_unlock: |
969 | if (lock_flags) | |
970 | xfs_iunlock(ip, lock_flags); | |
971 | return error; | |
972 | ||
c4ed4243 | 973 | out_trans_cancel: |
4906e215 | 974 | xfs_trans_cancel(tp); |
c4ed4243 CH |
975 | goto out_unlock; |
976 | } | |
977 | ||
69bca807 JK |
978 | int |
979 | xfs_vn_setattr_size( | |
f736d93d | 980 | struct user_namespace *mnt_userns, |
69bca807 JK |
981 | struct dentry *dentry, |
982 | struct iattr *iattr) | |
983 | { | |
984 | struct xfs_inode *ip = XFS_I(d_inode(dentry)); | |
985 | int error; | |
986 | ||
987 | trace_xfs_setattr(ip); | |
988 | ||
f736d93d | 989 | error = xfs_vn_change_ok(mnt_userns, dentry, iattr); |
69bca807 JK |
990 | if (error) |
991 | return error; | |
138060ba | 992 | return xfs_setattr_size(mnt_userns, dentry, ip, iattr); |
69bca807 JK |
993 | } |
994 | ||
1da177e4 | 995 | STATIC int |
416c6d5b | 996 | xfs_vn_setattr( |
549c7297 | 997 | struct user_namespace *mnt_userns, |
76ca4c23 CH |
998 | struct dentry *dentry, |
999 | struct iattr *iattr) | |
1da177e4 | 1000 | { |
26f88363 CH |
1001 | struct inode *inode = d_inode(dentry); |
1002 | struct xfs_inode *ip = XFS_I(inode); | |
76ca4c23 CH |
1003 | int error; |
1004 | ||
1005 | if (iattr->ia_valid & ATTR_SIZE) { | |
c63a8eae | 1006 | uint iolock; |
781355c6 | 1007 | |
c63a8eae DW |
1008 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
1009 | iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; | |
781355c6 | 1010 | |
69eb5fa1 | 1011 | error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); |
c63a8eae DW |
1012 | if (error) { |
1013 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); | |
65523218 | 1014 | return error; |
c63a8eae | 1015 | } |
e8e9ad42 | 1016 | |
f736d93d | 1017 | error = xfs_vn_setattr_size(mnt_userns, dentry, iattr); |
65523218 | 1018 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
76ca4c23 | 1019 | } else { |
26f88363 CH |
1020 | trace_xfs_setattr(ip); |
1021 | ||
f736d93d | 1022 | error = xfs_vn_change_ok(mnt_userns, dentry, iattr); |
26f88363 | 1023 | if (!error) |
138060ba | 1024 | error = xfs_setattr_nonsize(mnt_userns, dentry, ip, iattr); |
76ca4c23 CH |
1025 | } |
1026 | ||
2451337d | 1027 | return error; |
1da177e4 LT |
1028 | } |
1029 | ||
69ff2826 CH |
1030 | STATIC int |
1031 | xfs_vn_update_time( | |
1032 | struct inode *inode, | |
95582b00 | 1033 | struct timespec64 *now, |
69ff2826 CH |
1034 | int flags) |
1035 | { | |
1036 | struct xfs_inode *ip = XFS_I(inode); | |
1037 | struct xfs_mount *mp = ip->i_mount; | |
c3b1b131 | 1038 | int log_flags = XFS_ILOG_TIMESTAMP; |
69ff2826 CH |
1039 | struct xfs_trans *tp; |
1040 | int error; | |
1041 | ||
1042 | trace_xfs_update_time(ip); | |
1043 | ||
c3b1b131 CH |
1044 | if (inode->i_sb->s_flags & SB_LAZYTIME) { |
1045 | if (!((flags & S_VERSION) && | |
1046 | inode_maybe_inc_iversion(inode, false))) | |
1047 | return generic_update_time(inode, now, flags); | |
1048 | ||
1049 | /* Capture the iversion update that just occurred */ | |
1050 | log_flags |= XFS_ILOG_CORE; | |
1051 | } | |
1052 | ||
253f4911 CH |
1053 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
1054 | if (error) | |
2451337d | 1055 | return error; |
69ff2826 CH |
1056 | |
1057 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3987848c | 1058 | if (flags & S_CTIME) |
69ff2826 | 1059 | inode->i_ctime = *now; |
3987848c | 1060 | if (flags & S_MTIME) |
69ff2826 | 1061 | inode->i_mtime = *now; |
3987848c | 1062 | if (flags & S_ATIME) |
69ff2826 | 1063 | inode->i_atime = *now; |
3987848c | 1064 | |
69ff2826 | 1065 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
c3b1b131 | 1066 | xfs_trans_log_inode(tp, ip, log_flags); |
70393313 | 1067 | return xfs_trans_commit(tp); |
69ff2826 CH |
1068 | } |
1069 | ||
f35642e2 ES |
1070 | STATIC int |
1071 | xfs_vn_fiemap( | |
1072 | struct inode *inode, | |
1073 | struct fiemap_extent_info *fieinfo, | |
1074 | u64 start, | |
1075 | u64 length) | |
1076 | { | |
f35642e2 ES |
1077 | int error; |
1078 | ||
d2bb140e | 1079 | xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); |
1d4795e7 CH |
1080 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { |
1081 | fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; | |
1082 | error = iomap_fiemap(inode, fieinfo, start, length, | |
1083 | &xfs_xattr_iomap_ops); | |
1084 | } else { | |
1085 | error = iomap_fiemap(inode, fieinfo, start, length, | |
690c2a38 | 1086 | &xfs_read_iomap_ops); |
1d4795e7 | 1087 | } |
d2bb140e | 1088 | xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); |
f35642e2 | 1089 | |
d2bb140e | 1090 | return error; |
f35642e2 ES |
1091 | } |
1092 | ||
99b6436b ZYW |
1093 | STATIC int |
1094 | xfs_vn_tmpfile( | |
549c7297 CB |
1095 | struct user_namespace *mnt_userns, |
1096 | struct inode *dir, | |
863f144f | 1097 | struct file *file, |
549c7297 | 1098 | umode_t mode) |
99b6436b | 1099 | { |
863f144f MS |
1100 | int err = xfs_generic_create(mnt_userns, dir, file->f_path.dentry, mode, 0, file); |
1101 | ||
1102 | return finish_open_simple(file, err); | |
99b6436b ZYW |
1103 | } |
1104 | ||
41be8bed | 1105 | static const struct inode_operations xfs_inode_operations = { |
cac2f8b8 | 1106 | .get_inode_acl = xfs_get_acl, |
2401dc29 | 1107 | .set_acl = xfs_set_acl, |
416c6d5b NS |
1108 | .getattr = xfs_vn_getattr, |
1109 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1110 | .listxattr = xfs_vn_listxattr, |
f35642e2 | 1111 | .fiemap = xfs_vn_fiemap, |
69ff2826 | 1112 | .update_time = xfs_vn_update_time, |
9fefd5db MS |
1113 | .fileattr_get = xfs_fileattr_get, |
1114 | .fileattr_set = xfs_fileattr_set, | |
1da177e4 LT |
1115 | }; |
1116 | ||
41be8bed | 1117 | static const struct inode_operations xfs_dir_inode_operations = { |
416c6d5b NS |
1118 | .create = xfs_vn_create, |
1119 | .lookup = xfs_vn_lookup, | |
1120 | .link = xfs_vn_link, | |
1121 | .unlink = xfs_vn_unlink, | |
1122 | .symlink = xfs_vn_symlink, | |
1123 | .mkdir = xfs_vn_mkdir, | |
8f112e3b CH |
1124 | /* |
1125 | * Yes, XFS uses the same method for rmdir and unlink. | |
1126 | * | |
1127 | * There are some subtile differences deeper in the code, | |
1128 | * but we use S_ISDIR to check for those. | |
1129 | */ | |
1130 | .rmdir = xfs_vn_unlink, | |
416c6d5b | 1131 | .mknod = xfs_vn_mknod, |
2773bf00 | 1132 | .rename = xfs_vn_rename, |
cac2f8b8 | 1133 | .get_inode_acl = xfs_get_acl, |
2401dc29 | 1134 | .set_acl = xfs_set_acl, |
416c6d5b NS |
1135 | .getattr = xfs_vn_getattr, |
1136 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1137 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1138 | .update_time = xfs_vn_update_time, |
99b6436b | 1139 | .tmpfile = xfs_vn_tmpfile, |
9fefd5db MS |
1140 | .fileattr_get = xfs_fileattr_get, |
1141 | .fileattr_set = xfs_fileattr_set, | |
1da177e4 LT |
1142 | }; |
1143 | ||
41be8bed | 1144 | static const struct inode_operations xfs_dir_ci_inode_operations = { |
384f3ced BN |
1145 | .create = xfs_vn_create, |
1146 | .lookup = xfs_vn_ci_lookup, | |
1147 | .link = xfs_vn_link, | |
1148 | .unlink = xfs_vn_unlink, | |
1149 | .symlink = xfs_vn_symlink, | |
1150 | .mkdir = xfs_vn_mkdir, | |
8f112e3b CH |
1151 | /* |
1152 | * Yes, XFS uses the same method for rmdir and unlink. | |
1153 | * | |
1154 | * There are some subtile differences deeper in the code, | |
1155 | * but we use S_ISDIR to check for those. | |
1156 | */ | |
1157 | .rmdir = xfs_vn_unlink, | |
384f3ced | 1158 | .mknod = xfs_vn_mknod, |
2773bf00 | 1159 | .rename = xfs_vn_rename, |
cac2f8b8 | 1160 | .get_inode_acl = xfs_get_acl, |
2401dc29 | 1161 | .set_acl = xfs_set_acl, |
384f3ced BN |
1162 | .getattr = xfs_vn_getattr, |
1163 | .setattr = xfs_vn_setattr, | |
384f3ced | 1164 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1165 | .update_time = xfs_vn_update_time, |
99b6436b | 1166 | .tmpfile = xfs_vn_tmpfile, |
9fefd5db MS |
1167 | .fileattr_get = xfs_fileattr_get, |
1168 | .fileattr_set = xfs_fileattr_set, | |
384f3ced BN |
1169 | }; |
1170 | ||
41be8bed | 1171 | static const struct inode_operations xfs_symlink_inode_operations = { |
6b255391 | 1172 | .get_link = xfs_vn_get_link, |
416c6d5b NS |
1173 | .getattr = xfs_vn_getattr, |
1174 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1175 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1176 | .update_time = xfs_vn_update_time, |
1da177e4 | 1177 | }; |
41be8bed | 1178 | |
ba23cba9 DW |
1179 | /* Figure out if this file actually supports DAX. */ |
1180 | static bool | |
1181 | xfs_inode_supports_dax( | |
1182 | struct xfs_inode *ip) | |
1183 | { | |
1184 | struct xfs_mount *mp = ip->i_mount; | |
1185 | ||
32dbc565 IW |
1186 | /* Only supported on regular files. */ |
1187 | if (!S_ISREG(VFS_I(ip)->i_mode)) | |
ba23cba9 DW |
1188 | return false; |
1189 | ||
ba23cba9 DW |
1190 | /* Block size must match page size */ |
1191 | if (mp->m_sb.sb_blocksize != PAGE_SIZE) | |
1192 | return false; | |
1193 | ||
1194 | /* Device has to support DAX too. */ | |
30fa529e | 1195 | return xfs_inode_buftarg(ip)->bt_daxdev != NULL; |
ba23cba9 DW |
1196 | } |
1197 | ||
32dbc565 IW |
1198 | static bool |
1199 | xfs_inode_should_enable_dax( | |
1200 | struct xfs_inode *ip) | |
1201 | { | |
1202 | if (!IS_ENABLED(CONFIG_FS_DAX)) | |
1203 | return false; | |
0560f31a | 1204 | if (xfs_has_dax_never(ip->i_mount)) |
32dbc565 IW |
1205 | return false; |
1206 | if (!xfs_inode_supports_dax(ip)) | |
1207 | return false; | |
0560f31a | 1208 | if (xfs_has_dax_always(ip->i_mount)) |
32dbc565 | 1209 | return true; |
3e09ab8f | 1210 | if (ip->i_diflags2 & XFS_DIFLAG2_DAX) |
32dbc565 IW |
1211 | return true; |
1212 | return false; | |
1213 | } | |
1214 | ||
840d493d | 1215 | void |
41be8bed | 1216 | xfs_diflags_to_iflags( |
840d493d IW |
1217 | struct xfs_inode *ip, |
1218 | bool init) | |
41be8bed | 1219 | { |
840d493d IW |
1220 | struct inode *inode = VFS_I(ip); |
1221 | unsigned int xflags = xfs_ip2xflags(ip); | |
1222 | unsigned int flags = 0; | |
1223 | ||
1224 | ASSERT(!(IS_DAX(inode) && init)); | |
1225 | ||
1226 | if (xflags & FS_XFLAG_IMMUTABLE) | |
1227 | flags |= S_IMMUTABLE; | |
1228 | if (xflags & FS_XFLAG_APPEND) | |
1229 | flags |= S_APPEND; | |
1230 | if (xflags & FS_XFLAG_SYNC) | |
1231 | flags |= S_SYNC; | |
1232 | if (xflags & FS_XFLAG_NOATIME) | |
1233 | flags |= S_NOATIME; | |
1234 | if (init && xfs_inode_should_enable_dax(ip)) | |
1235 | flags |= S_DAX; | |
1236 | ||
1237 | /* | |
1238 | * S_DAX can only be set during inode initialization and is never set by | |
1239 | * the VFS, so we cannot mask off S_DAX in i_flags. | |
1240 | */ | |
1241 | inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME); | |
1242 | inode->i_flags |= flags; | |
41be8bed CH |
1243 | } |
1244 | ||
1245 | /* | |
2b3d1d41 | 1246 | * Initialize the Linux inode. |
bf904248 | 1247 | * |
58c90473 | 1248 | * When reading existing inodes from disk this is called directly from xfs_iget, |
132c460e YX |
1249 | * when creating a new inode it is called from xfs_init_new_inode after setting |
1250 | * up the inode. These callers have different criteria for clearing XFS_INEW, so | |
1251 | * leave it up to the caller to deal with unlocking the inode appropriately. | |
41be8bed CH |
1252 | */ |
1253 | void | |
1254 | xfs_setup_inode( | |
1255 | struct xfs_inode *ip) | |
1256 | { | |
bf904248 | 1257 | struct inode *inode = &ip->i_vnode; |
ad22c7a0 | 1258 | gfp_t gfp_mask; |
bf904248 DC |
1259 | |
1260 | inode->i_ino = ip->i_ino; | |
f38a032b | 1261 | inode->i_state |= I_NEW; |
646ec461 CH |
1262 | |
1263 | inode_sb_list_add(inode); | |
c6f6cd06 | 1264 | /* make the inode look hashed for the writeback code */ |
5bef9151 | 1265 | inode_fake_hash(inode); |
41be8bed | 1266 | |
13d2c10b | 1267 | i_size_write(inode, ip->i_disk_size); |
840d493d | 1268 | xfs_diflags_to_iflags(ip, true); |
41be8bed | 1269 | |
2b3d1d41 | 1270 | if (S_ISDIR(inode->i_mode)) { |
ef215e39 DC |
1271 | /* |
1272 | * We set the i_rwsem class here to avoid potential races with | |
1273 | * lockdep_annotate_inode_mutex_key() reinitialising the lock | |
1274 | * after a filehandle lookup has already found the inode in | |
1275 | * cache before it has been unlocked via unlock_new_inode(). | |
1276 | */ | |
1277 | lockdep_set_class(&inode->i_rwsem, | |
1278 | &inode->i_sb->s_type->i_mutex_dir_key); | |
93a8614e | 1279 | lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class); |
2b3d1d41 | 1280 | } else { |
2b3d1d41 | 1281 | lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class); |
41be8bed CH |
1282 | } |
1283 | ||
ad22c7a0 DC |
1284 | /* |
1285 | * Ensure all page cache allocations are done from GFP_NOFS context to | |
1286 | * prevent direct reclaim recursion back into the filesystem and blowing | |
1287 | * stacks or deadlocking. | |
1288 | */ | |
1289 | gfp_mask = mapping_gfp_mask(inode->i_mapping); | |
1290 | mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); | |
1291 | ||
510792ee CH |
1292 | /* |
1293 | * If there is no attribute fork no ACL can exist on this inode, | |
1294 | * and it can't have any file capabilities attached to it either. | |
1295 | */ | |
932b42c6 | 1296 | if (!xfs_inode_has_attr_fork(ip)) { |
510792ee | 1297 | inode_has_no_xattr(inode); |
6311b108 | 1298 | cache_no_acl(inode); |
510792ee | 1299 | } |
41be8bed | 1300 | } |
2b3d1d41 CH |
1301 | |
1302 | void | |
1303 | xfs_setup_iops( | |
1304 | struct xfs_inode *ip) | |
1305 | { | |
1306 | struct inode *inode = &ip->i_vnode; | |
1307 | ||
41be8bed CH |
1308 | switch (inode->i_mode & S_IFMT) { |
1309 | case S_IFREG: | |
1310 | inode->i_op = &xfs_inode_operations; | |
1311 | inode->i_fop = &xfs_file_operations; | |
6e2608df DW |
1312 | if (IS_DAX(inode)) |
1313 | inode->i_mapping->a_ops = &xfs_dax_aops; | |
1314 | else | |
1315 | inode->i_mapping->a_ops = &xfs_address_space_operations; | |
41be8bed CH |
1316 | break; |
1317 | case S_IFDIR: | |
38c26bfd | 1318 | if (xfs_has_asciici(XFS_M(inode->i_sb))) |
41be8bed CH |
1319 | inode->i_op = &xfs_dir_ci_inode_operations; |
1320 | else | |
1321 | inode->i_op = &xfs_dir_inode_operations; | |
1322 | inode->i_fop = &xfs_dir_file_operations; | |
1323 | break; | |
1324 | case S_IFLNK: | |
7b7820b8 | 1325 | inode->i_op = &xfs_symlink_inode_operations; |
41be8bed CH |
1326 | break; |
1327 | default: | |
1328 | inode->i_op = &xfs_inode_operations; | |
1329 | init_special_inode(inode, inode->i_mode, inode->i_rdev); | |
1330 | break; | |
1331 | } | |
41be8bed | 1332 | } |