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 LT |
24 | |
25 | #include <linux/xattr.h> | |
ef14f0c1 | 26 | #include <linux/posix_acl.h> |
446ada4a | 27 | #include <linux/security.h> |
c3b1b131 | 28 | #include <linux/iversion.h> |
1da177e4 | 29 | |
93a8614e DC |
30 | /* |
31 | * Directories have different lock order w.r.t. mmap_sem compared to regular | |
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 | |
35 | * mmap_sem is taken during the page fault, and then we lock the ilock to do | |
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++) { | |
2451337d | 53 | error = xfs_attr_set(ip, xattr->name, xattr->value, |
a5a14de2 | 54 | xattr->value_len, ATTR_SECURE); |
9d8f13ba MZ |
55 | if (error < 0) |
56 | break; | |
57 | } | |
58 | return error; | |
59 | } | |
60 | ||
446ada4a NS |
61 | /* |
62 | * Hook in SELinux. This is not quite correct yet, what we really need | |
63 | * here (as we do for default ACLs) is a mechanism by which creation of | |
64 | * these attrs can be journalled at inode creation time (along with the | |
65 | * inode, of course, such that log replay can't cause these to be lost). | |
66 | */ | |
9d8f13ba | 67 | |
446ada4a | 68 | STATIC int |
416c6d5b | 69 | xfs_init_security( |
af048193 | 70 | struct inode *inode, |
2a7dba39 EP |
71 | struct inode *dir, |
72 | const struct qstr *qstr) | |
446ada4a | 73 | { |
2451337d | 74 | return security_inode_init_security(inode, dir, qstr, |
a5a14de2 | 75 | &xfs_initxattrs, NULL); |
446ada4a NS |
76 | } |
77 | ||
556b8b16 BN |
78 | static void |
79 | xfs_dentry_to_name( | |
fab8eef8 AG |
80 | struct xfs_name *namep, |
81 | struct dentry *dentry) | |
82 | { | |
83 | namep->name = dentry->d_name.name; | |
84 | namep->len = dentry->d_name.len; | |
85 | namep->type = XFS_DIR3_FT_UNKNOWN; | |
86 | } | |
87 | ||
88 | static int | |
89 | xfs_dentry_mode_to_name( | |
556b8b16 | 90 | struct xfs_name *namep, |
0cb97766 DC |
91 | struct dentry *dentry, |
92 | int mode) | |
556b8b16 BN |
93 | { |
94 | namep->name = dentry->d_name.name; | |
95 | namep->len = dentry->d_name.len; | |
1fc4d33f | 96 | namep->type = xfs_mode_to_ftype(mode); |
fab8eef8 AG |
97 | |
98 | if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN)) | |
99 | return -EFSCORRUPTED; | |
100 | ||
101 | return 0; | |
556b8b16 BN |
102 | } |
103 | ||
7989cb8e | 104 | STATIC void |
416c6d5b | 105 | xfs_cleanup_inode( |
739bfb2a | 106 | struct inode *dir, |
af048193 | 107 | struct inode *inode, |
8f112e3b | 108 | struct dentry *dentry) |
3a69c7dc | 109 | { |
556b8b16 | 110 | struct xfs_name teardown; |
3a69c7dc YL |
111 | |
112 | /* Oh, the horror. | |
220b5284 | 113 | * If we can't add the ACL or we fail in |
416c6d5b | 114 | * xfs_init_security we must back out. |
3a69c7dc YL |
115 | * ENOSPC can hit here, among other things. |
116 | */ | |
fab8eef8 | 117 | xfs_dentry_to_name(&teardown, dentry); |
3a69c7dc | 118 | |
8f112e3b | 119 | xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); |
3a69c7dc YL |
120 | } |
121 | ||
1da177e4 | 122 | STATIC int |
d540e43b | 123 | xfs_generic_create( |
1da177e4 LT |
124 | struct inode *dir, |
125 | struct dentry *dentry, | |
1a67aafb | 126 | umode_t mode, |
d540e43b BF |
127 | dev_t rdev, |
128 | bool tmpfile) /* unnamed file */ | |
1da177e4 | 129 | { |
db0bb7ba | 130 | struct inode *inode; |
979ebab1 | 131 | struct xfs_inode *ip = NULL; |
2401dc29 | 132 | struct posix_acl *default_acl, *acl; |
556b8b16 | 133 | struct xfs_name name; |
1da177e4 LT |
134 | int error; |
135 | ||
136 | /* | |
137 | * Irix uses Missed'em'V split, but doesn't want to see | |
138 | * the upper 5 bits of (14bit) major. | |
139 | */ | |
517b5e8c CH |
140 | if (S_ISCHR(mode) || S_ISBLK(mode)) { |
141 | if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff)) | |
142 | return -EINVAL; | |
517b5e8c CH |
143 | } else { |
144 | rdev = 0; | |
145 | } | |
1da177e4 | 146 | |
2401dc29 CH |
147 | error = posix_acl_create(dir, &mode, &default_acl, &acl); |
148 | if (error) | |
149 | return error; | |
1da177e4 | 150 | |
fab8eef8 AG |
151 | /* Verify mode is valid also for tmpfile case */ |
152 | error = xfs_dentry_mode_to_name(&name, dentry, mode); | |
153 | if (unlikely(error)) | |
154 | goto out_free_acl; | |
155 | ||
d540e43b | 156 | if (!tmpfile) { |
d540e43b BF |
157 | error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip); |
158 | } else { | |
a1f69417 | 159 | error = xfs_create_tmpfile(XFS_I(dir), mode, &ip); |
d540e43b | 160 | } |
db0bb7ba CH |
161 | if (unlikely(error)) |
162 | goto out_free_acl; | |
446ada4a | 163 | |
01651646 | 164 | inode = VFS_I(ip); |
979ebab1 | 165 | |
2a7dba39 | 166 | error = xfs_init_security(inode, dir, &dentry->d_name); |
db0bb7ba CH |
167 | if (unlikely(error)) |
168 | goto out_cleanup_inode; | |
169 | ||
2401dc29 | 170 | #ifdef CONFIG_XFS_POSIX_ACL |
db0bb7ba | 171 | if (default_acl) { |
8ba35875 | 172 | error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); |
2401dc29 | 173 | if (error) |
db0bb7ba | 174 | goto out_cleanup_inode; |
1da177e4 | 175 | } |
2401dc29 | 176 | if (acl) { |
8ba35875 | 177 | error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); |
2401dc29 CH |
178 | if (error) |
179 | goto out_cleanup_inode; | |
180 | } | |
181 | #endif | |
1da177e4 | 182 | |
2b3d1d41 CH |
183 | xfs_setup_iops(ip); |
184 | ||
c4a6bf7f DW |
185 | if (tmpfile) { |
186 | /* | |
187 | * The VFS requires that any inode fed to d_tmpfile must have | |
188 | * nlink == 1 so that it can decrement the nlink in d_tmpfile. | |
189 | * However, we created the temp file with nlink == 0 because | |
190 | * we're not allowed to put an inode with nlink > 0 on the | |
191 | * unlinked list. Therefore we have to set nlink to 1 so that | |
192 | * d_tmpfile can immediately set it back to zero. | |
193 | */ | |
194 | set_nlink(inode, 1); | |
d540e43b | 195 | d_tmpfile(dentry, inode); |
c4a6bf7f | 196 | } else |
d540e43b BF |
197 | d_instantiate(dentry, inode); |
198 | ||
58c90473 DC |
199 | xfs_finish_inode_setup(ip); |
200 | ||
2401dc29 CH |
201 | out_free_acl: |
202 | if (default_acl) | |
203 | posix_acl_release(default_acl); | |
204 | if (acl) | |
205 | posix_acl_release(acl); | |
2451337d | 206 | return error; |
db0bb7ba CH |
207 | |
208 | out_cleanup_inode: | |
58c90473 | 209 | xfs_finish_inode_setup(ip); |
d540e43b BF |
210 | if (!tmpfile) |
211 | xfs_cleanup_inode(dir, inode, dentry); | |
44a8736b | 212 | xfs_irele(ip); |
2401dc29 | 213 | goto out_free_acl; |
1da177e4 LT |
214 | } |
215 | ||
d540e43b BF |
216 | STATIC int |
217 | xfs_vn_mknod( | |
218 | struct inode *dir, | |
219 | struct dentry *dentry, | |
220 | umode_t mode, | |
221 | dev_t rdev) | |
222 | { | |
223 | return xfs_generic_create(dir, dentry, mode, rdev, false); | |
224 | } | |
225 | ||
1da177e4 | 226 | STATIC int |
416c6d5b | 227 | xfs_vn_create( |
1da177e4 LT |
228 | struct inode *dir, |
229 | struct dentry *dentry, | |
4acdaf27 | 230 | umode_t mode, |
ebfc3b49 | 231 | bool flags) |
1da177e4 | 232 | { |
416c6d5b | 233 | return xfs_vn_mknod(dir, dentry, mode, 0); |
1da177e4 LT |
234 | } |
235 | ||
236 | STATIC int | |
416c6d5b | 237 | xfs_vn_mkdir( |
1da177e4 LT |
238 | struct inode *dir, |
239 | struct dentry *dentry, | |
18bb1db3 | 240 | umode_t mode) |
1da177e4 | 241 | { |
416c6d5b | 242 | return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0); |
1da177e4 LT |
243 | } |
244 | ||
245 | STATIC struct dentry * | |
416c6d5b | 246 | xfs_vn_lookup( |
1da177e4 LT |
247 | struct inode *dir, |
248 | struct dentry *dentry, | |
00cd8dd3 | 249 | unsigned int flags) |
1da177e4 | 250 | { |
b113a6d3 | 251 | struct inode *inode; |
ef1f5e7a | 252 | struct xfs_inode *cip; |
556b8b16 | 253 | struct xfs_name name; |
1da177e4 LT |
254 | int error; |
255 | ||
256 | if (dentry->d_name.len >= MAXNAMELEN) | |
257 | return ERR_PTR(-ENAMETOOLONG); | |
258 | ||
fab8eef8 | 259 | xfs_dentry_to_name(&name, dentry); |
384f3ced | 260 | error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); |
b113a6d3 AV |
261 | if (likely(!error)) |
262 | inode = VFS_I(cip); | |
263 | else if (likely(error == -ENOENT)) | |
264 | inode = NULL; | |
265 | else | |
266 | inode = ERR_PTR(error); | |
267 | return d_splice_alias(inode, dentry); | |
1da177e4 LT |
268 | } |
269 | ||
384f3ced BN |
270 | STATIC struct dentry * |
271 | xfs_vn_ci_lookup( | |
272 | struct inode *dir, | |
273 | struct dentry *dentry, | |
00cd8dd3 | 274 | unsigned int flags) |
384f3ced BN |
275 | { |
276 | struct xfs_inode *ip; | |
277 | struct xfs_name xname; | |
278 | struct xfs_name ci_name; | |
279 | struct qstr dname; | |
280 | int error; | |
281 | ||
282 | if (dentry->d_name.len >= MAXNAMELEN) | |
283 | return ERR_PTR(-ENAMETOOLONG); | |
284 | ||
fab8eef8 | 285 | xfs_dentry_to_name(&xname, dentry); |
384f3ced BN |
286 | error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); |
287 | if (unlikely(error)) { | |
2451337d DC |
288 | if (unlikely(error != -ENOENT)) |
289 | return ERR_PTR(error); | |
866d5dc9 BN |
290 | /* |
291 | * call d_add(dentry, NULL) here when d_drop_negative_children | |
292 | * is called in xfs_vn_mknod (ie. allow negative dentries | |
293 | * with CI filesystems). | |
294 | */ | |
384f3ced BN |
295 | return NULL; |
296 | } | |
297 | ||
298 | /* if exact match, just splice and exit */ | |
299 | if (!ci_name.name) | |
01651646 | 300 | return d_splice_alias(VFS_I(ip), dentry); |
384f3ced BN |
301 | |
302 | /* else case-insensitive match... */ | |
303 | dname.name = ci_name.name; | |
304 | dname.len = ci_name.len; | |
e45b590b | 305 | dentry = d_add_ci(dentry, VFS_I(ip), &dname); |
384f3ced BN |
306 | kmem_free(ci_name.name); |
307 | return dentry; | |
308 | } | |
309 | ||
1da177e4 | 310 | STATIC int |
416c6d5b | 311 | xfs_vn_link( |
1da177e4 LT |
312 | struct dentry *old_dentry, |
313 | struct inode *dir, | |
314 | struct dentry *dentry) | |
315 | { | |
2b0143b5 | 316 | struct inode *inode = d_inode(old_dentry); |
556b8b16 | 317 | struct xfs_name name; |
1da177e4 LT |
318 | int error; |
319 | ||
fab8eef8 AG |
320 | error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode); |
321 | if (unlikely(error)) | |
322 | return error; | |
1da177e4 | 323 | |
556b8b16 | 324 | error = xfs_link(XFS_I(dir), XFS_I(inode), &name); |
d9424b3c | 325 | if (unlikely(error)) |
2451337d | 326 | return error; |
a3da7896 | 327 | |
7de9c6ee | 328 | ihold(inode); |
a3da7896 CH |
329 | d_instantiate(dentry, inode); |
330 | return 0; | |
1da177e4 LT |
331 | } |
332 | ||
333 | STATIC int | |
416c6d5b | 334 | xfs_vn_unlink( |
1da177e4 LT |
335 | struct inode *dir, |
336 | struct dentry *dentry) | |
337 | { | |
556b8b16 | 338 | struct xfs_name name; |
1da177e4 LT |
339 | int error; |
340 | ||
fab8eef8 | 341 | xfs_dentry_to_name(&name, dentry); |
1da177e4 | 342 | |
2b0143b5 | 343 | error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry))); |
e5700704 CH |
344 | if (error) |
345 | return error; | |
346 | ||
347 | /* | |
348 | * With unlink, the VFS makes the dentry "negative": no inode, | |
349 | * but still hashed. This is incompatible with case-insensitive | |
350 | * mode, so invalidate (unhash) the dentry in CI-mode. | |
351 | */ | |
352 | if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb)) | |
353 | d_invalidate(dentry); | |
354 | return 0; | |
1da177e4 LT |
355 | } |
356 | ||
357 | STATIC int | |
416c6d5b | 358 | xfs_vn_symlink( |
1da177e4 LT |
359 | struct inode *dir, |
360 | struct dentry *dentry, | |
361 | const char *symname) | |
362 | { | |
3937be5b CH |
363 | struct inode *inode; |
364 | struct xfs_inode *cip = NULL; | |
556b8b16 | 365 | struct xfs_name name; |
1da177e4 | 366 | int error; |
576b1d67 | 367 | umode_t mode; |
1da177e4 | 368 | |
3e5daf05 | 369 | mode = S_IFLNK | |
ce3b0f8d | 370 | (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); |
fab8eef8 AG |
371 | error = xfs_dentry_mode_to_name(&name, dentry, mode); |
372 | if (unlikely(error)) | |
373 | goto out; | |
1da177e4 | 374 | |
6c77b0ea | 375 | error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip); |
3937be5b CH |
376 | if (unlikely(error)) |
377 | goto out; | |
378 | ||
01651646 | 379 | inode = VFS_I(cip); |
3937be5b | 380 | |
2a7dba39 | 381 | error = xfs_init_security(inode, dir, &dentry->d_name); |
3937be5b CH |
382 | if (unlikely(error)) |
383 | goto out_cleanup_inode; | |
384 | ||
2b3d1d41 CH |
385 | xfs_setup_iops(cip); |
386 | ||
3937be5b | 387 | d_instantiate(dentry, inode); |
58c90473 | 388 | xfs_finish_inode_setup(cip); |
3937be5b CH |
389 | return 0; |
390 | ||
391 | out_cleanup_inode: | |
58c90473 | 392 | xfs_finish_inode_setup(cip); |
8f112e3b | 393 | xfs_cleanup_inode(dir, inode, dentry); |
44a8736b | 394 | xfs_irele(cip); |
3937be5b | 395 | out: |
2451337d | 396 | return error; |
1da177e4 LT |
397 | } |
398 | ||
1da177e4 | 399 | STATIC int |
416c6d5b | 400 | xfs_vn_rename( |
1da177e4 LT |
401 | struct inode *odir, |
402 | struct dentry *odentry, | |
403 | struct inode *ndir, | |
dbe1b5ca CM |
404 | struct dentry *ndentry, |
405 | unsigned int flags) | |
1da177e4 | 406 | { |
2b0143b5 | 407 | struct inode *new_inode = d_inode(ndentry); |
d31a1825 | 408 | int omode = 0; |
fab8eef8 | 409 | int error; |
556b8b16 BN |
410 | struct xfs_name oname; |
411 | struct xfs_name nname; | |
1da177e4 | 412 | |
7dcf5c3e | 413 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
dbe1b5ca CM |
414 | return -EINVAL; |
415 | ||
d31a1825 CM |
416 | /* if we are exchanging files, we need to set i_mode of both files */ |
417 | if (flags & RENAME_EXCHANGE) | |
2b0143b5 | 418 | omode = d_inode(ndentry)->i_mode; |
d31a1825 | 419 | |
fab8eef8 AG |
420 | error = xfs_dentry_mode_to_name(&oname, odentry, omode); |
421 | if (omode && unlikely(error)) | |
422 | return error; | |
423 | ||
424 | error = xfs_dentry_mode_to_name(&nname, ndentry, | |
425 | d_inode(odentry)->i_mode); | |
426 | if (unlikely(error)) | |
427 | return error; | |
556b8b16 | 428 | |
2b0143b5 | 429 | return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)), |
dbe1b5ca | 430 | XFS_I(ndir), &nname, |
d31a1825 | 431 | new_inode ? XFS_I(new_inode) : NULL, flags); |
1da177e4 LT |
432 | } |
433 | ||
434 | /* | |
435 | * careful here - this function can get called recursively, so | |
436 | * we need to be very careful about how much stack we use. | |
437 | * uio is kmalloced for this reason... | |
438 | */ | |
680baacb | 439 | STATIC const char * |
6b255391 | 440 | xfs_vn_get_link( |
1da177e4 | 441 | struct dentry *dentry, |
6b255391 | 442 | struct inode *inode, |
fceef393 | 443 | struct delayed_call *done) |
1da177e4 | 444 | { |
1da177e4 | 445 | char *link; |
804c83c3 | 446 | int error = -ENOMEM; |
1da177e4 | 447 | |
6b255391 AV |
448 | if (!dentry) |
449 | return ERR_PTR(-ECHILD); | |
450 | ||
6eb0b8df | 451 | link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); |
804c83c3 CH |
452 | if (!link) |
453 | goto out_err; | |
1da177e4 | 454 | |
2b0143b5 | 455 | error = xfs_readlink(XFS_I(d_inode(dentry)), link); |
804c83c3 CH |
456 | if (unlikely(error)) |
457 | goto out_kfree; | |
1da177e4 | 458 | |
fceef393 AV |
459 | set_delayed_call(done, kfree_link, link); |
460 | return link; | |
804c83c3 CH |
461 | |
462 | out_kfree: | |
463 | kfree(link); | |
464 | out_err: | |
680baacb | 465 | return ERR_PTR(error); |
1da177e4 LT |
466 | } |
467 | ||
30ee052e CH |
468 | STATIC const char * |
469 | xfs_vn_get_link_inline( | |
470 | struct dentry *dentry, | |
471 | struct inode *inode, | |
472 | struct delayed_call *done) | |
473 | { | |
a5155b87 | 474 | struct xfs_inode *ip = XFS_I(inode); |
ae294787 DW |
475 | char *link; |
476 | ||
a5155b87 | 477 | ASSERT(ip->i_df.if_flags & XFS_IFINLINE); |
ae294787 DW |
478 | |
479 | /* | |
480 | * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if | |
481 | * if_data is junk. | |
482 | */ | |
a5155b87 | 483 | link = ip->i_df.if_u1.if_data; |
a71895c5 | 484 | if (XFS_IS_CORRUPT(ip->i_mount, !link)) |
ae294787 DW |
485 | return ERR_PTR(-EFSCORRUPTED); |
486 | return link; | |
30ee052e CH |
487 | } |
488 | ||
dd2d535e CH |
489 | static uint32_t |
490 | xfs_stat_blksize( | |
491 | struct xfs_inode *ip) | |
492 | { | |
493 | struct xfs_mount *mp = ip->i_mount; | |
494 | ||
495 | /* | |
496 | * If the file blocks are being allocated from a realtime volume, then | |
497 | * always return the realtime extent size. | |
498 | */ | |
499 | if (XFS_IS_REALTIME_INODE(ip)) | |
500 | return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog; | |
501 | ||
502 | /* | |
503 | * Allow large block sizes to be reported to userspace programs if the | |
504 | * "largeio" mount option is used. | |
505 | * | |
506 | * If compatibility mode is specified, simply return the basic unit of | |
507 | * caching so that we don't get inefficient read/modify/write I/O from | |
508 | * user apps. Otherwise.... | |
509 | * | |
510 | * If the underlying volume is a stripe, then return the stripe width in | |
511 | * bytes as the recommended I/O size. It is not a stripe and we've set a | |
512 | * default buffered I/O size, return that, otherwise return the compat | |
513 | * default. | |
514 | */ | |
7c6b94b1 | 515 | if (mp->m_flags & XFS_MOUNT_LARGEIO) { |
dd2d535e CH |
516 | if (mp->m_swidth) |
517 | return mp->m_swidth << mp->m_sb.sb_blocklog; | |
3274d008 | 518 | if (mp->m_flags & XFS_MOUNT_ALLOCSIZE) |
5da8a07c | 519 | return 1U << mp->m_allocsize_log; |
dd2d535e CH |
520 | } |
521 | ||
522 | return PAGE_SIZE; | |
523 | } | |
524 | ||
1da177e4 | 525 | STATIC int |
416c6d5b | 526 | xfs_vn_getattr( |
a528d35e DH |
527 | const struct path *path, |
528 | struct kstat *stat, | |
529 | u32 request_mask, | |
530 | unsigned int query_flags) | |
1da177e4 | 531 | { |
a528d35e | 532 | struct inode *inode = d_inode(path->dentry); |
c43f4087 CH |
533 | struct xfs_inode *ip = XFS_I(inode); |
534 | struct xfs_mount *mp = ip->i_mount; | |
535 | ||
cca28fb8 | 536 | trace_xfs_getattr(ip); |
c43f4087 CH |
537 | |
538 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 539 | return -EIO; |
c43f4087 CH |
540 | |
541 | stat->size = XFS_ISIZE(ip); | |
542 | stat->dev = inode->i_sb->s_dev; | |
c19b3b05 | 543 | stat->mode = inode->i_mode; |
54d7b5c1 | 544 | stat->nlink = inode->i_nlink; |
7aab1b28 DE |
545 | stat->uid = inode->i_uid; |
546 | stat->gid = inode->i_gid; | |
c43f4087 | 547 | stat->ino = ip->i_ino; |
c43f4087 | 548 | stat->atime = inode->i_atime; |
f9581b14 CH |
549 | stat->mtime = inode->i_mtime; |
550 | stat->ctime = inode->i_ctime; | |
c43f4087 CH |
551 | stat->blocks = |
552 | XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); | |
553 | ||
5f955f26 DW |
554 | if (ip->i_d.di_version == 3) { |
555 | if (request_mask & STATX_BTIME) { | |
556 | stat->result_mask |= STATX_BTIME; | |
557 | stat->btime.tv_sec = ip->i_d.di_crtime.t_sec; | |
558 | stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec; | |
559 | } | |
560 | } | |
561 | ||
1b9598c8 LR |
562 | /* |
563 | * Note: If you add another clause to set an attribute flag, please | |
564 | * update attributes_mask below. | |
565 | */ | |
5f955f26 DW |
566 | if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE) |
567 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
568 | if (ip->i_d.di_flags & XFS_DIFLAG_APPEND) | |
569 | stat->attributes |= STATX_ATTR_APPEND; | |
570 | if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP) | |
571 | stat->attributes |= STATX_ATTR_NODUMP; | |
c43f4087 | 572 | |
1b9598c8 LR |
573 | stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | |
574 | STATX_ATTR_APPEND | | |
575 | STATX_ATTR_NODUMP); | |
576 | ||
c43f4087 CH |
577 | switch (inode->i_mode & S_IFMT) { |
578 | case S_IFBLK: | |
579 | case S_IFCHR: | |
580 | stat->blksize = BLKDEV_IOSIZE; | |
66f36464 | 581 | stat->rdev = inode->i_rdev; |
c43f4087 CH |
582 | break; |
583 | default: | |
dd2d535e | 584 | stat->blksize = xfs_stat_blksize(ip); |
c43f4087 CH |
585 | stat->rdev = 0; |
586 | break; | |
69e23b9a | 587 | } |
c43f4087 CH |
588 | |
589 | return 0; | |
1da177e4 LT |
590 | } |
591 | ||
56c19e89 DC |
592 | static void |
593 | xfs_setattr_mode( | |
56c19e89 DC |
594 | struct xfs_inode *ip, |
595 | struct iattr *iattr) | |
596 | { | |
0c3d88df CH |
597 | struct inode *inode = VFS_I(ip); |
598 | umode_t mode = iattr->ia_mode; | |
56c19e89 | 599 | |
56c19e89 DC |
600 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
601 | ||
56c19e89 DC |
602 | inode->i_mode &= S_IFMT; |
603 | inode->i_mode |= mode & ~S_IFMT; | |
604 | } | |
605 | ||
52785112 | 606 | void |
c91c46c1 CH |
607 | xfs_setattr_time( |
608 | struct xfs_inode *ip, | |
609 | struct iattr *iattr) | |
610 | { | |
611 | struct inode *inode = VFS_I(ip); | |
612 | ||
613 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
614 | ||
3987848c | 615 | if (iattr->ia_valid & ATTR_ATIME) |
c91c46c1 | 616 | inode->i_atime = iattr->ia_atime; |
3987848c | 617 | if (iattr->ia_valid & ATTR_CTIME) |
c91c46c1 | 618 | inode->i_ctime = iattr->ia_ctime; |
3987848c | 619 | if (iattr->ia_valid & ATTR_MTIME) |
c91c46c1 | 620 | inode->i_mtime = iattr->ia_mtime; |
c91c46c1 CH |
621 | } |
622 | ||
69bca807 JK |
623 | static int |
624 | xfs_vn_change_ok( | |
625 | struct dentry *dentry, | |
626 | struct iattr *iattr) | |
627 | { | |
31051c85 | 628 | struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; |
69bca807 JK |
629 | |
630 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
631 | return -EROFS; | |
632 | ||
633 | if (XFS_FORCED_SHUTDOWN(mp)) | |
634 | return -EIO; | |
635 | ||
31051c85 | 636 | return setattr_prepare(dentry, iattr); |
69bca807 JK |
637 | } |
638 | ||
639 | /* | |
640 | * Set non-size attributes of an inode. | |
641 | * | |
642 | * Caution: The caller of this function is responsible for calling | |
31051c85 | 643 | * setattr_prepare() or otherwise verifying the change is fine. |
69bca807 | 644 | */ |
c4ed4243 CH |
645 | int |
646 | xfs_setattr_nonsize( | |
647 | struct xfs_inode *ip, | |
648 | struct iattr *iattr, | |
649 | int flags) | |
650 | { | |
651 | xfs_mount_t *mp = ip->i_mount; | |
652 | struct inode *inode = VFS_I(ip); | |
653 | int mask = iattr->ia_valid; | |
654 | xfs_trans_t *tp; | |
655 | int error; | |
7aab1b28 DE |
656 | kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID; |
657 | kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID; | |
c4ed4243 CH |
658 | struct xfs_dquot *udqp = NULL, *gdqp = NULL; |
659 | struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL; | |
660 | ||
c4ed4243 CH |
661 | ASSERT((mask & ATTR_SIZE) == 0); |
662 | ||
663 | /* | |
664 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
665 | * before we start any other transactions. Trying to do this later | |
666 | * is messy. We don't care to take a readlock to look at the ids | |
667 | * in inode here, because we can't hold it across the trans_reserve. | |
668 | * If the IDs do change before we take the ilock, we're covered | |
669 | * because the i_*dquot fields will get updated anyway. | |
670 | */ | |
671 | if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { | |
672 | uint qflags = 0; | |
673 | ||
674 | if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { | |
675 | uid = iattr->ia_uid; | |
676 | qflags |= XFS_QMOPT_UQUOTA; | |
677 | } else { | |
7aab1b28 | 678 | uid = inode->i_uid; |
c4ed4243 CH |
679 | } |
680 | if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { | |
681 | gid = iattr->ia_gid; | |
682 | qflags |= XFS_QMOPT_GQUOTA; | |
683 | } else { | |
7aab1b28 | 684 | gid = inode->i_gid; |
c4ed4243 CH |
685 | } |
686 | ||
687 | /* | |
688 | * We take a reference when we initialize udqp and gdqp, | |
689 | * so it is important that we never blindly double trip on | |
690 | * the same variable. See xfs_create() for an example. | |
691 | */ | |
692 | ASSERT(udqp == NULL); | |
693 | ASSERT(gdqp == NULL); | |
7aab1b28 DE |
694 | error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid), |
695 | xfs_kgid_to_gid(gid), | |
696 | xfs_get_projid(ip), | |
697 | qflags, &udqp, &gdqp, NULL); | |
c4ed4243 CH |
698 | if (error) |
699 | return error; | |
700 | } | |
701 | ||
253f4911 | 702 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp); |
c4ed4243 | 703 | if (error) |
253f4911 | 704 | goto out_dqrele; |
c4ed4243 CH |
705 | |
706 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
253f4911 | 707 | xfs_trans_ijoin(tp, ip, 0); |
c4ed4243 CH |
708 | |
709 | /* | |
710 | * Change file ownership. Must be the owner or privileged. | |
711 | */ | |
712 | if (mask & (ATTR_UID|ATTR_GID)) { | |
713 | /* | |
714 | * These IDs could have changed since we last looked at them. | |
715 | * But, we're assured that if the ownership did change | |
716 | * while we didn't have the inode locked, inode's dquot(s) | |
717 | * would have changed also. | |
718 | */ | |
7aab1b28 DE |
719 | iuid = inode->i_uid; |
720 | igid = inode->i_gid; | |
c4ed4243 CH |
721 | gid = (mask & ATTR_GID) ? iattr->ia_gid : igid; |
722 | uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid; | |
723 | ||
724 | /* | |
725 | * Do a quota reservation only if uid/gid is actually | |
726 | * going to change. | |
727 | */ | |
728 | if (XFS_IS_QUOTA_RUNNING(mp) && | |
7aab1b28 DE |
729 | ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) || |
730 | (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) { | |
c4ed4243 CH |
731 | ASSERT(tp); |
732 | error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp, | |
92f8ff73 | 733 | NULL, capable(CAP_FOWNER) ? |
c4ed4243 CH |
734 | XFS_QMOPT_FORCE_RES : 0); |
735 | if (error) /* out of quota */ | |
253f4911 | 736 | goto out_cancel; |
c4ed4243 CH |
737 | } |
738 | } | |
739 | ||
c4ed4243 CH |
740 | /* |
741 | * Change file ownership. Must be the owner or privileged. | |
742 | */ | |
743 | if (mask & (ATTR_UID|ATTR_GID)) { | |
744 | /* | |
745 | * CAP_FSETID overrides the following restrictions: | |
746 | * | |
747 | * The set-user-ID and set-group-ID bits of a file will be | |
748 | * cleared upon successful return from chown() | |
749 | */ | |
c19b3b05 | 750 | if ((inode->i_mode & (S_ISUID|S_ISGID)) && |
c4ed4243 | 751 | !capable(CAP_FSETID)) |
c19b3b05 | 752 | inode->i_mode &= ~(S_ISUID|S_ISGID); |
c4ed4243 CH |
753 | |
754 | /* | |
755 | * Change the ownerships and register quota modifications | |
756 | * in the transaction. | |
757 | */ | |
7aab1b28 | 758 | if (!uid_eq(iuid, uid)) { |
c4ed4243 CH |
759 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) { |
760 | ASSERT(mask & ATTR_UID); | |
761 | ASSERT(udqp); | |
762 | olddquot1 = xfs_qm_vop_chown(tp, ip, | |
763 | &ip->i_udquot, udqp); | |
764 | } | |
7aab1b28 | 765 | ip->i_d.di_uid = xfs_kuid_to_uid(uid); |
c4ed4243 CH |
766 | inode->i_uid = uid; |
767 | } | |
7aab1b28 | 768 | if (!gid_eq(igid, gid)) { |
c4ed4243 | 769 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) { |
5a01dd54 JL |
770 | ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) || |
771 | !XFS_IS_PQUOTA_ON(mp)); | |
c4ed4243 CH |
772 | ASSERT(mask & ATTR_GID); |
773 | ASSERT(gdqp); | |
774 | olddquot2 = xfs_qm_vop_chown(tp, ip, | |
775 | &ip->i_gdquot, gdqp); | |
776 | } | |
7aab1b28 | 777 | ip->i_d.di_gid = xfs_kgid_to_gid(gid); |
c4ed4243 CH |
778 | inode->i_gid = gid; |
779 | } | |
780 | } | |
781 | ||
56c19e89 | 782 | if (mask & ATTR_MODE) |
0c3d88df | 783 | xfs_setattr_mode(ip, iattr); |
c91c46c1 CH |
784 | if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) |
785 | xfs_setattr_time(ip, iattr); | |
c4ed4243 CH |
786 | |
787 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
788 | ||
ff6d6af2 | 789 | XFS_STATS_INC(mp, xs_ig_attrchg); |
c4ed4243 CH |
790 | |
791 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
792 | xfs_trans_set_sync(tp); | |
70393313 | 793 | error = xfs_trans_commit(tp); |
c4ed4243 CH |
794 | |
795 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
796 | ||
797 | /* | |
798 | * Release any dquot(s) the inode had kept before chown. | |
799 | */ | |
800 | xfs_qm_dqrele(olddquot1); | |
801 | xfs_qm_dqrele(olddquot2); | |
802 | xfs_qm_dqrele(udqp); | |
803 | xfs_qm_dqrele(gdqp); | |
804 | ||
805 | if (error) | |
b474c7ae | 806 | return error; |
c4ed4243 CH |
807 | |
808 | /* | |
809 | * XXX(hch): Updating the ACL entries is not atomic vs the i_mode | |
810 | * update. We could avoid this with linked transactions | |
811 | * and passing down the transaction pointer all the way | |
812 | * to attr_set. No previous user of the generic | |
813 | * Posix ACL code seems to care about this issue either. | |
814 | */ | |
815 | if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) { | |
2451337d | 816 | error = posix_acl_chmod(inode, inode->i_mode); |
c4ed4243 | 817 | if (error) |
b474c7ae | 818 | return error; |
c4ed4243 CH |
819 | } |
820 | ||
821 | return 0; | |
822 | ||
253f4911 | 823 | out_cancel: |
4906e215 | 824 | xfs_trans_cancel(tp); |
1fb254aa | 825 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
253f4911 | 826 | out_dqrele: |
c4ed4243 CH |
827 | xfs_qm_dqrele(udqp); |
828 | xfs_qm_dqrele(gdqp); | |
829 | return error; | |
830 | } | |
831 | ||
69bca807 JK |
832 | int |
833 | xfs_vn_setattr_nonsize( | |
834 | struct dentry *dentry, | |
835 | struct iattr *iattr) | |
836 | { | |
837 | struct xfs_inode *ip = XFS_I(d_inode(dentry)); | |
838 | int error; | |
839 | ||
840 | trace_xfs_setattr(ip); | |
841 | ||
842 | error = xfs_vn_change_ok(dentry, iattr); | |
843 | if (error) | |
844 | return error; | |
845 | return xfs_setattr_nonsize(ip, iattr, 0); | |
846 | } | |
847 | ||
c4ed4243 CH |
848 | /* |
849 | * Truncate file. Must have write permission and not be a directory. | |
69bca807 JK |
850 | * |
851 | * Caution: The caller of this function is responsible for calling | |
31051c85 | 852 | * setattr_prepare() or otherwise verifying the change is fine. |
c4ed4243 | 853 | */ |
7bf7a193 | 854 | STATIC int |
c4ed4243 CH |
855 | xfs_setattr_size( |
856 | struct xfs_inode *ip, | |
76ca4c23 | 857 | struct iattr *iattr) |
c4ed4243 CH |
858 | { |
859 | struct xfs_mount *mp = ip->i_mount; | |
860 | struct inode *inode = VFS_I(ip); | |
673e8e59 | 861 | xfs_off_t oldsize, newsize; |
c4ed4243 CH |
862 | struct xfs_trans *tp; |
863 | int error; | |
f38996f5 | 864 | uint lock_flags = 0; |
5885ebda | 865 | bool did_zeroing = false; |
c4ed4243 | 866 | |
76ca4c23 | 867 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
e8e9ad42 | 868 | ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); |
c19b3b05 | 869 | ASSERT(S_ISREG(inode->i_mode)); |
fe60a8a0 CH |
870 | ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| |
871 | ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0); | |
c4ed4243 | 872 | |
ce7ae151 | 873 | oldsize = inode->i_size; |
673e8e59 CH |
874 | newsize = iattr->ia_size; |
875 | ||
c4ed4243 CH |
876 | /* |
877 | * Short circuit the truncate case for zero length files. | |
878 | */ | |
673e8e59 | 879 | if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) { |
fe60a8a0 | 880 | if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) |
76ca4c23 | 881 | return 0; |
681b1200 CH |
882 | |
883 | /* | |
884 | * Use the regular setattr path to update the timestamps. | |
885 | */ | |
681b1200 CH |
886 | iattr->ia_valid &= ~ATTR_SIZE; |
887 | return xfs_setattr_nonsize(ip, iattr, 0); | |
c4ed4243 CH |
888 | } |
889 | ||
890 | /* | |
891 | * Make sure that the dquots are attached to the inode. | |
892 | */ | |
c14cfcca | 893 | error = xfs_qm_dqattach(ip); |
c4ed4243 | 894 | if (error) |
76ca4c23 | 895 | return error; |
c4ed4243 | 896 | |
f0c6bcba CH |
897 | /* |
898 | * Wait for all direct I/O to complete. | |
899 | */ | |
900 | inode_dio_wait(inode); | |
901 | ||
c4ed4243 | 902 | /* |
5885ebda DC |
903 | * File data changes must be complete before we start the transaction to |
904 | * modify the inode. This needs to be done before joining the inode to | |
905 | * the transaction because the inode cannot be unlocked once it is a | |
906 | * part of the transaction. | |
907 | * | |
f0c6bcba CH |
908 | * Start with zeroing any data beyond EOF that we may expose on file |
909 | * extension, or zeroing out the rest of the block on a downward | |
910 | * truncate. | |
c4ed4243 | 911 | */ |
673e8e59 | 912 | if (newsize > oldsize) { |
f5c54717 CH |
913 | trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); |
914 | error = iomap_zero_range(inode, oldsize, newsize - oldsize, | |
f150b423 | 915 | &did_zeroing, &xfs_buffered_write_iomap_ops); |
f0c6bcba | 916 | } else { |
459f0fbc | 917 | error = iomap_truncate_page(inode, newsize, &did_zeroing, |
f150b423 | 918 | &xfs_buffered_write_iomap_ops); |
c4ed4243 | 919 | } |
c4ed4243 | 920 | |
f0c6bcba CH |
921 | if (error) |
922 | return error; | |
923 | ||
49abc3a8 | 924 | /* |
0f9160b4 DC |
925 | * We've already locked out new page faults, so now we can safely remove |
926 | * pages from the page cache knowing they won't get refaulted until we | |
927 | * drop the XFS_MMAP_EXCL lock after the extent manipulations are | |
928 | * complete. The truncate_setsize() call also cleans partial EOF page | |
929 | * PTEs on extending truncates and hence ensures sub-page block size | |
930 | * filesystems are correctly handled, too. | |
49abc3a8 | 931 | * |
0f9160b4 DC |
932 | * We have to do all the page cache truncate work outside the |
933 | * transaction context as the "lock" order is page lock->log space | |
934 | * reservation as defined by extent allocation in the writeback path. | |
253f4911 | 935 | * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but |
0f9160b4 DC |
936 | * having already truncated the in-memory version of the file (i.e. made |
937 | * user visible changes). There's not much we can do about this, except | |
938 | * to hope that the caller sees ENOMEM and retries the truncate | |
939 | * operation. | |
350976ae EG |
940 | * |
941 | * And we update in-core i_size and truncate page cache beyond newsize | |
942 | * before writeback the [di_size, newsize] range, so we're guaranteed | |
943 | * not to write stale data past the new EOF on truncate down. | |
49abc3a8 | 944 | */ |
49abc3a8 | 945 | truncate_setsize(inode, newsize); |
c4ed4243 | 946 | |
350976ae EG |
947 | /* |
948 | * We are going to log the inode size change in this transaction so | |
949 | * any previous writes that are beyond the on disk EOF and the new | |
950 | * EOF that have not been written out need to be written here. If we | |
951 | * do not write the data out, we expose ourselves to the null files | |
952 | * problem. Note that this includes any block zeroing we did above; | |
953 | * otherwise those blocks may not be zeroed after a crash. | |
954 | */ | |
955 | if (did_zeroing || | |
956 | (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) { | |
957 | error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, | |
958 | ip->i_d.di_size, newsize - 1); | |
959 | if (error) | |
960 | return error; | |
961 | } | |
962 | ||
253f4911 | 963 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); |
c4ed4243 | 964 | if (error) |
253f4911 | 965 | return error; |
c4ed4243 | 966 | |
c4ed4243 | 967 | lock_flags |= XFS_ILOCK_EXCL; |
c4ed4243 | 968 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
ddc3415a | 969 | xfs_trans_ijoin(tp, ip, 0); |
c4ed4243 CH |
970 | |
971 | /* | |
972 | * Only change the c/mtime if we are changing the size or we are | |
973 | * explicitly asked to change it. This handles the semantic difference | |
974 | * between truncate() and ftruncate() as implemented in the VFS. | |
975 | * | |
976 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
977 | * special case where we need to update the times despite not having | |
978 | * these flags set. For all other operations the VFS set these flags | |
979 | * explicitly if it wants a timestamp update. | |
980 | */ | |
fe60a8a0 CH |
981 | if (newsize != oldsize && |
982 | !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { | |
c4ed4243 | 983 | iattr->ia_ctime = iattr->ia_mtime = |
c2050a45 | 984 | current_time(inode); |
fe60a8a0 | 985 | iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; |
c4ed4243 CH |
986 | } |
987 | ||
673e8e59 CH |
988 | /* |
989 | * The first thing we do is set the size to new_size permanently on | |
990 | * disk. This way we don't have to worry about anyone ever being able | |
991 | * to look at the data being freed even in the face of a crash. | |
992 | * What we're getting around here is the case where we free a block, it | |
993 | * is allocated to another file, it is written to, and then we crash. | |
994 | * If the new data gets written to the file but the log buffers | |
995 | * containing the free and reallocation don't, then we'd end up with | |
996 | * garbage in the blocks being freed. As long as we make the new size | |
997 | * permanent before actually freeing any blocks it doesn't matter if | |
998 | * they get written to. | |
999 | */ | |
1000 | ip->i_d.di_size = newsize; | |
673e8e59 CH |
1001 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1002 | ||
1003 | if (newsize <= oldsize) { | |
1004 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); | |
c4ed4243 | 1005 | if (error) |
4906e215 | 1006 | goto out_trans_cancel; |
c4ed4243 CH |
1007 | |
1008 | /* | |
1009 | * Truncated "down", so we're removing references to old data | |
1010 | * here - if we delay flushing for a long time, we expose | |
1011 | * ourselves unduly to the notorious NULL files problem. So, | |
1012 | * we mark this inode and flush it when the file is closed, | |
1013 | * and do not wait the usual (long) time for writeout. | |
1014 | */ | |
1015 | xfs_iflags_set(ip, XFS_ITRUNCATED); | |
27b52867 BF |
1016 | |
1017 | /* A truncate down always removes post-EOF blocks. */ | |
1018 | xfs_inode_clear_eofblocks_tag(ip); | |
c4ed4243 CH |
1019 | } |
1020 | ||
fe60a8a0 | 1021 | if (iattr->ia_valid & ATTR_MODE) |
0c3d88df | 1022 | xfs_setattr_mode(ip, iattr); |
fe60a8a0 | 1023 | if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) |
c91c46c1 | 1024 | xfs_setattr_time(ip, iattr); |
c4ed4243 CH |
1025 | |
1026 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1027 | ||
ff6d6af2 | 1028 | XFS_STATS_INC(mp, xs_ig_attrchg); |
c4ed4243 CH |
1029 | |
1030 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1031 | xfs_trans_set_sync(tp); | |
1032 | ||
70393313 | 1033 | error = xfs_trans_commit(tp); |
c4ed4243 CH |
1034 | out_unlock: |
1035 | if (lock_flags) | |
1036 | xfs_iunlock(ip, lock_flags); | |
1037 | return error; | |
1038 | ||
c4ed4243 | 1039 | out_trans_cancel: |
4906e215 | 1040 | xfs_trans_cancel(tp); |
c4ed4243 CH |
1041 | goto out_unlock; |
1042 | } | |
1043 | ||
69bca807 JK |
1044 | int |
1045 | xfs_vn_setattr_size( | |
1046 | struct dentry *dentry, | |
1047 | struct iattr *iattr) | |
1048 | { | |
1049 | struct xfs_inode *ip = XFS_I(d_inode(dentry)); | |
1050 | int error; | |
1051 | ||
1052 | trace_xfs_setattr(ip); | |
1053 | ||
1054 | error = xfs_vn_change_ok(dentry, iattr); | |
1055 | if (error) | |
1056 | return error; | |
1057 | return xfs_setattr_size(ip, iattr); | |
1058 | } | |
1059 | ||
1da177e4 | 1060 | STATIC int |
416c6d5b | 1061 | xfs_vn_setattr( |
76ca4c23 CH |
1062 | struct dentry *dentry, |
1063 | struct iattr *iattr) | |
1da177e4 | 1064 | { |
76ca4c23 CH |
1065 | int error; |
1066 | ||
1067 | if (iattr->ia_valid & ATTR_SIZE) { | |
69eb5fa1 DW |
1068 | struct inode *inode = d_inode(dentry); |
1069 | struct xfs_inode *ip = XFS_I(inode); | |
c63a8eae | 1070 | uint iolock; |
781355c6 | 1071 | |
c63a8eae DW |
1072 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
1073 | iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; | |
781355c6 | 1074 | |
69eb5fa1 | 1075 | error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); |
c63a8eae DW |
1076 | if (error) { |
1077 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); | |
65523218 | 1078 | return error; |
c63a8eae | 1079 | } |
e8e9ad42 | 1080 | |
0c187dc5 | 1081 | error = xfs_vn_setattr_size(dentry, iattr); |
65523218 | 1082 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
76ca4c23 | 1083 | } else { |
69bca807 | 1084 | error = xfs_vn_setattr_nonsize(dentry, iattr); |
76ca4c23 CH |
1085 | } |
1086 | ||
2451337d | 1087 | return error; |
1da177e4 LT |
1088 | } |
1089 | ||
69ff2826 CH |
1090 | STATIC int |
1091 | xfs_vn_update_time( | |
1092 | struct inode *inode, | |
95582b00 | 1093 | struct timespec64 *now, |
69ff2826 CH |
1094 | int flags) |
1095 | { | |
1096 | struct xfs_inode *ip = XFS_I(inode); | |
1097 | struct xfs_mount *mp = ip->i_mount; | |
c3b1b131 | 1098 | int log_flags = XFS_ILOG_TIMESTAMP; |
69ff2826 CH |
1099 | struct xfs_trans *tp; |
1100 | int error; | |
1101 | ||
1102 | trace_xfs_update_time(ip); | |
1103 | ||
c3b1b131 CH |
1104 | if (inode->i_sb->s_flags & SB_LAZYTIME) { |
1105 | if (!((flags & S_VERSION) && | |
1106 | inode_maybe_inc_iversion(inode, false))) | |
1107 | return generic_update_time(inode, now, flags); | |
1108 | ||
1109 | /* Capture the iversion update that just occurred */ | |
1110 | log_flags |= XFS_ILOG_CORE; | |
1111 | } | |
1112 | ||
253f4911 CH |
1113 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
1114 | if (error) | |
2451337d | 1115 | return error; |
69ff2826 CH |
1116 | |
1117 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3987848c | 1118 | if (flags & S_CTIME) |
69ff2826 | 1119 | inode->i_ctime = *now; |
3987848c | 1120 | if (flags & S_MTIME) |
69ff2826 | 1121 | inode->i_mtime = *now; |
3987848c | 1122 | if (flags & S_ATIME) |
69ff2826 | 1123 | inode->i_atime = *now; |
3987848c | 1124 | |
69ff2826 | 1125 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
c3b1b131 | 1126 | xfs_trans_log_inode(tp, ip, log_flags); |
70393313 | 1127 | return xfs_trans_commit(tp); |
69ff2826 CH |
1128 | } |
1129 | ||
f35642e2 ES |
1130 | STATIC int |
1131 | xfs_vn_fiemap( | |
1132 | struct inode *inode, | |
1133 | struct fiemap_extent_info *fieinfo, | |
1134 | u64 start, | |
1135 | u64 length) | |
1136 | { | |
f35642e2 ES |
1137 | int error; |
1138 | ||
d2bb140e | 1139 | xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); |
1d4795e7 CH |
1140 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { |
1141 | fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; | |
1142 | error = iomap_fiemap(inode, fieinfo, start, length, | |
1143 | &xfs_xattr_iomap_ops); | |
1144 | } else { | |
1145 | error = iomap_fiemap(inode, fieinfo, start, length, | |
690c2a38 | 1146 | &xfs_read_iomap_ops); |
1d4795e7 | 1147 | } |
d2bb140e | 1148 | xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); |
f35642e2 | 1149 | |
d2bb140e | 1150 | return error; |
f35642e2 ES |
1151 | } |
1152 | ||
99b6436b ZYW |
1153 | STATIC int |
1154 | xfs_vn_tmpfile( | |
1155 | struct inode *dir, | |
1156 | struct dentry *dentry, | |
1157 | umode_t mode) | |
1158 | { | |
d540e43b | 1159 | return xfs_generic_create(dir, dentry, mode, 0, true); |
99b6436b ZYW |
1160 | } |
1161 | ||
41be8bed | 1162 | static const struct inode_operations xfs_inode_operations = { |
4e34e719 | 1163 | .get_acl = xfs_get_acl, |
2401dc29 | 1164 | .set_acl = xfs_set_acl, |
416c6d5b NS |
1165 | .getattr = xfs_vn_getattr, |
1166 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1167 | .listxattr = xfs_vn_listxattr, |
f35642e2 | 1168 | .fiemap = xfs_vn_fiemap, |
69ff2826 | 1169 | .update_time = xfs_vn_update_time, |
1da177e4 LT |
1170 | }; |
1171 | ||
41be8bed | 1172 | static const struct inode_operations xfs_dir_inode_operations = { |
416c6d5b NS |
1173 | .create = xfs_vn_create, |
1174 | .lookup = xfs_vn_lookup, | |
1175 | .link = xfs_vn_link, | |
1176 | .unlink = xfs_vn_unlink, | |
1177 | .symlink = xfs_vn_symlink, | |
1178 | .mkdir = xfs_vn_mkdir, | |
8f112e3b CH |
1179 | /* |
1180 | * Yes, XFS uses the same method for rmdir and unlink. | |
1181 | * | |
1182 | * There are some subtile differences deeper in the code, | |
1183 | * but we use S_ISDIR to check for those. | |
1184 | */ | |
1185 | .rmdir = xfs_vn_unlink, | |
416c6d5b | 1186 | .mknod = xfs_vn_mknod, |
2773bf00 | 1187 | .rename = xfs_vn_rename, |
4e34e719 | 1188 | .get_acl = xfs_get_acl, |
2401dc29 | 1189 | .set_acl = xfs_set_acl, |
416c6d5b NS |
1190 | .getattr = xfs_vn_getattr, |
1191 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1192 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1193 | .update_time = xfs_vn_update_time, |
99b6436b | 1194 | .tmpfile = xfs_vn_tmpfile, |
1da177e4 LT |
1195 | }; |
1196 | ||
41be8bed | 1197 | static const struct inode_operations xfs_dir_ci_inode_operations = { |
384f3ced BN |
1198 | .create = xfs_vn_create, |
1199 | .lookup = xfs_vn_ci_lookup, | |
1200 | .link = xfs_vn_link, | |
1201 | .unlink = xfs_vn_unlink, | |
1202 | .symlink = xfs_vn_symlink, | |
1203 | .mkdir = xfs_vn_mkdir, | |
8f112e3b CH |
1204 | /* |
1205 | * Yes, XFS uses the same method for rmdir and unlink. | |
1206 | * | |
1207 | * There are some subtile differences deeper in the code, | |
1208 | * but we use S_ISDIR to check for those. | |
1209 | */ | |
1210 | .rmdir = xfs_vn_unlink, | |
384f3ced | 1211 | .mknod = xfs_vn_mknod, |
2773bf00 | 1212 | .rename = xfs_vn_rename, |
4e34e719 | 1213 | .get_acl = xfs_get_acl, |
2401dc29 | 1214 | .set_acl = xfs_set_acl, |
384f3ced BN |
1215 | .getattr = xfs_vn_getattr, |
1216 | .setattr = xfs_vn_setattr, | |
384f3ced | 1217 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1218 | .update_time = xfs_vn_update_time, |
99b6436b | 1219 | .tmpfile = xfs_vn_tmpfile, |
384f3ced BN |
1220 | }; |
1221 | ||
41be8bed | 1222 | static const struct inode_operations xfs_symlink_inode_operations = { |
6b255391 | 1223 | .get_link = xfs_vn_get_link, |
416c6d5b NS |
1224 | .getattr = xfs_vn_getattr, |
1225 | .setattr = xfs_vn_setattr, | |
416c6d5b | 1226 | .listxattr = xfs_vn_listxattr, |
69ff2826 | 1227 | .update_time = xfs_vn_update_time, |
1da177e4 | 1228 | }; |
41be8bed | 1229 | |
30ee052e | 1230 | static const struct inode_operations xfs_inline_symlink_inode_operations = { |
30ee052e CH |
1231 | .get_link = xfs_vn_get_link_inline, |
1232 | .getattr = xfs_vn_getattr, | |
1233 | .setattr = xfs_vn_setattr, | |
30ee052e CH |
1234 | .listxattr = xfs_vn_listxattr, |
1235 | .update_time = xfs_vn_update_time, | |
1236 | }; | |
1237 | ||
ba23cba9 DW |
1238 | /* Figure out if this file actually supports DAX. */ |
1239 | static bool | |
1240 | xfs_inode_supports_dax( | |
1241 | struct xfs_inode *ip) | |
1242 | { | |
1243 | struct xfs_mount *mp = ip->i_mount; | |
1244 | ||
1245 | /* Only supported on non-reflinked files. */ | |
1246 | if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip)) | |
1247 | return false; | |
1248 | ||
1249 | /* DAX mount option or DAX iflag must be set. */ | |
1250 | if (!(mp->m_flags & XFS_MOUNT_DAX) && | |
1251 | !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)) | |
1252 | return false; | |
1253 | ||
1254 | /* Block size must match page size */ | |
1255 | if (mp->m_sb.sb_blocksize != PAGE_SIZE) | |
1256 | return false; | |
1257 | ||
1258 | /* Device has to support DAX too. */ | |
30fa529e | 1259 | return xfs_inode_buftarg(ip)->bt_daxdev != NULL; |
ba23cba9 DW |
1260 | } |
1261 | ||
41be8bed CH |
1262 | STATIC void |
1263 | xfs_diflags_to_iflags( | |
1264 | struct inode *inode, | |
1265 | struct xfs_inode *ip) | |
1266 | { | |
cbe4dab1 DC |
1267 | uint16_t flags = ip->i_d.di_flags; |
1268 | ||
1269 | inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | | |
1270 | S_NOATIME | S_DAX); | |
1271 | ||
1272 | if (flags & XFS_DIFLAG_IMMUTABLE) | |
41be8bed | 1273 | inode->i_flags |= S_IMMUTABLE; |
cbe4dab1 | 1274 | if (flags & XFS_DIFLAG_APPEND) |
41be8bed | 1275 | inode->i_flags |= S_APPEND; |
cbe4dab1 | 1276 | if (flags & XFS_DIFLAG_SYNC) |
41be8bed | 1277 | inode->i_flags |= S_SYNC; |
cbe4dab1 | 1278 | if (flags & XFS_DIFLAG_NOATIME) |
41be8bed | 1279 | inode->i_flags |= S_NOATIME; |
ba23cba9 | 1280 | if (xfs_inode_supports_dax(ip)) |
cbe4dab1 | 1281 | inode->i_flags |= S_DAX; |
41be8bed CH |
1282 | } |
1283 | ||
1284 | /* | |
2b3d1d41 | 1285 | * Initialize the Linux inode. |
bf904248 | 1286 | * |
58c90473 DC |
1287 | * When reading existing inodes from disk this is called directly from xfs_iget, |
1288 | * when creating a new inode it is called from xfs_ialloc after setting up the | |
1289 | * inode. These callers have different criteria for clearing XFS_INEW, so leave | |
1290 | * it up to the caller to deal with unlocking the inode appropriately. | |
41be8bed CH |
1291 | */ |
1292 | void | |
1293 | xfs_setup_inode( | |
1294 | struct xfs_inode *ip) | |
1295 | { | |
bf904248 | 1296 | struct inode *inode = &ip->i_vnode; |
ad22c7a0 | 1297 | gfp_t gfp_mask; |
bf904248 DC |
1298 | |
1299 | inode->i_ino = ip->i_ino; | |
eaff8079 | 1300 | inode->i_state = I_NEW; |
646ec461 CH |
1301 | |
1302 | inode_sb_list_add(inode); | |
c6f6cd06 | 1303 | /* make the inode look hashed for the writeback code */ |
5bef9151 | 1304 | inode_fake_hash(inode); |
41be8bed | 1305 | |
7aab1b28 DE |
1306 | inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid); |
1307 | inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid); | |
41be8bed | 1308 | |
41be8bed | 1309 | i_size_write(inode, ip->i_d.di_size); |
41be8bed | 1310 | xfs_diflags_to_iflags(inode, ip); |
41be8bed | 1311 | |
2b3d1d41 | 1312 | if (S_ISDIR(inode->i_mode)) { |
ef215e39 DC |
1313 | /* |
1314 | * We set the i_rwsem class here to avoid potential races with | |
1315 | * lockdep_annotate_inode_mutex_key() reinitialising the lock | |
1316 | * after a filehandle lookup has already found the inode in | |
1317 | * cache before it has been unlocked via unlock_new_inode(). | |
1318 | */ | |
1319 | lockdep_set_class(&inode->i_rwsem, | |
1320 | &inode->i_sb->s_type->i_mutex_dir_key); | |
93a8614e | 1321 | lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class); |
2b3d1d41 | 1322 | } else { |
2b3d1d41 | 1323 | lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class); |
41be8bed CH |
1324 | } |
1325 | ||
ad22c7a0 DC |
1326 | /* |
1327 | * Ensure all page cache allocations are done from GFP_NOFS context to | |
1328 | * prevent direct reclaim recursion back into the filesystem and blowing | |
1329 | * stacks or deadlocking. | |
1330 | */ | |
1331 | gfp_mask = mapping_gfp_mask(inode->i_mapping); | |
1332 | mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); | |
1333 | ||
510792ee CH |
1334 | /* |
1335 | * If there is no attribute fork no ACL can exist on this inode, | |
1336 | * and it can't have any file capabilities attached to it either. | |
1337 | */ | |
1338 | if (!XFS_IFORK_Q(ip)) { | |
1339 | inode_has_no_xattr(inode); | |
6311b108 | 1340 | cache_no_acl(inode); |
510792ee | 1341 | } |
41be8bed | 1342 | } |
2b3d1d41 CH |
1343 | |
1344 | void | |
1345 | xfs_setup_iops( | |
1346 | struct xfs_inode *ip) | |
1347 | { | |
1348 | struct inode *inode = &ip->i_vnode; | |
1349 | ||
41be8bed CH |
1350 | switch (inode->i_mode & S_IFMT) { |
1351 | case S_IFREG: | |
1352 | inode->i_op = &xfs_inode_operations; | |
1353 | inode->i_fop = &xfs_file_operations; | |
6e2608df DW |
1354 | if (IS_DAX(inode)) |
1355 | inode->i_mapping->a_ops = &xfs_dax_aops; | |
1356 | else | |
1357 | inode->i_mapping->a_ops = &xfs_address_space_operations; | |
41be8bed CH |
1358 | break; |
1359 | case S_IFDIR: | |
1360 | if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb)) | |
1361 | inode->i_op = &xfs_dir_ci_inode_operations; | |
1362 | else | |
1363 | inode->i_op = &xfs_dir_inode_operations; | |
1364 | inode->i_fop = &xfs_dir_file_operations; | |
1365 | break; | |
1366 | case S_IFLNK: | |
30ee052e CH |
1367 | if (ip->i_df.if_flags & XFS_IFINLINE) |
1368 | inode->i_op = &xfs_inline_symlink_inode_operations; | |
1369 | else | |
1370 | inode->i_op = &xfs_symlink_inode_operations; | |
41be8bed CH |
1371 | break; |
1372 | default: | |
1373 | inode->i_op = &xfs_inode_operations; | |
1374 | init_special_inode(inode, inode->i_mode, inode->i_rdev); | |
1375 | break; | |
1376 | } | |
41be8bed | 1377 | } |