1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2012-2013 Red Hat, Inc.
5 * All rights reserved.
6 */
7 #include "xfs.h"
8 #include "xfs_shared.h"
9 #include "xfs_fs.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_bit.h"
14 #include "xfs_mount.h"
15 #include "xfs_dir2.h"
16 #include "xfs_inode.h"
17 #include "xfs_bmap.h"
18 #include "xfs_bmap_btree.h"
19 #include "xfs_quota.h"
20 #include "xfs_symlink.h"
21 #include "xfs_trans_space.h"
22 #include "xfs_trace.h"
23 #include "xfs_trans.h"
24 #include "xfs_ialloc.h"
25 #include "xfs_error.h"
26
27 /* ----- Kernel only functions below ----- */
28 int
xfs_readlink_bmap_ilocked(struct xfs_inode * ip,char * link)29 xfs_readlink_bmap_ilocked(
30 struct xfs_inode *ip,
31 char *link)
32 {
33 struct xfs_mount *mp = ip->i_mount;
34 struct xfs_bmbt_irec mval[XFS_SYMLINK_MAPS];
35 struct xfs_buf *bp;
36 xfs_daddr_t d;
37 char *cur_chunk;
38 int pathlen = ip->i_disk_size;
39 int nmaps = XFS_SYMLINK_MAPS;
40 int byte_cnt;
41 int n;
42 int error = 0;
43 int fsblocks = 0;
44 int offset;
45
46 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
47
48 fsblocks = xfs_symlink_blocks(mp, pathlen);
49 error = xfs_bmapi_read(ip, 0, fsblocks, mval, &nmaps, 0);
50 if (error)
51 goto out;
52
53 offset = 0;
54 for (n = 0; n < nmaps; n++) {
55 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
56 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
57
58 error = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0,
59 &bp, &xfs_symlink_buf_ops);
60 if (error)
61 return error;
62 byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt);
63 if (pathlen < byte_cnt)
64 byte_cnt = pathlen;
65
66 cur_chunk = bp->b_addr;
67 if (xfs_has_crc(mp)) {
68 if (!xfs_symlink_hdr_ok(ip->i_ino, offset,
69 byte_cnt, bp)) {
70 error = -EFSCORRUPTED;
71 xfs_alert(mp,
72 "symlink header does not match required off/len/owner (0x%x/Ox%x,0x%llx)",
73 offset, byte_cnt, ip->i_ino);
74 xfs_buf_relse(bp);
75 goto out;
76
77 }
78
79 cur_chunk += sizeof(struct xfs_dsymlink_hdr);
80 }
81
82 memcpy(link + offset, cur_chunk, byte_cnt);
83
84 pathlen -= byte_cnt;
85 offset += byte_cnt;
86
87 xfs_buf_relse(bp);
88 }
89 ASSERT(pathlen == 0);
90
91 link[ip->i_disk_size] = '\0';
92 error = 0;
93
94 out:
95 return error;
96 }
97
98 int
xfs_readlink(struct xfs_inode * ip,char * link)99 xfs_readlink(
100 struct xfs_inode *ip,
101 char *link)
102 {
103 struct xfs_mount *mp = ip->i_mount;
104 xfs_fsize_t pathlen;
105 int error = -EFSCORRUPTED;
106
107 trace_xfs_readlink(ip);
108
109 if (xfs_is_shutdown(mp))
110 return -EIO;
111
112 xfs_ilock(ip, XFS_ILOCK_SHARED);
113
114 pathlen = ip->i_disk_size;
115 if (!pathlen)
116 goto out;
117
118 if (pathlen < 0 || pathlen > XFS_SYMLINK_MAXLEN) {
119 xfs_alert(mp, "%s: inode (%llu) bad symlink length (%lld)",
120 __func__, (unsigned long long) ip->i_ino,
121 (long long) pathlen);
122 ASSERT(0);
123 goto out;
124 }
125
126 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
127 /*
128 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED
129 * if if_data is junk.
130 */
131 if (XFS_IS_CORRUPT(ip->i_mount, !ip->i_df.if_u1.if_data))
132 goto out;
133
134 memcpy(link, ip->i_df.if_u1.if_data, pathlen + 1);
135 error = 0;
136 } else {
137 error = xfs_readlink_bmap_ilocked(ip, link);
138 }
139
140 out:
141 xfs_iunlock(ip, XFS_ILOCK_SHARED);
142 return error;
143 }
144
145 int
xfs_symlink(struct mnt_idmap * idmap,struct xfs_inode * dp,struct xfs_name * link_name,const char * target_path,umode_t mode,struct xfs_inode ** ipp)146 xfs_symlink(
147 struct mnt_idmap *idmap,
148 struct xfs_inode *dp,
149 struct xfs_name *link_name,
150 const char *target_path,
151 umode_t mode,
152 struct xfs_inode **ipp)
153 {
154 struct xfs_mount *mp = dp->i_mount;
155 struct xfs_trans *tp = NULL;
156 struct xfs_inode *ip = NULL;
157 int error = 0;
158 int pathlen;
159 bool unlock_dp_on_error = false;
160 xfs_fileoff_t first_fsb;
161 xfs_filblks_t fs_blocks;
162 int nmaps;
163 struct xfs_bmbt_irec mval[XFS_SYMLINK_MAPS];
164 xfs_daddr_t d;
165 const char *cur_chunk;
166 int byte_cnt;
167 int n;
168 struct xfs_buf *bp;
169 prid_t prid;
170 struct xfs_dquot *udqp = NULL;
171 struct xfs_dquot *gdqp = NULL;
172 struct xfs_dquot *pdqp = NULL;
173 uint resblks;
174 xfs_ino_t ino;
175
176 *ipp = NULL;
177
178 trace_xfs_symlink(dp, link_name);
179
180 if (xfs_is_shutdown(mp))
181 return -EIO;
182
183 /*
184 * Check component lengths of the target path name.
185 */
186 pathlen = strlen(target_path);
187 if (pathlen >= XFS_SYMLINK_MAXLEN) /* total string too long */
188 return -ENAMETOOLONG;
189 ASSERT(pathlen > 0);
190
191 prid = xfs_get_initial_prid(dp);
192
193 /*
194 * Make sure that we have allocated dquot(s) on disk. The uid/gid
195 * computation code must match what the VFS uses to assign i_[ug]id.
196 * INHERIT adjusts the gid computation for setgid/grpid systems.
197 */
198 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, i_user_ns(VFS_I(dp))),
199 mapped_fsgid(idmap, i_user_ns(VFS_I(dp))), prid,
200 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
201 &udqp, &gdqp, &pdqp);
202 if (error)
203 return error;
204
205 /*
206 * The symlink will fit into the inode data fork?
207 * There can't be any attributes so we get the whole variable part.
208 */
209 if (pathlen <= XFS_LITINO(mp))
210 fs_blocks = 0;
211 else
212 fs_blocks = xfs_symlink_blocks(mp, pathlen);
213 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
214
215 error = xfs_trans_alloc_icreate(mp, &M_RES(mp)->tr_symlink, udqp, gdqp,
216 pdqp, resblks, &tp);
217 if (error)
218 goto out_release_dquots;
219
220 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
221 unlock_dp_on_error = true;
222
223 /*
224 * Check whether the directory allows new symlinks or not.
225 */
226 if (dp->i_diflags & XFS_DIFLAG_NOSYMLINKS) {
227 error = -EPERM;
228 goto out_trans_cancel;
229 }
230
231 /*
232 * Allocate an inode for the symlink.
233 */
234 error = xfs_dialloc(&tp, dp->i_ino, S_IFLNK, &ino);
235 if (!error)
236 error = xfs_init_new_inode(idmap, tp, dp, ino,
237 S_IFLNK | (mode & ~S_IFMT), 1, 0, prid,
238 false, &ip);
239 if (error)
240 goto out_trans_cancel;
241
242 /*
243 * Now we join the directory inode to the transaction. We do not do it
244 * earlier because xfs_dir_ialloc might commit the previous transaction
245 * (and release all the locks). An error from here on will result in
246 * the transaction cancel unlocking dp so don't do it explicitly in the
247 * error path.
248 */
249 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
250 unlock_dp_on_error = false;
251
252 /*
253 * Also attach the dquot(s) to it, if applicable.
254 */
255 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
256
257 resblks -= XFS_IALLOC_SPACE_RES(mp);
258 /*
259 * If the symlink will fit into the inode, write it inline.
260 */
261 if (pathlen <= xfs_inode_data_fork_size(ip)) {
262 xfs_init_local_fork(ip, XFS_DATA_FORK, target_path, pathlen);
263
264 ip->i_disk_size = pathlen;
265 ip->i_df.if_format = XFS_DINODE_FMT_LOCAL;
266 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
267 } else {
268 int offset;
269
270 first_fsb = 0;
271 nmaps = XFS_SYMLINK_MAPS;
272
273 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks,
274 XFS_BMAPI_METADATA, resblks, mval, &nmaps);
275 if (error)
276 goto out_trans_cancel;
277
278 resblks -= fs_blocks;
279 ip->i_disk_size = pathlen;
280 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
281
282 cur_chunk = target_path;
283 offset = 0;
284 for (n = 0; n < nmaps; n++) {
285 char *buf;
286
287 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
288 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
289 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
290 BTOBB(byte_cnt), 0, &bp);
291 if (error)
292 goto out_trans_cancel;
293 bp->b_ops = &xfs_symlink_buf_ops;
294
295 byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt);
296 byte_cnt = min(byte_cnt, pathlen);
297
298 buf = bp->b_addr;
299 buf += xfs_symlink_hdr_set(mp, ip->i_ino, offset,
300 byte_cnt, bp);
301
302 memcpy(buf, cur_chunk, byte_cnt);
303
304 cur_chunk += byte_cnt;
305 pathlen -= byte_cnt;
306 offset += byte_cnt;
307
308 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SYMLINK_BUF);
309 xfs_trans_log_buf(tp, bp, 0, (buf + byte_cnt - 1) -
310 (char *)bp->b_addr);
311 }
312 ASSERT(pathlen == 0);
313 }
314 i_size_write(VFS_I(ip), ip->i_disk_size);
315
316 /*
317 * Create the directory entry for the symlink.
318 */
319 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, resblks);
320 if (error)
321 goto out_trans_cancel;
322 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
323 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
324
325 /*
326 * If this is a synchronous mount, make sure that the
327 * symlink transaction goes to disk before returning to
328 * the user.
329 */
330 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
331 xfs_trans_set_sync(tp);
332
333 error = xfs_trans_commit(tp);
334 if (error)
335 goto out_release_inode;
336
337 xfs_qm_dqrele(udqp);
338 xfs_qm_dqrele(gdqp);
339 xfs_qm_dqrele(pdqp);
340
341 *ipp = ip;
342 return 0;
343
344 out_trans_cancel:
345 xfs_trans_cancel(tp);
346 out_release_inode:
347 /*
348 * Wait until after the current transaction is aborted to finish the
349 * setup of the inode and release the inode. This prevents recursive
350 * transactions and deadlocks from xfs_inactive.
351 */
352 if (ip) {
353 xfs_finish_inode_setup(ip);
354 xfs_irele(ip);
355 }
356 out_release_dquots:
357 xfs_qm_dqrele(udqp);
358 xfs_qm_dqrele(gdqp);
359 xfs_qm_dqrele(pdqp);
360
361 if (unlock_dp_on_error)
362 xfs_iunlock(dp, XFS_ILOCK_EXCL);
363 return error;
364 }
365
366 /*
367 * Free a symlink that has blocks associated with it.
368 *
369 * Note: zero length symlinks are not allowed to exist. When we set the size to
370 * zero, also change it to a regular file so that it does not get written to
371 * disk as a zero length symlink. The inode is on the unlinked list already, so
372 * userspace cannot find this inode anymore, so this change is not user visible
373 * but allows us to catch corrupt zero-length symlinks in the verifiers.
374 */
375 STATIC int
xfs_inactive_symlink_rmt(struct xfs_inode * ip)376 xfs_inactive_symlink_rmt(
377 struct xfs_inode *ip)
378 {
379 struct xfs_buf *bp;
380 int done;
381 int error;
382 int i;
383 xfs_mount_t *mp;
384 xfs_bmbt_irec_t mval[XFS_SYMLINK_MAPS];
385 int nmaps;
386 int size;
387 xfs_trans_t *tp;
388
389 mp = ip->i_mount;
390 ASSERT(!xfs_need_iread_extents(&ip->i_df));
391 /*
392 * We're freeing a symlink that has some
393 * blocks allocated to it. Free the
394 * blocks here. We know that we've got
395 * either 1 or 2 extents and that we can
396 * free them all in one bunmapi call.
397 */
398 ASSERT(ip->i_df.if_nextents > 0 && ip->i_df.if_nextents <= 2);
399
400 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
401 if (error)
402 return error;
403
404 xfs_ilock(ip, XFS_ILOCK_EXCL);
405 xfs_trans_ijoin(tp, ip, 0);
406
407 /*
408 * Lock the inode, fix the size, turn it into a regular file and join it
409 * to the transaction. Hold it so in the normal path, we still have it
410 * locked for the second transaction. In the error paths we need it
411 * held so the cancel won't rele it, see below.
412 */
413 size = (int)ip->i_disk_size;
414 ip->i_disk_size = 0;
415 VFS_I(ip)->i_mode = (VFS_I(ip)->i_mode & ~S_IFMT) | S_IFREG;
416 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
417 /*
418 * Find the block(s) so we can inval and unmap them.
419 */
420 done = 0;
421 nmaps = ARRAY_SIZE(mval);
422 error = xfs_bmapi_read(ip, 0, xfs_symlink_blocks(mp, size),
423 mval, &nmaps, 0);
424 if (error)
425 goto error_trans_cancel;
426 /*
427 * Invalidate the block(s). No validation is done.
428 */
429 for (i = 0; i < nmaps; i++) {
430 error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
431 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
432 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0,
433 &bp);
434 if (error)
435 goto error_trans_cancel;
436 xfs_trans_binval(tp, bp);
437 }
438 /*
439 * Unmap the dead block(s) to the dfops.
440 */
441 error = xfs_bunmapi(tp, ip, 0, size, 0, nmaps, &done);
442 if (error)
443 goto error_trans_cancel;
444 ASSERT(done);
445
446 /*
447 * Commit the transaction. This first logs the EFI and the inode, then
448 * rolls and commits the transaction that frees the extents.
449 */
450 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
451 error = xfs_trans_commit(tp);
452 if (error) {
453 ASSERT(xfs_is_shutdown(mp));
454 goto error_unlock;
455 }
456
457 /*
458 * Remove the memory for extent descriptions (just bookkeeping).
459 */
460 if (ip->i_df.if_bytes)
461 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
462 ASSERT(ip->i_df.if_bytes == 0);
463
464 xfs_iunlock(ip, XFS_ILOCK_EXCL);
465 return 0;
466
467 error_trans_cancel:
468 xfs_trans_cancel(tp);
469 error_unlock:
470 xfs_iunlock(ip, XFS_ILOCK_EXCL);
471 return error;
472 }
473
474 /*
475 * xfs_inactive_symlink - free a symlink
476 */
477 int
xfs_inactive_symlink(struct xfs_inode * ip)478 xfs_inactive_symlink(
479 struct xfs_inode *ip)
480 {
481 struct xfs_mount *mp = ip->i_mount;
482 int pathlen;
483
484 trace_xfs_inactive_symlink(ip);
485
486 if (xfs_is_shutdown(mp))
487 return -EIO;
488
489 xfs_ilock(ip, XFS_ILOCK_EXCL);
490 pathlen = (int)ip->i_disk_size;
491 ASSERT(pathlen);
492
493 if (pathlen <= 0 || pathlen > XFS_SYMLINK_MAXLEN) {
494 xfs_alert(mp, "%s: inode (0x%llx) bad symlink length (%d)",
495 __func__, (unsigned long long)ip->i_ino, pathlen);
496 xfs_iunlock(ip, XFS_ILOCK_EXCL);
497 ASSERT(0);
498 return -EFSCORRUPTED;
499 }
500
501 /*
502 * Inline fork state gets removed by xfs_difree() so we have nothing to
503 * do here in that case.
504 */
505 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
506 xfs_iunlock(ip, XFS_ILOCK_EXCL);
507 return 0;
508 }
509
510 xfs_iunlock(ip, XFS_ILOCK_EXCL);
511
512 /* remove the remote symlink */
513 return xfs_inactive_symlink_rmt(ip);
514 }
515