Merge branch 'uaccess-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[cascardo/linux.git] / fs / xfs / xfs_symlink.c
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * Copyright (c) 2012-2013 Red Hat, Inc.
4  * All rights reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_shared.h"
21 #include "xfs_fs.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_defer.h"
30 #include "xfs_dir2.h"
31 #include "xfs_inode.h"
32 #include "xfs_ialloc.h"
33 #include "xfs_alloc.h"
34 #include "xfs_bmap.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_bmap_util.h"
37 #include "xfs_error.h"
38 #include "xfs_quota.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_trace.h"
41 #include "xfs_symlink.h"
42 #include "xfs_trans.h"
43 #include "xfs_log.h"
44
45 /* ----- Kernel only functions below ----- */
46 STATIC int
47 xfs_readlink_bmap(
48         struct xfs_inode        *ip,
49         char                    *link)
50 {
51         struct xfs_mount        *mp = ip->i_mount;
52         struct xfs_bmbt_irec    mval[XFS_SYMLINK_MAPS];
53         struct xfs_buf          *bp;
54         xfs_daddr_t             d;
55         char                    *cur_chunk;
56         int                     pathlen = ip->i_d.di_size;
57         int                     nmaps = XFS_SYMLINK_MAPS;
58         int                     byte_cnt;
59         int                     n;
60         int                     error = 0;
61         int                     fsblocks = 0;
62         int                     offset;
63
64         fsblocks = xfs_symlink_blocks(mp, pathlen);
65         error = xfs_bmapi_read(ip, 0, fsblocks, mval, &nmaps, 0);
66         if (error)
67                 goto out;
68
69         offset = 0;
70         for (n = 0; n < nmaps; n++) {
71                 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
72                 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
73
74                 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0,
75                                   &xfs_symlink_buf_ops);
76                 if (!bp)
77                         return -ENOMEM;
78                 error = bp->b_error;
79                 if (error) {
80                         xfs_buf_ioerror_alert(bp, __func__);
81                         xfs_buf_relse(bp);
82
83                         /* bad CRC means corrupted metadata */
84                         if (error == -EFSBADCRC)
85                                 error = -EFSCORRUPTED;
86                         goto out;
87                 }
88                 byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt);
89                 if (pathlen < byte_cnt)
90                         byte_cnt = pathlen;
91
92                 cur_chunk = bp->b_addr;
93                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
94                         if (!xfs_symlink_hdr_ok(ip->i_ino, offset,
95                                                         byte_cnt, bp)) {
96                                 error = -EFSCORRUPTED;
97                                 xfs_alert(mp,
98 "symlink header does not match required off/len/owner (0x%x/Ox%x,0x%llx)",
99                                         offset, byte_cnt, ip->i_ino);
100                                 xfs_buf_relse(bp);
101                                 goto out;
102
103                         }
104
105                         cur_chunk += sizeof(struct xfs_dsymlink_hdr);
106                 }
107
108                 memcpy(link + offset, cur_chunk, byte_cnt);
109
110                 pathlen -= byte_cnt;
111                 offset += byte_cnt;
112
113                 xfs_buf_relse(bp);
114         }
115         ASSERT(pathlen == 0);
116
117         link[ip->i_d.di_size] = '\0';
118         error = 0;
119
120  out:
121         return error;
122 }
123
124 int
125 xfs_readlink(
126         struct xfs_inode *ip,
127         char            *link)
128 {
129         struct xfs_mount *mp = ip->i_mount;
130         xfs_fsize_t     pathlen;
131         int             error = 0;
132
133         trace_xfs_readlink(ip);
134
135         ASSERT(!(ip->i_df.if_flags & XFS_IFINLINE));
136
137         if (XFS_FORCED_SHUTDOWN(mp))
138                 return -EIO;
139
140         xfs_ilock(ip, XFS_ILOCK_SHARED);
141
142         pathlen = ip->i_d.di_size;
143         if (!pathlen)
144                 goto out;
145
146         if (pathlen < 0 || pathlen > MAXPATHLEN) {
147                 xfs_alert(mp, "%s: inode (%llu) bad symlink length (%lld)",
148                          __func__, (unsigned long long) ip->i_ino,
149                          (long long) pathlen);
150                 ASSERT(0);
151                 error = -EFSCORRUPTED;
152                 goto out;
153         }
154
155
156         error = xfs_readlink_bmap(ip, link);
157
158  out:
159         xfs_iunlock(ip, XFS_ILOCK_SHARED);
160         return error;
161 }
162
163 int
164 xfs_symlink(
165         struct xfs_inode        *dp,
166         struct xfs_name         *link_name,
167         const char              *target_path,
168         umode_t                 mode,
169         struct xfs_inode        **ipp)
170 {
171         struct xfs_mount        *mp = dp->i_mount;
172         struct xfs_trans        *tp = NULL;
173         struct xfs_inode        *ip = NULL;
174         int                     error = 0;
175         int                     pathlen;
176         struct xfs_defer_ops    dfops;
177         xfs_fsblock_t           first_block;
178         bool                    unlock_dp_on_error = false;
179         xfs_fileoff_t           first_fsb;
180         xfs_filblks_t           fs_blocks;
181         int                     nmaps;
182         struct xfs_bmbt_irec    mval[XFS_SYMLINK_MAPS];
183         xfs_daddr_t             d;
184         const char              *cur_chunk;
185         int                     byte_cnt;
186         int                     n;
187         xfs_buf_t               *bp;
188         prid_t                  prid;
189         struct xfs_dquot        *udqp = NULL;
190         struct xfs_dquot        *gdqp = NULL;
191         struct xfs_dquot        *pdqp = NULL;
192         uint                    resblks;
193
194         *ipp = NULL;
195
196         trace_xfs_symlink(dp, link_name);
197
198         if (XFS_FORCED_SHUTDOWN(mp))
199                 return -EIO;
200
201         /*
202          * Check component lengths of the target path name.
203          */
204         pathlen = strlen(target_path);
205         if (pathlen >= MAXPATHLEN)      /* total string too long */
206                 return -ENAMETOOLONG;
207
208         udqp = gdqp = NULL;
209         prid = xfs_get_initial_prid(dp);
210
211         /*
212          * Make sure that we have allocated dquot(s) on disk.
213          */
214         error = xfs_qm_vop_dqalloc(dp,
215                         xfs_kuid_to_uid(current_fsuid()),
216                         xfs_kgid_to_gid(current_fsgid()), prid,
217                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
218                         &udqp, &gdqp, &pdqp);
219         if (error)
220                 return error;
221
222         /*
223          * The symlink will fit into the inode data fork?
224          * There can't be any attributes so we get the whole variable part.
225          */
226         if (pathlen <= XFS_LITINO(mp, dp->i_d.di_version))
227                 fs_blocks = 0;
228         else
229                 fs_blocks = xfs_symlink_blocks(mp, pathlen);
230         resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
231
232         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_symlink, resblks, 0, 0, &tp);
233         if (error == -ENOSPC && fs_blocks == 0) {
234                 resblks = 0;
235                 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_symlink, 0, 0, 0,
236                                 &tp);
237         }
238         if (error)
239                 goto out_release_inode;
240
241         xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
242                       XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
243         unlock_dp_on_error = true;
244
245         /*
246          * Check whether the directory allows new symlinks or not.
247          */
248         if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
249                 error = -EPERM;
250                 goto out_trans_cancel;
251         }
252
253         /*
254          * Reserve disk quota : blocks and inode.
255          */
256         error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
257                                                 pdqp, resblks, 1, 0);
258         if (error)
259                 goto out_trans_cancel;
260
261         /*
262          * Check for ability to enter directory entry, if no space reserved.
263          */
264         if (!resblks) {
265                 error = xfs_dir_canenter(tp, dp, link_name);
266                 if (error)
267                         goto out_trans_cancel;
268         }
269         /*
270          * Initialize the bmap freelist prior to calling either
271          * bmapi or the directory create code.
272          */
273         xfs_defer_init(&dfops, &first_block);
274
275         /*
276          * Allocate an inode for the symlink.
277          */
278         error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
279                                prid, resblks > 0, &ip, NULL);
280         if (error)
281                 goto out_trans_cancel;
282
283         /*
284          * Now we join the directory inode to the transaction.  We do not do it
285          * earlier because xfs_dir_ialloc might commit the previous transaction
286          * (and release all the locks).  An error from here on will result in
287          * the transaction cancel unlocking dp so don't do it explicitly in the
288          * error path.
289          */
290         xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
291         unlock_dp_on_error = false;
292
293         /*
294          * Also attach the dquot(s) to it, if applicable.
295          */
296         xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
297
298         if (resblks)
299                 resblks -= XFS_IALLOC_SPACE_RES(mp);
300         /*
301          * If the symlink will fit into the inode, write it inline.
302          */
303         if (pathlen <= XFS_IFORK_DSIZE(ip)) {
304                 xfs_init_local_fork(ip, XFS_DATA_FORK, target_path, pathlen);
305
306                 ip->i_d.di_size = pathlen;
307                 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
308                 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
309         } else {
310                 int     offset;
311
312                 first_fsb = 0;
313                 nmaps = XFS_SYMLINK_MAPS;
314
315                 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks,
316                                   XFS_BMAPI_METADATA, &first_block, resblks,
317                                   mval, &nmaps, &dfops);
318                 if (error)
319                         goto out_bmap_cancel;
320
321                 if (resblks)
322                         resblks -= fs_blocks;
323                 ip->i_d.di_size = pathlen;
324                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
325
326                 cur_chunk = target_path;
327                 offset = 0;
328                 for (n = 0; n < nmaps; n++) {
329                         char    *buf;
330
331                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
332                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
333                         bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
334                                                BTOBB(byte_cnt), 0);
335                         if (!bp) {
336                                 error = -ENOMEM;
337                                 goto out_bmap_cancel;
338                         }
339                         bp->b_ops = &xfs_symlink_buf_ops;
340
341                         byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt);
342                         byte_cnt = min(byte_cnt, pathlen);
343
344                         buf = bp->b_addr;
345                         buf += xfs_symlink_hdr_set(mp, ip->i_ino, offset,
346                                                    byte_cnt, bp);
347
348                         memcpy(buf, cur_chunk, byte_cnt);
349
350                         cur_chunk += byte_cnt;
351                         pathlen -= byte_cnt;
352                         offset += byte_cnt;
353
354                         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SYMLINK_BUF);
355                         xfs_trans_log_buf(tp, bp, 0, (buf + byte_cnt - 1) -
356                                                         (char *)bp->b_addr);
357                 }
358                 ASSERT(pathlen == 0);
359         }
360
361         /*
362          * Create the directory entry for the symlink.
363          */
364         error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
365                                         &first_block, &dfops, resblks);
366         if (error)
367                 goto out_bmap_cancel;
368         xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
369         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
370
371         /*
372          * If this is a synchronous mount, make sure that the
373          * symlink transaction goes to disk before returning to
374          * the user.
375          */
376         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
377                 xfs_trans_set_sync(tp);
378         }
379
380         error = xfs_defer_finish(&tp, &dfops, NULL);
381         if (error)
382                 goto out_bmap_cancel;
383
384         error = xfs_trans_commit(tp);
385         if (error)
386                 goto out_release_inode;
387
388         xfs_qm_dqrele(udqp);
389         xfs_qm_dqrele(gdqp);
390         xfs_qm_dqrele(pdqp);
391
392         *ipp = ip;
393         return 0;
394
395 out_bmap_cancel:
396         xfs_defer_cancel(&dfops);
397 out_trans_cancel:
398         xfs_trans_cancel(tp);
399 out_release_inode:
400         /*
401          * Wait until after the current transaction is aborted to finish the
402          * setup of the inode and release the inode.  This prevents recursive
403          * transactions and deadlocks from xfs_inactive.
404          */
405         if (ip) {
406                 xfs_finish_inode_setup(ip);
407                 IRELE(ip);
408         }
409
410         xfs_qm_dqrele(udqp);
411         xfs_qm_dqrele(gdqp);
412         xfs_qm_dqrele(pdqp);
413
414         if (unlock_dp_on_error)
415                 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
416         return error;
417 }
418
419 /*
420  * Free a symlink that has blocks associated with it.
421  */
422 STATIC int
423 xfs_inactive_symlink_rmt(
424         struct xfs_inode *ip)
425 {
426         xfs_buf_t       *bp;
427         int             done;
428         int             error;
429         xfs_fsblock_t   first_block;
430         struct xfs_defer_ops    dfops;
431         int             i;
432         xfs_mount_t     *mp;
433         xfs_bmbt_irec_t mval[XFS_SYMLINK_MAPS];
434         int             nmaps;
435         int             size;
436         xfs_trans_t     *tp;
437
438         mp = ip->i_mount;
439         ASSERT(ip->i_df.if_flags & XFS_IFEXTENTS);
440         /*
441          * We're freeing a symlink that has some
442          * blocks allocated to it.  Free the
443          * blocks here.  We know that we've got
444          * either 1 or 2 extents and that we can
445          * free them all in one bunmapi call.
446          */
447         ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
448
449         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
450         if (error)
451                 return error;
452
453         xfs_ilock(ip, XFS_ILOCK_EXCL);
454         xfs_trans_ijoin(tp, ip, 0);
455
456         /*
457          * Lock the inode, fix the size, and join it to the transaction.
458          * Hold it so in the normal path, we still have it locked for
459          * the second transaction.  In the error paths we need it
460          * held so the cancel won't rele it, see below.
461          */
462         size = (int)ip->i_d.di_size;
463         ip->i_d.di_size = 0;
464         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
465         /*
466          * Find the block(s) so we can inval and unmap them.
467          */
468         done = 0;
469         xfs_defer_init(&dfops, &first_block);
470         nmaps = ARRAY_SIZE(mval);
471         error = xfs_bmapi_read(ip, 0, xfs_symlink_blocks(mp, size),
472                                 mval, &nmaps, 0);
473         if (error)
474                 goto error_trans_cancel;
475         /*
476          * Invalidate the block(s). No validation is done.
477          */
478         for (i = 0; i < nmaps; i++) {
479                 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
480                         XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
481                         XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
482                 if (!bp) {
483                         error = -ENOMEM;
484                         goto error_bmap_cancel;
485                 }
486                 xfs_trans_binval(tp, bp);
487         }
488         /*
489          * Unmap the dead block(s) to the dfops.
490          */
491         error = xfs_bunmapi(tp, ip, 0, size, 0, nmaps,
492                             &first_block, &dfops, &done);
493         if (error)
494                 goto error_bmap_cancel;
495         ASSERT(done);
496         /*
497          * Commit the first transaction.  This logs the EFI and the inode.
498          */
499         error = xfs_defer_finish(&tp, &dfops, ip);
500         if (error)
501                 goto error_bmap_cancel;
502         /*
503          * The first xact was committed, so add the inode to the new one.
504          * Mark it dirty so it will be logged and moved forward in the log as
505          * part of every commit.
506          */
507         xfs_trans_ijoin(tp, ip, 0);
508         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
509         /*
510          * Commit the transaction containing extent freeing and EFDs.
511          */
512         error = xfs_trans_commit(tp);
513         if (error) {
514                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
515                 goto error_unlock;
516         }
517
518         /*
519          * Remove the memory for extent descriptions (just bookkeeping).
520          */
521         if (ip->i_df.if_bytes)
522                 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
523         ASSERT(ip->i_df.if_bytes == 0);
524
525         xfs_iunlock(ip, XFS_ILOCK_EXCL);
526         return 0;
527
528 error_bmap_cancel:
529         xfs_defer_cancel(&dfops);
530 error_trans_cancel:
531         xfs_trans_cancel(tp);
532 error_unlock:
533         xfs_iunlock(ip, XFS_ILOCK_EXCL);
534         return error;
535 }
536
537 /*
538  * xfs_inactive_symlink - free a symlink
539  */
540 int
541 xfs_inactive_symlink(
542         struct xfs_inode        *ip)
543 {
544         struct xfs_mount        *mp = ip->i_mount;
545         int                     pathlen;
546
547         trace_xfs_inactive_symlink(ip);
548
549         if (XFS_FORCED_SHUTDOWN(mp))
550                 return -EIO;
551
552         xfs_ilock(ip, XFS_ILOCK_EXCL);
553
554         /*
555          * Zero length symlinks _can_ exist.
556          */
557         pathlen = (int)ip->i_d.di_size;
558         if (!pathlen) {
559                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
560                 return 0;
561         }
562
563         if (pathlen < 0 || pathlen > MAXPATHLEN) {
564                 xfs_alert(mp, "%s: inode (0x%llx) bad symlink length (%d)",
565                          __func__, (unsigned long long)ip->i_ino, pathlen);
566                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
567                 ASSERT(0);
568                 return -EFSCORRUPTED;
569         }
570
571         if (ip->i_df.if_flags & XFS_IFINLINE) {
572                 if (ip->i_df.if_bytes > 0) 
573                         xfs_idata_realloc(ip, -(ip->i_df.if_bytes),
574                                           XFS_DATA_FORK);
575                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
576                 ASSERT(ip->i_df.if_bytes == 0);
577                 return 0;
578         }
579
580         xfs_iunlock(ip, XFS_ILOCK_EXCL);
581
582         /* remove the remote symlink */
583         return xfs_inactive_symlink_rmt(ip);
584 }