Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / 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_dir2.h"
30 #include "xfs_inode.h"
31 #include "xfs_ialloc.h"
32 #include "xfs_alloc.h"
33 #include "xfs_bmap.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_bmap_util.h"
36 #include "xfs_error.h"
37 #include "xfs_quota.h"
38 #include "xfs_trans_space.h"
39 #include "xfs_trace.h"
40 #include "xfs_symlink.h"
41 #include "xfs_trans.h"
42 #include "xfs_log.h"
43
44 /* ----- Kernel only functions below ----- */
45 STATIC int
46 xfs_readlink_bmap(
47         struct xfs_inode        *ip,
48         char                    *link)
49 {
50         struct xfs_mount        *mp = ip->i_mount;
51         struct xfs_bmbt_irec    mval[XFS_SYMLINK_MAPS];
52         struct xfs_buf          *bp;
53         xfs_daddr_t             d;
54         char                    *cur_chunk;
55         int                     pathlen = ip->i_d.di_size;
56         int                     nmaps = XFS_SYMLINK_MAPS;
57         int                     byte_cnt;
58         int                     n;
59         int                     error = 0;
60         int                     fsblocks = 0;
61         int                     offset;
62
63         fsblocks = xfs_symlink_blocks(mp, pathlen);
64         error = xfs_bmapi_read(ip, 0, fsblocks, mval, &nmaps, 0);
65         if (error)
66                 goto out;
67
68         offset = 0;
69         for (n = 0; n < nmaps; n++) {
70                 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
71                 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
72
73                 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0,
74                                   &xfs_symlink_buf_ops);
75                 if (!bp)
76                         return -ENOMEM;
77                 error = bp->b_error;
78                 if (error) {
79                         xfs_buf_ioerror_alert(bp, __func__);
80                         xfs_buf_relse(bp);
81
82                         /* bad CRC means corrupted metadata */
83                         if (error == -EFSBADCRC)
84                                 error = -EFSCORRUPTED;
85                         goto out;
86                 }
87                 byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt);
88                 if (pathlen < byte_cnt)
89                         byte_cnt = pathlen;
90
91                 cur_chunk = bp->b_addr;
92                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
93                         if (!xfs_symlink_hdr_ok(ip->i_ino, offset,
94                                                         byte_cnt, bp)) {
95                                 error = -EFSCORRUPTED;
96                                 xfs_alert(mp,
97 "symlink header does not match required off/len/owner (0x%x/Ox%x,0x%llx)",
98                                         offset, byte_cnt, ip->i_ino);
99                                 xfs_buf_relse(bp);
100                                 goto out;
101
102                         }
103
104                         cur_chunk += sizeof(struct xfs_dsymlink_hdr);
105                 }
106
107                 memcpy(link + offset, bp->b_addr, byte_cnt);
108
109                 pathlen -= byte_cnt;
110                 offset += byte_cnt;
111
112                 xfs_buf_relse(bp);
113         }
114         ASSERT(pathlen == 0);
115
116         link[ip->i_d.di_size] = '\0';
117         error = 0;
118
119  out:
120         return error;
121 }
122
123 int
124 xfs_readlink(
125         struct xfs_inode *ip,
126         char            *link)
127 {
128         struct xfs_mount *mp = ip->i_mount;
129         xfs_fsize_t     pathlen;
130         int             error = 0;
131
132         trace_xfs_readlink(ip);
133
134         if (XFS_FORCED_SHUTDOWN(mp))
135                 return -EIO;
136
137         xfs_ilock(ip, XFS_ILOCK_SHARED);
138
139         pathlen = ip->i_d.di_size;
140         if (!pathlen)
141                 goto out;
142
143         if (pathlen < 0 || pathlen > MAXPATHLEN) {
144                 xfs_alert(mp, "%s: inode (%llu) bad symlink length (%lld)",
145                          __func__, (unsigned long long) ip->i_ino,
146                          (long long) pathlen);
147                 ASSERT(0);
148                 error = -EFSCORRUPTED;
149                 goto out;
150         }
151
152
153         if (ip->i_df.if_flags & XFS_IFINLINE) {
154                 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
155                 link[pathlen] = '\0';
156         } else {
157                 error = xfs_readlink_bmap(ip, link);
158         }
159
160  out:
161         xfs_iunlock(ip, XFS_ILOCK_SHARED);
162         return error;
163 }
164
165 int
166 xfs_symlink(
167         struct xfs_inode        *dp,
168         struct xfs_name         *link_name,
169         const char              *target_path,
170         umode_t                 mode,
171         struct xfs_inode        **ipp)
172 {
173         struct xfs_mount        *mp = dp->i_mount;
174         struct xfs_trans        *tp = NULL;
175         struct xfs_inode        *ip = NULL;
176         int                     error = 0;
177         int                     pathlen;
178         struct xfs_bmap_free    free_list;
179         xfs_fsblock_t           first_block;
180         bool                    unlock_dp_on_error = false;
181         uint                    cancel_flags;
182         int                     committed;
183         xfs_fileoff_t           first_fsb;
184         xfs_filblks_t           fs_blocks;
185         int                     nmaps;
186         struct xfs_bmbt_irec    mval[XFS_SYMLINK_MAPS];
187         xfs_daddr_t             d;
188         const char              *cur_chunk;
189         int                     byte_cnt;
190         int                     n;
191         xfs_buf_t               *bp;
192         prid_t                  prid;
193         struct xfs_dquot        *udqp = NULL;
194         struct xfs_dquot        *gdqp = NULL;
195         struct xfs_dquot        *pdqp = NULL;
196         uint                    resblks;
197
198         *ipp = NULL;
199
200         trace_xfs_symlink(dp, link_name);
201
202         if (XFS_FORCED_SHUTDOWN(mp))
203                 return -EIO;
204
205         /*
206          * Check component lengths of the target path name.
207          */
208         pathlen = strlen(target_path);
209         if (pathlen >= MAXPATHLEN)      /* total string too long */
210                 return -ENAMETOOLONG;
211
212         udqp = gdqp = NULL;
213         prid = xfs_get_initial_prid(dp);
214
215         /*
216          * Make sure that we have allocated dquot(s) on disk.
217          */
218         error = xfs_qm_vop_dqalloc(dp,
219                         xfs_kuid_to_uid(current_fsuid()),
220                         xfs_kgid_to_gid(current_fsgid()), prid,
221                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
222                         &udqp, &gdqp, &pdqp);
223         if (error)
224                 return error;
225
226         tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
227         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
228         /*
229          * The symlink will fit into the inode data fork?
230          * There can't be any attributes so we get the whole variable part.
231          */
232         if (pathlen <= XFS_LITINO(mp, dp->i_d.di_version))
233                 fs_blocks = 0;
234         else
235                 fs_blocks = xfs_symlink_blocks(mp, pathlen);
236         resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
237         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_symlink, resblks, 0);
238         if (error == -ENOSPC && fs_blocks == 0) {
239                 resblks = 0;
240                 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_symlink, 0, 0);
241         }
242         if (error) {
243                 cancel_flags = 0;
244                 goto out_trans_cancel;
245         }
246
247         xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
248         unlock_dp_on_error = true;
249
250         /*
251          * Check whether the directory allows new symlinks or not.
252          */
253         if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
254                 error = -EPERM;
255                 goto out_trans_cancel;
256         }
257
258         /*
259          * Reserve disk quota : blocks and inode.
260          */
261         error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
262                                                 pdqp, resblks, 1, 0);
263         if (error)
264                 goto out_trans_cancel;
265
266         /*
267          * Check for ability to enter directory entry, if no space reserved.
268          */
269         if (!resblks) {
270                 error = xfs_dir_canenter(tp, dp, link_name);
271                 if (error)
272                         goto out_trans_cancel;
273         }
274         /*
275          * Initialize the bmap freelist prior to calling either
276          * bmapi or the directory create code.
277          */
278         xfs_bmap_init(&free_list, &first_block);
279
280         /*
281          * Allocate an inode for the symlink.
282          */
283         error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
284                                prid, resblks > 0, &ip, NULL);
285         if (error)
286                 goto out_trans_cancel;
287
288         /*
289          * Now we join the directory inode to the transaction.  We do not do it
290          * earlier because xfs_dir_ialloc might commit the previous transaction
291          * (and release all the locks).  An error from here on will result in
292          * the transaction cancel unlocking dp so don't do it explicitly in the
293          * error path.
294          */
295         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
296         unlock_dp_on_error = false;
297
298         /*
299          * Also attach the dquot(s) to it, if applicable.
300          */
301         xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
302
303         if (resblks)
304                 resblks -= XFS_IALLOC_SPACE_RES(mp);
305         /*
306          * If the symlink will fit into the inode, write it inline.
307          */
308         if (pathlen <= XFS_IFORK_DSIZE(ip)) {
309                 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
310                 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
311                 ip->i_d.di_size = pathlen;
312
313                 /*
314                  * The inode was initially created in extent format.
315                  */
316                 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
317                 ip->i_df.if_flags |= XFS_IFINLINE;
318
319                 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
320                 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
321
322         } else {
323                 int     offset;
324
325                 first_fsb = 0;
326                 nmaps = XFS_SYMLINK_MAPS;
327
328                 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks,
329                                   XFS_BMAPI_METADATA, &first_block, resblks,
330                                   mval, &nmaps, &free_list);
331                 if (error)
332                         goto out_bmap_cancel;
333
334                 if (resblks)
335                         resblks -= fs_blocks;
336                 ip->i_d.di_size = pathlen;
337                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
338
339                 cur_chunk = target_path;
340                 offset = 0;
341                 for (n = 0; n < nmaps; n++) {
342                         char    *buf;
343
344                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
345                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
346                         bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
347                                                BTOBB(byte_cnt), 0);
348                         if (!bp) {
349                                 error = -ENOMEM;
350                                 goto out_bmap_cancel;
351                         }
352                         bp->b_ops = &xfs_symlink_buf_ops;
353
354                         byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt);
355                         byte_cnt = min(byte_cnt, pathlen);
356
357                         buf = bp->b_addr;
358                         buf += xfs_symlink_hdr_set(mp, ip->i_ino, offset,
359                                                    byte_cnt, bp);
360
361                         memcpy(buf, cur_chunk, byte_cnt);
362
363                         cur_chunk += byte_cnt;
364                         pathlen -= byte_cnt;
365                         offset += byte_cnt;
366
367                         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SYMLINK_BUF);
368                         xfs_trans_log_buf(tp, bp, 0, (buf + byte_cnt - 1) -
369                                                         (char *)bp->b_addr);
370                 }
371                 ASSERT(pathlen == 0);
372         }
373
374         /*
375          * Create the directory entry for the symlink.
376          */
377         error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
378                                         &first_block, &free_list, resblks);
379         if (error)
380                 goto out_bmap_cancel;
381         xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
382         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
383
384         /*
385          * If this is a synchronous mount, make sure that the
386          * symlink transaction goes to disk before returning to
387          * the user.
388          */
389         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
390                 xfs_trans_set_sync(tp);
391         }
392
393         error = xfs_bmap_finish(&tp, &free_list, &committed);
394         if (error)
395                 goto out_bmap_cancel;
396
397         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
398         if (error)
399                 goto out_release_inode;
400
401         xfs_qm_dqrele(udqp);
402         xfs_qm_dqrele(gdqp);
403         xfs_qm_dqrele(pdqp);
404
405         *ipp = ip;
406         return 0;
407
408 out_bmap_cancel:
409         xfs_bmap_cancel(&free_list);
410         cancel_flags |= XFS_TRANS_ABORT;
411 out_trans_cancel:
412         xfs_trans_cancel(tp, cancel_flags);
413 out_release_inode:
414         /*
415          * Wait until after the current transaction is aborted to finish the
416          * setup of the inode and release the inode.  This prevents recursive
417          * transactions and deadlocks from xfs_inactive.
418          */
419         if (ip) {
420                 xfs_finish_inode_setup(ip);
421                 IRELE(ip);
422         }
423
424         xfs_qm_dqrele(udqp);
425         xfs_qm_dqrele(gdqp);
426         xfs_qm_dqrele(pdqp);
427
428         if (unlock_dp_on_error)
429                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
430         return error;
431 }
432
433 /*
434  * Free a symlink that has blocks associated with it.
435  */
436 STATIC int
437 xfs_inactive_symlink_rmt(
438         struct xfs_inode *ip)
439 {
440         xfs_buf_t       *bp;
441         int             committed;
442         int             done;
443         int             error;
444         xfs_fsblock_t   first_block;
445         xfs_bmap_free_t free_list;
446         int             i;
447         xfs_mount_t     *mp;
448         xfs_bmbt_irec_t mval[XFS_SYMLINK_MAPS];
449         int             nmaps;
450         int             size;
451         xfs_trans_t     *tp;
452
453         mp = ip->i_mount;
454         ASSERT(ip->i_df.if_flags & XFS_IFEXTENTS);
455         /*
456          * We're freeing a symlink that has some
457          * blocks allocated to it.  Free the
458          * blocks here.  We know that we've got
459          * either 1 or 2 extents and that we can
460          * free them all in one bunmapi call.
461          */
462         ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
463
464         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
465         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
466         if (error) {
467                 xfs_trans_cancel(tp, 0);
468                 return error;
469         }
470
471         xfs_ilock(ip, XFS_ILOCK_EXCL);
472         xfs_trans_ijoin(tp, ip, 0);
473
474         /*
475          * Lock the inode, fix the size, and join it to the transaction.
476          * Hold it so in the normal path, we still have it locked for
477          * the second transaction.  In the error paths we need it
478          * held so the cancel won't rele it, see below.
479          */
480         size = (int)ip->i_d.di_size;
481         ip->i_d.di_size = 0;
482         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
483         /*
484          * Find the block(s) so we can inval and unmap them.
485          */
486         done = 0;
487         xfs_bmap_init(&free_list, &first_block);
488         nmaps = ARRAY_SIZE(mval);
489         error = xfs_bmapi_read(ip, 0, xfs_symlink_blocks(mp, size),
490                                 mval, &nmaps, 0);
491         if (error)
492                 goto error_trans_cancel;
493         /*
494          * Invalidate the block(s). No validation is done.
495          */
496         for (i = 0; i < nmaps; i++) {
497                 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
498                         XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
499                         XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
500                 if (!bp) {
501                         error = -ENOMEM;
502                         goto error_bmap_cancel;
503                 }
504                 xfs_trans_binval(tp, bp);
505         }
506         /*
507          * Unmap the dead block(s) to the free_list.
508          */
509         error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
510                             &first_block, &free_list, &done);
511         if (error)
512                 goto error_bmap_cancel;
513         ASSERT(done);
514         /*
515          * Commit the first transaction.  This logs the EFI and the inode.
516          */
517         error = xfs_bmap_finish(&tp, &free_list, &committed);
518         if (error)
519                 goto error_bmap_cancel;
520         /*
521          * The transaction must have been committed, since there were
522          * actually extents freed by xfs_bunmapi.  See xfs_bmap_finish.
523          * The new tp has the extent freeing and EFDs.
524          */
525         ASSERT(committed);
526         /*
527          * The first xact was committed, so add the inode to the new one.
528          * Mark it dirty so it will be logged and moved forward in the log as
529          * part of every commit.
530          */
531         xfs_trans_ijoin(tp, ip, 0);
532         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
533         /*
534          * Commit the transaction containing extent freeing and EFDs.
535          */
536         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
537         if (error) {
538                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
539                 goto error_unlock;
540         }
541
542         /*
543          * Remove the memory for extent descriptions (just bookkeeping).
544          */
545         if (ip->i_df.if_bytes)
546                 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
547         ASSERT(ip->i_df.if_bytes == 0);
548
549         xfs_iunlock(ip, XFS_ILOCK_EXCL);
550         return 0;
551
552 error_bmap_cancel:
553         xfs_bmap_cancel(&free_list);
554 error_trans_cancel:
555         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
556 error_unlock:
557         xfs_iunlock(ip, XFS_ILOCK_EXCL);
558         return error;
559 }
560
561 /*
562  * xfs_inactive_symlink - free a symlink
563  */
564 int
565 xfs_inactive_symlink(
566         struct xfs_inode        *ip)
567 {
568         struct xfs_mount        *mp = ip->i_mount;
569         int                     pathlen;
570
571         trace_xfs_inactive_symlink(ip);
572
573         if (XFS_FORCED_SHUTDOWN(mp))
574                 return -EIO;
575
576         xfs_ilock(ip, XFS_ILOCK_EXCL);
577
578         /*
579          * Zero length symlinks _can_ exist.
580          */
581         pathlen = (int)ip->i_d.di_size;
582         if (!pathlen) {
583                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
584                 return 0;
585         }
586
587         if (pathlen < 0 || pathlen > MAXPATHLEN) {
588                 xfs_alert(mp, "%s: inode (0x%llx) bad symlink length (%d)",
589                          __func__, (unsigned long long)ip->i_ino, pathlen);
590                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
591                 ASSERT(0);
592                 return -EFSCORRUPTED;
593         }
594
595         if (ip->i_df.if_flags & XFS_IFINLINE) {
596                 if (ip->i_df.if_bytes > 0) 
597                         xfs_idata_realloc(ip, -(ip->i_df.if_bytes),
598                                           XFS_DATA_FORK);
599                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
600                 ASSERT(ip->i_df.if_bytes == 0);
601                 return 0;
602         }
603
604         xfs_iunlock(ip, XFS_ILOCK_EXCL);
605
606         /* remove the remote symlink */
607         return xfs_inactive_symlink_rmt(ip);
608 }