libceph: move r_reply_op_{len,result} into struct ceph_osd_req_op
[linux-2.6-block.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_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, cur_chunk, 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         xfs_fileoff_t           first_fsb;
182         xfs_filblks_t           fs_blocks;
183         int                     nmaps;
184         struct xfs_bmbt_irec    mval[XFS_SYMLINK_MAPS];
185         xfs_daddr_t             d;
186         const char              *cur_chunk;
187         int                     byte_cnt;
188         int                     n;
189         xfs_buf_t               *bp;
190         prid_t                  prid;
191         struct xfs_dquot        *udqp = NULL;
192         struct xfs_dquot        *gdqp = NULL;
193         struct xfs_dquot        *pdqp = NULL;
194         uint                    resblks;
195
196         *ipp = NULL;
197
198         trace_xfs_symlink(dp, link_name);
199
200         if (XFS_FORCED_SHUTDOWN(mp))
201                 return -EIO;
202
203         /*
204          * Check component lengths of the target path name.
205          */
206         pathlen = strlen(target_path);
207         if (pathlen >= MAXPATHLEN)      /* total string too long */
208                 return -ENAMETOOLONG;
209
210         udqp = gdqp = NULL;
211         prid = xfs_get_initial_prid(dp);
212
213         /*
214          * Make sure that we have allocated dquot(s) on disk.
215          */
216         error = xfs_qm_vop_dqalloc(dp,
217                         xfs_kuid_to_uid(current_fsuid()),
218                         xfs_kgid_to_gid(current_fsgid()), prid,
219                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
220                         &udqp, &gdqp, &pdqp);
221         if (error)
222                 return error;
223
224         tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
225         /*
226          * The symlink will fit into the inode data fork?
227          * There can't be any attributes so we get the whole variable part.
228          */
229         if (pathlen <= XFS_LITINO(mp, dp->i_d.di_version))
230                 fs_blocks = 0;
231         else
232                 fs_blocks = xfs_symlink_blocks(mp, pathlen);
233         resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
234         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_symlink, resblks, 0);
235         if (error == -ENOSPC && fs_blocks == 0) {
236                 resblks = 0;
237                 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_symlink, 0, 0);
238         }
239         if (error)
240                 goto out_trans_cancel;
241
242         xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
243                       XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
244         unlock_dp_on_error = true;
245
246         /*
247          * Check whether the directory allows new symlinks or not.
248          */
249         if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
250                 error = -EPERM;
251                 goto out_trans_cancel;
252         }
253
254         /*
255          * Reserve disk quota : blocks and inode.
256          */
257         error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
258                                                 pdqp, resblks, 1, 0);
259         if (error)
260                 goto out_trans_cancel;
261
262         /*
263          * Check for ability to enter directory entry, if no space reserved.
264          */
265         if (!resblks) {
266                 error = xfs_dir_canenter(tp, dp, link_name);
267                 if (error)
268                         goto out_trans_cancel;
269         }
270         /*
271          * Initialize the bmap freelist prior to calling either
272          * bmapi or the directory create code.
273          */
274         xfs_bmap_init(&free_list, &first_block);
275
276         /*
277          * Allocate an inode for the symlink.
278          */
279         error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
280                                prid, resblks > 0, &ip, NULL);
281         if (error)
282                 goto out_trans_cancel;
283
284         /*
285          * Now we join the directory inode to the transaction.  We do not do it
286          * earlier because xfs_dir_ialloc might commit the previous transaction
287          * (and release all the locks).  An error from here on will result in
288          * the transaction cancel unlocking dp so don't do it explicitly in the
289          * error path.
290          */
291         xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
292         unlock_dp_on_error = false;
293
294         /*
295          * Also attach the dquot(s) to it, if applicable.
296          */
297         xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
298
299         if (resblks)
300                 resblks -= XFS_IALLOC_SPACE_RES(mp);
301         /*
302          * If the symlink will fit into the inode, write it inline.
303          */
304         if (pathlen <= XFS_IFORK_DSIZE(ip)) {
305                 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
306                 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
307                 ip->i_d.di_size = pathlen;
308
309                 /*
310                  * The inode was initially created in extent format.
311                  */
312                 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
313                 ip->i_df.if_flags |= XFS_IFINLINE;
314
315                 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
316                 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
317
318         } else {
319                 int     offset;
320
321                 first_fsb = 0;
322                 nmaps = XFS_SYMLINK_MAPS;
323
324                 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks,
325                                   XFS_BMAPI_METADATA, &first_block, resblks,
326                                   mval, &nmaps, &free_list);
327                 if (error)
328                         goto out_bmap_cancel;
329
330                 if (resblks)
331                         resblks -= fs_blocks;
332                 ip->i_d.di_size = pathlen;
333                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
334
335                 cur_chunk = target_path;
336                 offset = 0;
337                 for (n = 0; n < nmaps; n++) {
338                         char    *buf;
339
340                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
341                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
342                         bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
343                                                BTOBB(byte_cnt), 0);
344                         if (!bp) {
345                                 error = -ENOMEM;
346                                 goto out_bmap_cancel;
347                         }
348                         bp->b_ops = &xfs_symlink_buf_ops;
349
350                         byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt);
351                         byte_cnt = min(byte_cnt, pathlen);
352
353                         buf = bp->b_addr;
354                         buf += xfs_symlink_hdr_set(mp, ip->i_ino, offset,
355                                                    byte_cnt, bp);
356
357                         memcpy(buf, cur_chunk, byte_cnt);
358
359                         cur_chunk += byte_cnt;
360                         pathlen -= byte_cnt;
361                         offset += byte_cnt;
362
363                         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SYMLINK_BUF);
364                         xfs_trans_log_buf(tp, bp, 0, (buf + byte_cnt - 1) -
365                                                         (char *)bp->b_addr);
366                 }
367                 ASSERT(pathlen == 0);
368         }
369
370         /*
371          * Create the directory entry for the symlink.
372          */
373         error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
374                                         &first_block, &free_list, resblks);
375         if (error)
376                 goto out_bmap_cancel;
377         xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
378         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
379
380         /*
381          * If this is a synchronous mount, make sure that the
382          * symlink transaction goes to disk before returning to
383          * the user.
384          */
385         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
386                 xfs_trans_set_sync(tp);
387         }
388
389         error = xfs_bmap_finish(&tp, &free_list, NULL);
390         if (error)
391                 goto out_bmap_cancel;
392
393         error = xfs_trans_commit(tp);
394         if (error)
395                 goto out_release_inode;
396
397         xfs_qm_dqrele(udqp);
398         xfs_qm_dqrele(gdqp);
399         xfs_qm_dqrele(pdqp);
400
401         *ipp = ip;
402         return 0;
403
404 out_bmap_cancel:
405         xfs_bmap_cancel(&free_list);
406 out_trans_cancel:
407         xfs_trans_cancel(tp);
408 out_release_inode:
409         /*
410          * Wait until after the current transaction is aborted to finish the
411          * setup of the inode and release the inode.  This prevents recursive
412          * transactions and deadlocks from xfs_inactive.
413          */
414         if (ip) {
415                 xfs_finish_inode_setup(ip);
416                 IRELE(ip);
417         }
418
419         xfs_qm_dqrele(udqp);
420         xfs_qm_dqrele(gdqp);
421         xfs_qm_dqrele(pdqp);
422
423         if (unlock_dp_on_error)
424                 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
425         return error;
426 }
427
428 /*
429  * Free a symlink that has blocks associated with it.
430  */
431 STATIC int
432 xfs_inactive_symlink_rmt(
433         struct xfs_inode *ip)
434 {
435         xfs_buf_t       *bp;
436         int             done;
437         int             error;
438         xfs_fsblock_t   first_block;
439         xfs_bmap_free_t free_list;
440         int             i;
441         xfs_mount_t     *mp;
442         xfs_bmbt_irec_t mval[XFS_SYMLINK_MAPS];
443         int             nmaps;
444         int             size;
445         xfs_trans_t     *tp;
446
447         mp = ip->i_mount;
448         ASSERT(ip->i_df.if_flags & XFS_IFEXTENTS);
449         /*
450          * We're freeing a symlink that has some
451          * blocks allocated to it.  Free the
452          * blocks here.  We know that we've got
453          * either 1 or 2 extents and that we can
454          * free them all in one bunmapi call.
455          */
456         ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
457
458         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
459         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
460         if (error) {
461                 xfs_trans_cancel(tp);
462                 return error;
463         }
464
465         xfs_ilock(ip, XFS_ILOCK_EXCL);
466         xfs_trans_ijoin(tp, ip, 0);
467
468         /*
469          * Lock the inode, fix the size, and join it to the transaction.
470          * Hold it so in the normal path, we still have it locked for
471          * the second transaction.  In the error paths we need it
472          * held so the cancel won't rele it, see below.
473          */
474         size = (int)ip->i_d.di_size;
475         ip->i_d.di_size = 0;
476         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
477         /*
478          * Find the block(s) so we can inval and unmap them.
479          */
480         done = 0;
481         xfs_bmap_init(&free_list, &first_block);
482         nmaps = ARRAY_SIZE(mval);
483         error = xfs_bmapi_read(ip, 0, xfs_symlink_blocks(mp, size),
484                                 mval, &nmaps, 0);
485         if (error)
486                 goto error_trans_cancel;
487         /*
488          * Invalidate the block(s). No validation is done.
489          */
490         for (i = 0; i < nmaps; i++) {
491                 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
492                         XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
493                         XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
494                 if (!bp) {
495                         error = -ENOMEM;
496                         goto error_bmap_cancel;
497                 }
498                 xfs_trans_binval(tp, bp);
499         }
500         /*
501          * Unmap the dead block(s) to the free_list.
502          */
503         error = xfs_bunmapi(tp, ip, 0, size, 0, nmaps,
504                             &first_block, &free_list, &done);
505         if (error)
506                 goto error_bmap_cancel;
507         ASSERT(done);
508         /*
509          * Commit the first transaction.  This logs the EFI and the inode.
510          */
511         error = xfs_bmap_finish(&tp, &free_list, ip);
512         if (error)
513                 goto error_bmap_cancel;
514         /*
515          * The first xact was committed, so add the inode to the new one.
516          * Mark it dirty so it will be logged and moved forward in the log as
517          * part of every commit.
518          */
519         xfs_trans_ijoin(tp, ip, 0);
520         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
521         /*
522          * Commit the transaction containing extent freeing and EFDs.
523          */
524         error = xfs_trans_commit(tp);
525         if (error) {
526                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
527                 goto error_unlock;
528         }
529
530         /*
531          * Remove the memory for extent descriptions (just bookkeeping).
532          */
533         if (ip->i_df.if_bytes)
534                 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
535         ASSERT(ip->i_df.if_bytes == 0);
536
537         xfs_iunlock(ip, XFS_ILOCK_EXCL);
538         return 0;
539
540 error_bmap_cancel:
541         xfs_bmap_cancel(&free_list);
542 error_trans_cancel:
543         xfs_trans_cancel(tp);
544 error_unlock:
545         xfs_iunlock(ip, XFS_ILOCK_EXCL);
546         return error;
547 }
548
549 /*
550  * xfs_inactive_symlink - free a symlink
551  */
552 int
553 xfs_inactive_symlink(
554         struct xfs_inode        *ip)
555 {
556         struct xfs_mount        *mp = ip->i_mount;
557         int                     pathlen;
558
559         trace_xfs_inactive_symlink(ip);
560
561         if (XFS_FORCED_SHUTDOWN(mp))
562                 return -EIO;
563
564         xfs_ilock(ip, XFS_ILOCK_EXCL);
565
566         /*
567          * Zero length symlinks _can_ exist.
568          */
569         pathlen = (int)ip->i_d.di_size;
570         if (!pathlen) {
571                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
572                 return 0;
573         }
574
575         if (pathlen < 0 || pathlen > MAXPATHLEN) {
576                 xfs_alert(mp, "%s: inode (0x%llx) bad symlink length (%d)",
577                          __func__, (unsigned long long)ip->i_ino, pathlen);
578                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
579                 ASSERT(0);
580                 return -EFSCORRUPTED;
581         }
582
583         if (ip->i_df.if_flags & XFS_IFINLINE) {
584                 if (ip->i_df.if_bytes > 0) 
585                         xfs_idata_realloc(ip, -(ip->i_df.if_bytes),
586                                           XFS_DATA_FORK);
587                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
588                 ASSERT(ip->i_df.if_bytes == 0);
589                 return 0;
590         }
591
592         xfs_iunlock(ip, XFS_ILOCK_EXCL);
593
594         /* remove the remote symlink */
595         return xfs_inactive_symlink_rmt(ip);
596 }