blob: 5b3f5b13a008eff6b4803106a5ed42c452c99698 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080018
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_log.h"
24#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110029#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_bmap_btree.h"
31#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_dinode.h"
33#include "xfs_inode.h"
34#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_itable.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_ialloc.h"
37#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include "xfs_bmap.h"
Christoph Hellwigef14f0c2009-06-10 17:07:47 +020039#include "xfs_acl.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_attr.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_quota.h"
43#include "xfs_utils.h"
Nathan Scotta844f452005-11-02 14:38:42 +110044#include "xfs_rtalloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include "xfs_log_priv.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100047#include "xfs_filestream.h"
Christoph Hellwig993386c2007-08-28 16:12:30 +100048#include "xfs_vnodeops.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000049#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
Nathan Scott7d4fb402006-06-09 15:27:16 +100052 * The maximum pathlen is 1024 bytes. Since the minimum file system
53 * blocksize is 512 bytes, we can get a max of 2 extents back from
54 * bmapi.
55 */
56#define SYMLINK_MAPS 2
57
Christoph Hellwig804c83c2007-08-28 13:59:03 +100058STATIC int
59xfs_readlink_bmap(
60 xfs_inode_t *ip,
61 char *link)
62{
63 xfs_mount_t *mp = ip->i_mount;
64 int pathlen = ip->i_d.di_size;
65 int nmaps = SYMLINK_MAPS;
66 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
67 xfs_daddr_t d;
68 int byte_cnt;
69 int n;
70 xfs_buf_t *bp;
71 int error = 0;
72
Dave Chinner5c8ed202011-09-18 20:40:45 +000073 error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, pathlen), mval, &nmaps,
74 0);
Christoph Hellwig804c83c2007-08-28 13:59:03 +100075 if (error)
76 goto out;
77
78 for (n = 0; n < nmaps; n++) {
79 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
80 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
81
Christoph Hellwig6ad112b2009-11-24 18:02:23 +000082 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt),
Dave Chinneraa5c1582012-04-23 15:58:56 +100083 XBF_MAPPED);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +000084 if (!bp)
85 return XFS_ERROR(ENOMEM);
Chandra Seetharamane5702802011-08-03 02:18:34 +000086 error = bp->b_error;
Christoph Hellwig804c83c2007-08-28 13:59:03 +100087 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +000088 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwig804c83c2007-08-28 13:59:03 +100089 xfs_buf_relse(bp);
90 goto out;
91 }
92 if (pathlen < byte_cnt)
93 byte_cnt = pathlen;
94 pathlen -= byte_cnt;
95
Chandra Seetharaman62926042011-07-22 23:40:15 +000096 memcpy(link, bp->b_addr, byte_cnt);
Christoph Hellwig804c83c2007-08-28 13:59:03 +100097 xfs_buf_relse(bp);
98 }
99
100 link[ip->i_d.di_size] = '\0';
101 error = 0;
102
103 out:
104 return error;
105}
106
Christoph Hellwig993386c2007-08-28 16:12:30 +1000107int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108xfs_readlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000109 xfs_inode_t *ip,
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000110 char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111{
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000112 xfs_mount_t *mp = ip->i_mount;
Carlos Maiolinob52a3602011-11-07 16:10:24 +0000113 xfs_fsize_t pathlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000116 trace_xfs_readlink(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
118 if (XFS_FORCED_SHUTDOWN(mp))
119 return XFS_ERROR(EIO);
120
121 xfs_ilock(ip, XFS_ILOCK_SHARED);
122
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000123 pathlen = ip->i_d.di_size;
124 if (!pathlen)
125 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Carlos Maiolinob52a3602011-11-07 16:10:24 +0000127 if (pathlen < 0 || pathlen > MAXPATHLEN) {
128 xfs_alert(mp, "%s: inode (%llu) bad symlink length (%lld)",
129 __func__, (unsigned long long) ip->i_ino,
130 (long long) pathlen);
131 ASSERT(0);
Jan Kara9b025eb2012-01-11 18:52:10 +0000132 error = XFS_ERROR(EFSCORRUPTED);
133 goto out;
Carlos Maiolinob52a3602011-11-07 16:10:24 +0000134 }
135
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 if (ip->i_df.if_flags & XFS_IFINLINE) {
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000138 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
139 link[pathlen] = '\0';
140 } else {
141 error = xfs_readlink_bmap(ip, link);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 }
143
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000144 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 return error;
147}
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149/*
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000150 * Flags for xfs_free_eofblocks
151 */
152#define XFS_FREE_EOF_TRYLOCK (1<<0)
153
154/*
David Chinner92dfe8d2007-05-24 15:22:19 +1000155 * This is called by xfs_inactive to free any blocks beyond eof
156 * when the link count isn't zero and by xfs_dm_punch_hole() when
157 * punching a hole to EOF.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 */
Eric Sandeend96f8f82009-07-02 00:09:33 -0500159STATIC int
David Chinner92dfe8d2007-05-24 15:22:19 +1000160xfs_free_eofblocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 xfs_mount_t *mp,
David Chinner92dfe8d2007-05-24 15:22:19 +1000162 xfs_inode_t *ip,
163 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
165 xfs_trans_t *tp;
166 int error;
167 xfs_fileoff_t end_fsb;
168 xfs_fileoff_t last_fsb;
169 xfs_filblks_t map_len;
170 int nimaps;
171 xfs_bmbt_irec_t imap;
172
173 /*
174 * Figure out if there are any blocks beyond the end
175 * of the file. If not, then there is nothing to do.
176 */
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000177 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
Kulikov Vasiliy3f348852010-07-20 17:54:28 +1000179 if (last_fsb <= end_fsb)
Jesper Juhl014c2542006-01-15 02:37:08 +0100180 return 0;
Kulikov Vasiliy3f348852010-07-20 17:54:28 +1000181 map_len = last_fsb - end_fsb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183 nimaps = 1;
184 xfs_ilock(ip, XFS_ILOCK_SHARED);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000185 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 xfs_iunlock(ip, XFS_ILOCK_SHARED);
187
188 if (!error && (nimaps != 0) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +1100189 (imap.br_startblock != HOLESTARTBLOCK ||
190 ip->i_delayed_blks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 /*
192 * Attach the dquots to the inode up front.
193 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200194 error = xfs_qm_dqattach(ip, 0);
195 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +0100196 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
198 /*
199 * There are blocks after the end of file.
200 * Free them up now by truncating the file to
201 * its current size.
202 */
203 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
204
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000205 if (flags & XFS_FREE_EOF_TRYLOCK) {
206 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
207 xfs_trans_cancel(tp, 0);
208 return 0;
209 }
210 } else {
David Chinner92dfe8d2007-05-24 15:22:19 +1000211 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000212 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
214 error = xfs_trans_reserve(tp, 0,
215 XFS_ITRUNCATE_LOG_RES(mp),
216 0, XFS_TRANS_PERM_LOG_RES,
217 XFS_ITRUNCATE_LOG_COUNT);
218 if (error) {
219 ASSERT(XFS_FORCED_SHUTDOWN(mp));
220 xfs_trans_cancel(tp, 0);
221 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +0100222 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 }
224
225 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000226 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Christoph Hellwig673e8e52011-12-18 20:00:04 +0000228 /*
229 * Do not update the on-disk file size. If we update the
230 * on-disk file size and then the system crashes before the
231 * contents of the file are flushed to disk then the files
232 * may be full of holes (ie NULL files bug).
233 */
234 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000235 XFS_ISIZE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 if (error) {
Christoph Hellwig8f04c472011-07-08 14:34:34 +0200237 /*
238 * If we get an error at this point we simply don't
239 * bother truncating the file.
240 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 xfs_trans_cancel(tp,
242 (XFS_TRANS_RELEASE_LOG_RES |
243 XFS_TRANS_ABORT));
244 } else {
245 error = xfs_trans_commit(tp,
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000246 XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000248 xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
Jesper Juhl014c2542006-01-15 02:37:08 +0100250 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
253/*
254 * Free a symlink that has blocks associated with it.
255 */
256STATIC int
257xfs_inactive_symlink_rmt(
258 xfs_inode_t *ip,
259 xfs_trans_t **tpp)
260{
261 xfs_buf_t *bp;
262 int committed;
263 int done;
264 int error;
265 xfs_fsblock_t first_block;
266 xfs_bmap_free_t free_list;
267 int i;
268 xfs_mount_t *mp;
269 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
270 int nmaps;
271 xfs_trans_t *ntp;
272 int size;
273 xfs_trans_t *tp;
274
275 tp = *tpp;
276 mp = ip->i_mount;
277 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
278 /*
279 * We're freeing a symlink that has some
280 * blocks allocated to it. Free the
281 * blocks here. We know that we've got
282 * either 1 or 2 extents and that we can
283 * free them all in one bunmapi call.
284 */
285 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
286 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
287 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
288 ASSERT(XFS_FORCED_SHUTDOWN(mp));
289 xfs_trans_cancel(tp, 0);
290 *tpp = NULL;
291 return error;
292 }
293 /*
294 * Lock the inode, fix the size, and join it to the transaction.
295 * Hold it so in the normal path, we still have it locked for
296 * the second transaction. In the error paths we need it
297 * held so the cancel won't rele it, see below.
298 */
299 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
300 size = (int)ip->i_d.di_size;
301 ip->i_d.di_size = 0;
Christoph Hellwigddc34152011-09-19 15:00:54 +0000302 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
304 /*
305 * Find the block(s) so we can inval and unmap them.
306 */
307 done = 0;
Eric Sandeen9d87c312009-01-14 23:22:07 -0600308 xfs_bmap_init(&free_list, &first_block);
Tobias Klausere8c96f82006-03-24 03:15:34 -0800309 nmaps = ARRAY_SIZE(mval);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000310 error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, size),
311 mval, &nmaps, 0);
312 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 goto error0;
314 /*
315 * Invalidate the block(s).
316 */
317 for (i = 0; i < nmaps; i++) {
318 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
319 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
320 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000321 if (!bp) {
322 error = ENOMEM;
323 goto error1;
324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 xfs_trans_binval(tp, bp);
326 }
327 /*
328 * Unmap the dead block(s) to the free_list.
329 */
330 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000331 &first_block, &free_list, &done)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 goto error1;
333 ASSERT(done);
334 /*
335 * Commit the first transaction. This logs the EFI and the inode.
336 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100337 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 goto error1;
339 /*
340 * The transaction must have been committed, since there were
341 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
342 * The new tp has the extent freeing and EFDs.
343 */
344 ASSERT(committed);
345 /*
346 * The first xact was committed, so add the inode to the new one.
347 * Mark it dirty so it will be logged and moved forward in the log as
348 * part of every commit.
349 */
Christoph Hellwigddc34152011-09-19 15:00:54 +0000350 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
352 /*
353 * Get a new, empty transaction to return to our caller.
354 */
355 ntp = xfs_trans_dup(tp);
356 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000357 * Commit the transaction containing extent freeing and EFDs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 * If we get an error on the commit here or on the reserve below,
359 * we need to unlock the inode since the new transaction doesn't
360 * have the inode attached.
361 */
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000362 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 tp = ntp;
364 if (error) {
365 ASSERT(XFS_FORCED_SHUTDOWN(mp));
366 goto error0;
367 }
368 /*
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100369 * transaction commit worked ok so we can drop the extra ticket
370 * reference that we gained in xfs_trans_dup()
371 */
372 xfs_log_ticket_put(tp->t_ticket);
373
374 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * Remove the memory for extent descriptions (just bookkeeping).
376 */
377 if (ip->i_df.if_bytes)
378 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
379 ASSERT(ip->i_df.if_bytes == 0);
380 /*
381 * Put an itruncate log reservation in the new transaction
382 * for our caller.
383 */
384 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
385 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
386 ASSERT(XFS_FORCED_SHUTDOWN(mp));
387 goto error0;
388 }
389 /*
390 * Return with the inode locked but not joined to the transaction.
391 */
392 *tpp = tp;
393 return 0;
394
395 error1:
396 xfs_bmap_cancel(&free_list);
397 error0:
398 /*
399 * Have to come here with the inode locked and either
400 * (held and in the transaction) or (not in the transaction).
401 * If the inode isn't held then cancel would iput it, but
402 * that's wrong since this is inactive and the vnode ref
403 * count is 0 already.
404 * Cancel won't do anything to the inode if held, but it still
405 * needs to be locked until the cancel is done, if it was
406 * joined to the transaction.
407 */
408 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
409 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
410 *tpp = NULL;
411 return error;
412
413}
414
415STATIC int
416xfs_inactive_symlink_local(
417 xfs_inode_t *ip,
418 xfs_trans_t **tpp)
419{
420 int error;
421
422 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
423 /*
424 * We're freeing a symlink which fit into
425 * the inode. Just free the memory used
426 * to hold the old symlink.
427 */
428 error = xfs_trans_reserve(*tpp, 0,
429 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
430 0, XFS_TRANS_PERM_LOG_RES,
431 XFS_ITRUNCATE_LOG_COUNT);
432
433 if (error) {
434 xfs_trans_cancel(*tpp, 0);
435 *tpp = NULL;
Jesper Juhl014c2542006-01-15 02:37:08 +0100436 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 }
438 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
439
440 /*
441 * Zero length symlinks _can_ exist.
442 */
443 if (ip->i_df.if_bytes > 0) {
444 xfs_idata_realloc(ip,
445 -(ip->i_df.if_bytes),
446 XFS_DATA_FORK);
447 ASSERT(ip->i_df.if_bytes == 0);
448 }
Jesper Juhl014c2542006-01-15 02:37:08 +0100449 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450}
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452STATIC int
453xfs_inactive_attrs(
454 xfs_inode_t *ip,
455 xfs_trans_t **tpp)
456{
457 xfs_trans_t *tp;
458 int error;
459 xfs_mount_t *mp;
460
Christoph Hellwig579aa9ca2008-04-22 17:34:00 +1000461 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 tp = *tpp;
463 mp = ip->i_mount;
464 ASSERT(ip->i_d.di_forkoff != 0);
David Chinnere5720ee2008-04-10 12:21:18 +1000465 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 xfs_iunlock(ip, XFS_ILOCK_EXCL);
David Chinnere5720ee2008-04-10 12:21:18 +1000467 if (error)
468 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 error = xfs_attr_inactive(ip);
David Chinnere5720ee2008-04-10 12:21:18 +1000471 if (error)
472 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
475 error = xfs_trans_reserve(tp, 0,
476 XFS_IFREE_LOG_RES(mp),
477 0, XFS_TRANS_PERM_LOG_RES,
478 XFS_INACTIVE_LOG_COUNT);
David Chinnere5720ee2008-04-10 12:21:18 +1000479 if (error)
480 goto error_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000483 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
485
486 ASSERT(ip->i_d.di_anextents == 0);
487
488 *tpp = tp;
Jesper Juhl014c2542006-01-15 02:37:08 +0100489 return 0;
David Chinnere5720ee2008-04-10 12:21:18 +1000490
491error_cancel:
492 ASSERT(XFS_FORCED_SHUTDOWN(mp));
493 xfs_trans_cancel(tp, 0);
494error_unlock:
495 *tpp = NULL;
496 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
497 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498}
499
Christoph Hellwig993386c2007-08-28 16:12:30 +1000500int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501xfs_release(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000502 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503{
Christoph Hellwig993386c2007-08-28 16:12:30 +1000504 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 int error;
506
Christoph Hellwig42173f62008-04-22 17:33:25 +1000507 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000511 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return 0;
513
David Chinner2a82b8b2007-07-11 11:09:12 +1000514 if (!XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000515 int truncated;
516
David Chinner2a82b8b2007-07-11 11:09:12 +1000517 /*
518 * If we are using filestreams, and we have an unlinked
519 * file that we are processing the last close on, then nothing
520 * will be able to reopen and write to this file. Purge this
521 * inode from the filestreams cache so that it doesn't delay
522 * teardown of the inode.
523 */
524 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
525 xfs_filestream_deassociate(ip);
526
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +1000527 /*
528 * If we previously truncated this file and removed old data
529 * in the process, we want to initiate "early" writeout on
530 * the last close. This is an attempt to combat the notorious
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300531 * NULL files problem which is particularly noticeable from a
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +1000532 * truncate down, buffered (re-)write (delalloc), followed by
533 * a crash. What we are effectively doing here is
534 * significantly reducing the time window where we'd otherwise
535 * be exposed to that problem.
536 */
Christoph Hellwig09262b42007-08-29 11:44:50 +1000537 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
Dave Chinnerdf4368a2011-06-23 01:35:00 +0000538 if (truncated) {
539 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
540 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
541 xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE);
542 }
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +1000543 }
544
Dave Chinner6e857562010-12-23 12:02:31 +1100545 if (ip->i_d.di_nlink == 0)
546 return 0;
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000547
Al Viroabbede12011-07-26 02:31:30 -0400548 if ((S_ISREG(ip->i_d.di_mode) &&
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000549 (VFS_I(ip)->i_size > 0 ||
550 (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) &&
Dave Chinner6e857562010-12-23 12:02:31 +1100551 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
552 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
553
554 /*
555 * If we can't get the iolock just skip truncating the blocks
556 * past EOF because we could deadlock with the mmap_sem
557 * otherwise. We'll get another chance to drop them once the
558 * last reference to the inode is dropped, so we'll never leak
559 * blocks permanently.
560 *
561 * Further, check if the inode is being opened, written and
562 * closed frequently and we have delayed allocation blocks
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300563 * outstanding (e.g. streaming writes from the NFS server),
Dave Chinner6e857562010-12-23 12:02:31 +1100564 * truncating the blocks past EOF will cause fragmentation to
565 * occur.
566 *
567 * In this case don't do the truncation, either, but we have to
568 * be careful how we detect this case. Blocks beyond EOF show
569 * up as i_delayed_blks even when the inode is clean, so we
570 * need to truncate them away first before checking for a dirty
571 * release. Hence on the first dirty close we will still remove
572 * the speculative allocation, but after that we will leave it
573 * in place.
574 */
575 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
576 return 0;
577
578 error = xfs_free_eofblocks(mp, ip,
579 XFS_FREE_EOF_TRYLOCK);
580 if (error)
581 return error;
582
583 /* delalloc blocks after truncation means it really is dirty */
584 if (ip->i_delayed_blks)
585 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 return 0;
588}
589
590/*
591 * xfs_inactive
592 *
593 * This is called when the vnode reference count for the vnode
594 * goes to zero. If the file has been unlinked, then it must
595 * now be truncated. Also, we clear all of the read-ahead state
596 * kept for the inode here since the file is now closed.
597 */
Christoph Hellwig993386c2007-08-28 16:12:30 +1000598int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599xfs_inactive(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000600 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000602 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 xfs_fsblock_t first_block;
604 int committed;
605 xfs_trans_t *tp;
606 xfs_mount_t *mp;
607 int error;
608 int truncate;
609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 /*
611 * If the inode is already free, then there can be nothing
612 * to clean up here.
613 */
Christoph Hellwigcb4c8cc2009-03-16 08:25:25 +0100614 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 ASSERT(ip->i_df.if_real_bytes == 0);
616 ASSERT(ip->i_df.if_broot_bytes == 0);
617 return VN_INACTIVE_CACHE;
618 }
619
620 /*
621 * Only do a truncate if it's a regular file with
622 * some actual space in it. It's OK to look at the
623 * inode's fields without the lock because we're the
624 * only one with a reference to the inode.
625 */
626 truncate = ((ip->i_d.di_nlink == 0) &&
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000627 ((ip->i_d.di_size != 0) || XFS_ISIZE(ip) != 0 ||
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000628 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
Al Viroabbede12011-07-26 02:31:30 -0400629 S_ISREG(ip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
631 mp = ip->i_mount;
632
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 error = 0;
634
635 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000636 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 goto out;
638
639 if (ip->i_d.di_nlink != 0) {
Al Viroabbede12011-07-26 02:31:30 -0400640 if ((S_ISREG(ip->i_d.di_mode) &&
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000641 (VFS_I(ip)->i_size > 0 ||
642 (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) &&
643 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
644 (!(ip->i_d.di_flags &
Nathan Scottdd9f4382006-01-11 15:28:28 +1100645 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000646 ip->i_delayed_blks != 0))) {
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000647 error = xfs_free_eofblocks(mp, ip, 0);
David Chinner92dfe8d2007-05-24 15:22:19 +1000648 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +0100649 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 }
651 goto out;
652 }
653
654 ASSERT(ip->i_d.di_nlink == 0);
655
Christoph Hellwig7d095252009-06-08 15:33:32 +0200656 error = xfs_qm_dqattach(ip, 0);
657 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +0100658 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
661 if (truncate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 xfs_ilock(ip, XFS_IOLOCK_EXCL);
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 error = xfs_trans_reserve(tp, 0,
665 XFS_ITRUNCATE_LOG_RES(mp),
666 0, XFS_TRANS_PERM_LOG_RES,
667 XFS_ITRUNCATE_LOG_COUNT);
668 if (error) {
669 /* Don't call itruncate_cleanup */
670 ASSERT(XFS_FORCED_SHUTDOWN(mp));
671 xfs_trans_cancel(tp, 0);
672 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +0100673 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 }
675
676 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000677 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Christoph Hellwig673e8e52011-12-18 20:00:04 +0000679 ip->i_d.di_size = 0;
Christoph Hellwig673e8e52011-12-18 20:00:04 +0000680 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
681
682 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 if (error) {
684 xfs_trans_cancel(tp,
685 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
686 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +0100687 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 }
Christoph Hellwig673e8e52011-12-18 20:00:04 +0000689
690 ASSERT(ip->i_d.di_nextents == 0);
Al Viroabbede12011-07-26 02:31:30 -0400691 } else if (S_ISLNK(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
693 /*
694 * If we get an error while cleaning up a
695 * symlink we bail out.
696 */
697 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
698 xfs_inactive_symlink_rmt(ip, &tp) :
699 xfs_inactive_symlink_local(ip, &tp);
700
701 if (error) {
702 ASSERT(tp == NULL);
Jesper Juhl014c2542006-01-15 02:37:08 +0100703 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 }
705
Christoph Hellwigddc34152011-09-19 15:00:54 +0000706 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 } else {
708 error = xfs_trans_reserve(tp, 0,
709 XFS_IFREE_LOG_RES(mp),
710 0, XFS_TRANS_PERM_LOG_RES,
711 XFS_INACTIVE_LOG_COUNT);
712 if (error) {
713 ASSERT(XFS_FORCED_SHUTDOWN(mp));
714 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +0100715 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 }
717
718 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +0000719 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 }
721
722 /*
723 * If there are attributes associated with the file
724 * then blow them away now. The code calls a routine
725 * that recursively deconstructs the attribute fork.
726 * We need to just commit the current transaction
727 * because we can't use it for xfs_attr_inactive().
728 */
729 if (ip->i_d.di_anextents > 0) {
730 error = xfs_inactive_attrs(ip, &tp);
731 /*
732 * If we got an error, the transaction is already
733 * cancelled, and the inode is unlocked. Just get out.
734 */
735 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +0100736 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 } else if (ip->i_afp) {
738 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
739 }
740
741 /*
742 * Free the inode.
743 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600744 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 error = xfs_ifree(tp, ip, &free_list);
746 if (error) {
747 /*
748 * If we fail to free the inode, shut down. The cancel
749 * might do that, we need to make sure. Otherwise the
750 * inode might be lost for a long time or forever.
751 */
752 if (!XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100753 xfs_notice(mp, "%s: xfs_ifree returned error %d",
754 __func__, error);
Nathan Scott7d04a332006-06-09 14:58:38 +1000755 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 }
757 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
758 } else {
759 /*
760 * Credit the quota account(s). The inode is gone.
761 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200762 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
764 /*
David Chinner78e9da72008-04-10 12:24:17 +1000765 * Just ignore errors at this point. There is nothing we can
766 * do except to try to keep going. Make sure it's not a silent
767 * error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 */
David Chinner78e9da72008-04-10 12:24:17 +1000769 error = xfs_bmap_finish(&tp, &free_list, &committed);
770 if (error)
Dave Chinner53487782011-03-07 10:05:35 +1100771 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
772 __func__, error);
David Chinner78e9da72008-04-10 12:24:17 +1000773 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
774 if (error)
Dave Chinner53487782011-03-07 10:05:35 +1100775 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
776 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
Christoph Hellwig7d095252009-06-08 15:33:32 +0200778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 /*
780 * Release the dquots held by inode, if any.
781 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200782 xfs_qm_dqdetach(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
784
785 out:
786 return VN_INACTIVE_CACHE;
787}
788
Barry Naujok384f3ce2008-05-21 16:58:22 +1000789/*
790 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
791 * is allowed, otherwise it has to be an exact match. If a CI match is found,
792 * ci_name->name will point to a the actual name (caller must free) or
793 * will be set to NULL if an exact match is found.
794 */
Christoph Hellwig993386c2007-08-28 16:12:30 +1000795int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796xfs_lookup(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000797 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +1000798 struct xfs_name *name,
Barry Naujok384f3ce2008-05-21 16:58:22 +1000799 xfs_inode_t **ipp,
800 struct xfs_name *ci_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801{
Christoph Hellwigeca450b2008-04-22 17:33:52 +1000802 xfs_ino_t inum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 int error;
804 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000806 trace_xfs_lookup(dp, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
808 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
809 return XFS_ERROR(EIO);
810
811 lock_mode = xfs_ilock_map_shared(dp);
Barry Naujok384f3ce2008-05-21 16:58:22 +1000812 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 xfs_iunlock_map_shared(dp, lock_mode);
Christoph Hellwigeca450b2008-04-22 17:33:52 +1000814
815 if (error)
816 goto out;
817
Dave Chinner7b6259e2010-06-24 11:35:17 +1000818 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
Christoph Hellwigeca450b2008-04-22 17:33:52 +1000819 if (error)
Barry Naujok384f3ce2008-05-21 16:58:22 +1000820 goto out_free_name;
Christoph Hellwigeca450b2008-04-22 17:33:52 +1000821
Christoph Hellwigeca450b2008-04-22 17:33:52 +1000822 return 0;
823
Barry Naujok384f3ce2008-05-21 16:58:22 +1000824out_free_name:
825 if (ci_name)
826 kmem_free(ci_name->name);
827out:
Christoph Hellwigeca450b2008-04-22 17:33:52 +1000828 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 return error;
830}
831
Christoph Hellwig993386c2007-08-28 16:12:30 +1000832int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833xfs_create(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000834 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +1000835 struct xfs_name *name,
Al Viro576b1d62011-07-26 02:50:15 -0400836 umode_t mode,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +1000837 xfs_dev_t rdev,
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +0000838 xfs_inode_t **ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100840 int is_dir = S_ISDIR(mode);
841 struct xfs_mount *mp = dp->i_mount;
842 struct xfs_inode *ip = NULL;
843 struct xfs_trans *tp = NULL;
Barry Naujok556b8b12008-04-10 12:22:07 +1000844 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 xfs_bmap_free_t free_list;
846 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +1000847 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 uint cancel_flags;
849 int committed;
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000850 prid_t prid;
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100851 struct xfs_dquot *udqp = NULL;
852 struct xfs_dquot *gdqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 uint resblks;
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100854 uint log_res;
855 uint log_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000857 trace_xfs_create(dp, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100859 if (XFS_FORCED_SHUTDOWN(mp))
860 return XFS_ERROR(EIO);
861
Nathan Scott365ca832005-06-21 15:39:12 +1000862 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000863 prid = xfs_get_projid(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 else
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000865 prid = XFS_PROJID_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
867 /*
868 * Make sure that we have allocated dquot(s) on disk.
869 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200870 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100871 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 if (error)
Christoph Hellwigec3ba852011-02-13 13:26:42 +0000873 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100875 if (is_dir) {
876 rdev = 0;
877 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
878 log_res = XFS_MKDIR_LOG_RES(mp);
879 log_count = XFS_MKDIR_LOG_COUNT;
880 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
881 } else {
882 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
883 log_res = XFS_CREATE_LOG_RES(mp);
884 log_count = XFS_CREATE_LOG_COUNT;
885 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100889
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 /*
891 * Initially assume that the file does not exist and
892 * reserve the resources for that case. If that is not
893 * the case we'll drop the one we have and get a more
894 * appropriate transaction later.
895 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100896 error = xfs_trans_reserve(tp, resblks, log_res, 0,
897 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 if (error == ENOSPC) {
Dave Chinner153fec42009-04-06 18:48:30 +0200899 /* flush outstanding delalloc blocks and retry */
900 xfs_flush_inodes(dp);
Christoph Hellwig4734d402009-09-09 18:19:02 -0500901 error = xfs_trans_reserve(tp, resblks, log_res, 0,
902 XFS_TRANS_PERM_LOG_RES, log_count);
Dave Chinner153fec42009-04-06 18:48:30 +0200903 }
904 if (error == ENOSPC) {
905 /* No space at all so try a "no-allocation" reservation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 resblks = 0;
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100907 error = xfs_trans_reserve(tp, 0, log_res, 0,
908 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 }
910 if (error) {
911 cancel_flags = 0;
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100912 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 }
914
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000915 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +1000916 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100918 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920 /*
921 * Reserve disk quota and the inode.
922 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200923 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 if (error)
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100925 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
Barry Naujok556b8b12008-04-10 12:22:07 +1000927 error = xfs_dir_canenter(tp, dp, name, resblks);
928 if (error)
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100929 goto out_trans_cancel;
930
931 /*
932 * A newly created regular or special file just has one directory
933 * entry pointing to them, but a directory also the "." entry
934 * pointing to itself.
935 */
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +0000936 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100937 prid, resblks > 0, &ip, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 if (error) {
939 if (error == ENOSPC)
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100940 goto out_trans_cancel;
941 goto out_trans_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 /*
Christoph Hellwig993386c2007-08-28 16:12:30 +1000945 * Now we join the directory inode to the transaction. We do not do it
946 * earlier because xfs_dir_ialloc might commit the previous transaction
947 * (and release all the locks). An error from here on will result in
948 * the transaction cancel unlocking dp so don't do it explicitly in the
949 * error path.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 */
Christoph Hellwigddc34152011-09-19 15:00:54 +0000951 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +1000952 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Barry Naujok556b8b12008-04-10 12:22:07 +1000954 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000955 &first_block, &free_list, resblks ?
956 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 if (error) {
958 ASSERT(error != ENOSPC);
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100959 goto out_trans_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 }
Dave Chinnerdcd79a12010-09-28 12:27:25 +1000961 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
963
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100964 if (is_dir) {
965 error = xfs_dir_init(tp, ip, dp);
966 if (error)
967 goto out_bmap_cancel;
968
969 error = xfs_bumplink(tp, dp);
970 if (error)
971 goto out_bmap_cancel;
972 }
973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 /*
975 * If this is a synchronous mount, make sure that the
976 * create transaction goes to disk before returning to
977 * the user.
978 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100979 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 /*
983 * Attach the dquot(s) to the inodes and modify them incore.
984 * These ids of the inode couldn't have changed since the new
985 * inode has been locked ever since it was created.
986 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200987 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100989 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig517b5e82009-02-09 08:38:02 +0100990 if (error)
Christoph Hellwigec3ba852011-02-13 13:26:42 +0000991 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000993 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwigec3ba852011-02-13 13:26:42 +0000994 if (error)
995 goto out_release_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
Christoph Hellwig7d095252009-06-08 15:33:32 +0200997 xfs_qm_dqrele(udqp);
998 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001000 *ipp = ip;
Christoph Hellwig288699f2010-06-23 18:11:15 +10001001 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001003 out_bmap_cancel:
1004 xfs_bmap_cancel(&free_list);
1005 out_trans_abort:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 cancel_flags |= XFS_TRANS_ABORT;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001007 out_trans_cancel:
1008 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001009 out_release_inode:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 /*
1011 * Wait until after the current transaction is aborted to
1012 * release the inode. This prevents recursive transactions
1013 * and deadlocks from xfs_inactive.
1014 */
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001015 if (ip)
1016 IRELE(ip);
1017
1018 xfs_qm_dqrele(udqp);
1019 xfs_qm_dqrele(gdqp);
1020
1021 if (unlock_dp_on_error)
1022 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1023 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024}
1025
1026#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027int xfs_locked_n;
1028int xfs_small_retries;
1029int xfs_middle_retries;
1030int xfs_lots_retries;
1031int xfs_lock_delays;
1032#endif
1033
1034/*
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001035 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1036 * a different value
1037 */
1038static inline int
1039xfs_lock_inumorder(int lock_mode, int subclass)
1040{
1041 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10001042 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001043 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10001044 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001045
1046 return lock_mode;
1047}
1048
1049/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 * The following routine will lock n inodes in exclusive mode.
1051 * We assume the caller calls us with the inodes in i_ino order.
1052 *
1053 * We need to detect deadlock where an inode that we lock
1054 * is in the AIL and we start waiting for another inode that is locked
1055 * by a thread in a long running transaction (such as truncate). This can
1056 * result in deadlock since the long running trans might need to wait
1057 * for the inode we just locked in order to push the tail and free space
1058 * in the log.
1059 */
1060void
1061xfs_lock_inodes(
1062 xfs_inode_t **ips,
1063 int inodes,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 uint lock_mode)
1065{
1066 int attempts = 0, i, j, try_lock;
1067 xfs_log_item_t *lp;
1068
1069 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1070
Christoph Hellwigcfa853e2008-04-22 17:34:06 +10001071 try_lock = 0;
1072 i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
1074again:
1075 for (; i < inodes; i++) {
1076 ASSERT(ips[i]);
1077
1078 if (i && (ips[i] == ips[i-1])) /* Already locked */
1079 continue;
1080
1081 /*
1082 * If try_lock is not set yet, make sure all locked inodes
1083 * are not in the AIL.
1084 * If any are, set try_lock to be used later.
1085 */
1086
1087 if (!try_lock) {
1088 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1089 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1090 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1091 try_lock++;
1092 }
1093 }
1094 }
1095
1096 /*
1097 * If any of the previous locks we have locked is in the AIL,
1098 * we must TRY to get the second and subsequent locks. If
1099 * we can't get any, we must release all we have
1100 * and try again.
1101 */
1102
1103 if (try_lock) {
1104 /* try_lock must be 0 if i is 0. */
1105 /*
1106 * try_lock means we have an inode locked
1107 * that is in the AIL.
1108 */
1109 ASSERT(i != 0);
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001110 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 attempts++;
1112
1113 /*
1114 * Unlock all previous guys and try again.
1115 * xfs_iunlock will try to push the tail
1116 * if the inode is in the AIL.
1117 */
1118
1119 for(j = i - 1; j >= 0; j--) {
1120
1121 /*
1122 * Check to see if we've already
1123 * unlocked this one.
1124 * Not the first one going back,
1125 * and the inode ptr is the same.
1126 */
1127 if ((j != (i - 1)) && ips[j] ==
1128 ips[j+1])
1129 continue;
1130
1131 xfs_iunlock(ips[j], lock_mode);
1132 }
1133
1134 if ((attempts % 5) == 0) {
1135 delay(1); /* Don't just spin the CPU */
1136#ifdef DEBUG
1137 xfs_lock_delays++;
1138#endif
1139 }
1140 i = 0;
1141 try_lock = 0;
1142 goto again;
1143 }
1144 } else {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001145 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 }
1147 }
1148
1149#ifdef DEBUG
1150 if (attempts) {
1151 if (attempts < 5) xfs_small_retries++;
1152 else if (attempts < 100) xfs_middle_retries++;
1153 else xfs_lots_retries++;
1154 } else {
1155 xfs_locked_n++;
1156 }
1157#endif
1158}
1159
David Chinnerf9114eb2008-09-17 16:51:21 +10001160/*
1161 * xfs_lock_two_inodes() can only be used to lock one type of lock
1162 * at a time - the iolock or the ilock, but not both at once. If
1163 * we lock both at once, lockdep will report false positives saying
1164 * we have violated locking orders.
1165 */
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001166void
1167xfs_lock_two_inodes(
1168 xfs_inode_t *ip0,
1169 xfs_inode_t *ip1,
1170 uint lock_mode)
1171{
1172 xfs_inode_t *temp;
1173 int attempts = 0;
1174 xfs_log_item_t *lp;
1175
David Chinnerf9114eb2008-09-17 16:51:21 +10001176 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1177 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001178 ASSERT(ip0->i_ino != ip1->i_ino);
1179
1180 if (ip0->i_ino > ip1->i_ino) {
1181 temp = ip0;
1182 ip0 = ip1;
1183 ip1 = temp;
1184 }
1185
1186 again:
1187 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1188
1189 /*
1190 * If the first lock we have locked is in the AIL, we must TRY to get
1191 * the second lock. If we can't get it, we must release the first one
1192 * and try again.
1193 */
1194 lp = (xfs_log_item_t *)ip0->i_itemp;
1195 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1196 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1197 xfs_iunlock(ip0, lock_mode);
1198 if ((++attempts % 5) == 0)
1199 delay(1); /* Don't just spin the CPU */
1200 goto again;
1201 }
1202 } else {
1203 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1204 }
1205}
1206
Christoph Hellwig993386c2007-08-28 16:12:30 +10001207int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208xfs_remove(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001209 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001210 struct xfs_name *name,
1211 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001213 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 xfs_trans_t *tp = NULL;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001215 int is_dir = S_ISDIR(ip->i_d.di_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 int error = 0;
1217 xfs_bmap_free_t free_list;
1218 xfs_fsblock_t first_block;
1219 int cancel_flags;
1220 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 int link_zero;
1222 uint resblks;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001223 uint log_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001225 trace_xfs_remove(dp, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 if (XFS_FORCED_SHUTDOWN(mp))
1228 return XFS_ERROR(EIO);
1229
Christoph Hellwig7d095252009-06-08 15:33:32 +02001230 error = xfs_qm_dqattach(dp, 0);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001231 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Christoph Hellwig7d095252009-06-08 15:33:32 +02001234 error = xfs_qm_dqattach(ip, 0);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001235 if (error)
1236 goto std_return;
1237
1238 if (is_dir) {
1239 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1240 log_count = XFS_DEFAULT_LOG_COUNT;
1241 } else {
1242 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1243 log_count = XFS_REMOVE_LOG_COUNT;
1244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001246
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 /*
1248 * We try to get the real space reservation first,
1249 * allowing for directory btree deletion(s) implying
1250 * possible bmap insert(s). If we can't get the space
1251 * reservation then we use 0 instead, and avoid the bmap
1252 * btree insert(s) in the directory code by, if the bmap
1253 * insert tries to happen, instead trimming the LAST
1254 * block from the directory.
1255 */
1256 resblks = XFS_REMOVE_SPACE_RES(mp);
1257 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001258 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 if (error == ENOSPC) {
1260 resblks = 0;
1261 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001262 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 }
1264 if (error) {
1265 ASSERT(error != ENOSPC);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001266 cancel_flags = 0;
1267 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 }
1269
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001270 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Christoph Hellwigddc34152011-09-19 15:00:54 +00001272 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1273 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
1275 /*
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001276 * If we're removing a directory perform some additional validation.
1277 */
1278 if (is_dir) {
1279 ASSERT(ip->i_d.di_nlink >= 2);
1280 if (ip->i_d.di_nlink != 2) {
1281 error = XFS_ERROR(ENOTEMPTY);
1282 goto out_trans_cancel;
1283 }
1284 if (!xfs_dir_isempty(ip)) {
1285 error = XFS_ERROR(ENOTEMPTY);
1286 goto out_trans_cancel;
1287 }
1288 }
1289
Eric Sandeen9d87c312009-01-14 23:22:07 -06001290 xfs_bmap_init(&free_list, &first_block);
Barry Naujok556b8b12008-04-10 12:22:07 +10001291 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
Christoph Hellwigd4377d82008-04-22 17:33:46 +10001292 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 if (error) {
1294 ASSERT(error != ENOENT);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001295 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 }
Dave Chinnerdcd79a12010-09-28 12:27:25 +10001297 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001299 if (is_dir) {
1300 /*
1301 * Drop the link from ip's "..".
1302 */
1303 error = xfs_droplink(tp, dp);
1304 if (error)
1305 goto out_bmap_cancel;
1306
1307 /*
Christoph Hellwig2b7035f2008-10-30 17:55:18 +11001308 * Drop the "." link from ip to self.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001309 */
1310 error = xfs_droplink(tp, ip);
1311 if (error)
1312 goto out_bmap_cancel;
1313 } else {
1314 /*
1315 * When removing a non-directory we need to log the parent
Dave Chinner26c52952008-11-28 14:23:37 +11001316 * inode here. For a directory this is done implicitly
1317 * by the xfs_droplink call for the ".." entry.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001318 */
1319 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 }
1321
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001322 /*
Christoph Hellwig2b7035f2008-10-30 17:55:18 +11001323 * Drop the link from dp to ip.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001324 */
1325 error = xfs_droplink(tp, ip);
1326 if (error)
1327 goto out_bmap_cancel;
1328
1329 /*
1330 * Determine if this is the last link while
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 * we are in the transaction.
1332 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001333 link_zero = (ip->i_d.di_nlink == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 * If this is a synchronous mount, make sure that the
1337 * remove transaction goes to disk before returning to
1338 * the user.
1339 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001340 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001343 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001344 if (error)
1345 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001347 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwig5df78e72008-04-22 17:34:24 +10001348 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
1351 /*
David Chinner2a82b8b2007-07-11 11:09:12 +10001352 * If we are using filestreams, kill the stream association.
1353 * If the file is still open it may get a new one but that
1354 * will get killed on last close in xfs_close() so we don't
1355 * have to worry about that.
1356 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001357 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
David Chinner2a82b8b2007-07-11 11:09:12 +10001358 xfs_filestream_deassociate(ip);
1359
Christoph Hellwig288699f2010-06-23 18:11:15 +10001360 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001362 out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 xfs_bmap_cancel(&free_list);
1364 cancel_flags |= XFS_TRANS_ABORT;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001365 out_trans_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig288699f2010-06-23 18:11:15 +10001367 std_return:
1368 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369}
1370
Christoph Hellwig993386c2007-08-28 16:12:30 +10001371int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372xfs_link(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001373 xfs_inode_t *tdp,
Christoph Hellwiga3da7892008-03-06 13:46:12 +11001374 xfs_inode_t *sip,
Barry Naujok556b8b12008-04-10 12:22:07 +10001375 struct xfs_name *target_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001377 xfs_mount_t *mp = tdp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 int error;
1380 xfs_bmap_free_t free_list;
1381 xfs_fsblock_t first_block;
1382 int cancel_flags;
1383 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 int resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001386 trace_xfs_link(tdp, target_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Christoph Hellwiga3da7892008-03-06 13:46:12 +11001388 ASSERT(!S_ISDIR(sip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 if (XFS_FORCED_SHUTDOWN(mp))
1391 return XFS_ERROR(EIO);
1392
Christoph Hellwig7d095252009-06-08 15:33:32 +02001393 error = xfs_qm_dqattach(sip, 0);
Christoph Hellwigcb3f35b2009-02-04 09:34:20 +01001394 if (error)
1395 goto std_return;
1396
Christoph Hellwig7d095252009-06-08 15:33:32 +02001397 error = xfs_qm_dqattach(tdp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 if (error)
1399 goto std_return;
1400
1401 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1402 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Barry Naujok556b8b12008-04-10 12:22:07 +10001403 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1405 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1406 if (error == ENOSPC) {
1407 resblks = 0;
1408 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1409 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1410 }
1411 if (error) {
1412 cancel_flags = 0;
1413 goto error_return;
1414 }
1415
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001416 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Christoph Hellwigddc34152011-09-19 15:00:54 +00001418 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1419 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
1421 /*
Nathan Scott365ca832005-06-21 15:39:12 +10001422 * If we are using project inheritance, we only allow hard link
1423 * creation in our tree when the project IDs are the same; else
1424 * the tree quota mechanism could be circumvented.
1425 */
1426 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001427 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
Nathan Scottb1ecdda2006-05-08 19:51:42 +10001428 error = XFS_ERROR(EXDEV);
Nathan Scott365ca832005-06-21 15:39:12 +10001429 goto error_return;
1430 }
1431
Barry Naujok556b8b12008-04-10 12:22:07 +10001432 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1433 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 goto error_return;
1435
Eric Sandeen9d87c312009-01-14 23:22:07 -06001436 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Barry Naujok556b8b12008-04-10 12:22:07 +10001438 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1439 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (error)
1441 goto abort_return;
Dave Chinnerdcd79a12010-09-28 12:27:25 +10001442 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1444
1445 error = xfs_bumplink(tp, sip);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10001446 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 goto abort_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
1449 /*
1450 * If this is a synchronous mount, make sure that the
1451 * link transaction goes to disk before returning to
1452 * the user.
1453 */
1454 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1455 xfs_trans_set_sync(tp);
1456 }
1457
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001458 error = xfs_bmap_finish (&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 if (error) {
1460 xfs_bmap_cancel(&free_list);
1461 goto abort_return;
1462 }
1463
Christoph Hellwig288699f2010-06-23 18:11:15 +10001464 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
1466 abort_return:
1467 cancel_flags |= XFS_TRANS_ABORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 error_return:
1469 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig288699f2010-06-23 18:11:15 +10001470 std_return:
1471 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472}
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10001473
Christoph Hellwig993386c2007-08-28 16:12:30 +10001474int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475xfs_symlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001476 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001477 struct xfs_name *link_name,
1478 const char *target_path,
Al Viro576b1d62011-07-26 02:50:15 -04001479 umode_t mode,
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00001480 xfs_inode_t **ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001482 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 xfs_inode_t *ip;
1485 int error;
1486 int pathlen;
1487 xfs_bmap_free_t free_list;
1488 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001489 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 uint cancel_flags;
1491 int committed;
1492 xfs_fileoff_t first_fsb;
1493 xfs_filblks_t fs_blocks;
1494 int nmaps;
1495 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1496 xfs_daddr_t d;
Barry Naujok556b8b12008-04-10 12:22:07 +10001497 const char *cur_chunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 int byte_cnt;
1499 int n;
1500 xfs_buf_t *bp;
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001501 prid_t prid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 struct xfs_dquot *udqp, *gdqp;
1503 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504
Christoph Hellwig3937be52008-03-06 13:46:19 +11001505 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 error = 0;
1507 ip = NULL;
1508 tp = NULL;
1509
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001510 trace_xfs_symlink(dp, link_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
1512 if (XFS_FORCED_SHUTDOWN(mp))
1513 return XFS_ERROR(EIO);
1514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 /*
1516 * Check component lengths of the target path name.
1517 */
1518 pathlen = strlen(target_path);
1519 if (pathlen >= MAXPATHLEN) /* total string too long */
1520 return XFS_ERROR(ENAMETOOLONG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10001523 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001524 prid = xfs_get_projid(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 else
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001526 prid = XFS_PROJID_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
1528 /*
1529 * Make sure that we have allocated dquot(s) on disk.
1530 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001531 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
1533 if (error)
1534 goto std_return;
1535
1536 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
1537 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1538 /*
1539 * The symlink will fit into the inode data fork?
1540 * There can't be any attributes so we get the whole variable part.
1541 */
1542 if (pathlen <= XFS_LITINO(mp))
1543 fs_blocks = 0;
1544 else
1545 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
Barry Naujok556b8b12008-04-10 12:22:07 +10001546 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
1548 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
1549 if (error == ENOSPC && fs_blocks == 0) {
1550 resblks = 0;
1551 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
1552 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
1553 }
1554 if (error) {
1555 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 goto error_return;
1557 }
1558
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001559 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001560 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562 /*
1563 * Check whether the directory allows new symlinks or not.
1564 */
1565 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
1566 error = XFS_ERROR(EPERM);
1567 goto error_return;
1568 }
1569
1570 /*
1571 * Reserve disk quota : blocks and inode.
1572 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001573 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 if (error)
1575 goto error_return;
1576
1577 /*
1578 * Check for ability to enter directory entry, if no space reserved.
1579 */
Barry Naujok556b8b12008-04-10 12:22:07 +10001580 error = xfs_dir_canenter(tp, dp, link_name, resblks);
1581 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 goto error_return;
1583 /*
1584 * Initialize the bmap freelist prior to calling either
1585 * bmapi or the directory create code.
1586 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06001587 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
1589 /*
1590 * Allocate an inode for the symlink.
1591 */
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00001592 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
1593 prid, resblks > 0, &ip, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (error) {
1595 if (error == ENOSPC)
1596 goto error_return;
1597 goto error1;
1598 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
Christoph Hellwig993386c2007-08-28 16:12:30 +10001600 /*
1601 * An error after we've joined dp to the transaction will result in the
1602 * transaction cancel unlocking dp so don't do it explicitly in the
1603 * error path.
1604 */
Christoph Hellwigddc34152011-09-19 15:00:54 +00001605 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001606 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
1608 /*
1609 * Also attach the dquot(s) to it, if applicable.
1610 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001611 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
1613 if (resblks)
1614 resblks -= XFS_IALLOC_SPACE_RES(mp);
1615 /*
1616 * If the symlink will fit into the inode, write it inline.
1617 */
1618 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
1619 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
1620 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
1621 ip->i_d.di_size = pathlen;
1622
1623 /*
1624 * The inode was initially created in extent format.
1625 */
1626 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
1627 ip->i_df.if_flags |= XFS_IFINLINE;
1628
1629 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
1630 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
1631
1632 } else {
1633 first_fsb = 0;
1634 nmaps = SYMLINK_MAPS;
1635
Dave Chinnerc0dc7822011-09-18 20:40:52 +00001636 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks,
1637 XFS_BMAPI_METADATA, &first_block, resblks,
1638 mval, &nmaps, &free_list);
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001639 if (error)
1640 goto error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
1642 if (resblks)
1643 resblks -= fs_blocks;
1644 ip->i_d.di_size = pathlen;
1645 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1646
1647 cur_chunk = target_path;
1648 for (n = 0; n < nmaps; n++) {
1649 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1650 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1651 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
1652 BTOBB(byte_cnt), 0);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00001653 if (!bp) {
1654 error = ENOMEM;
1655 goto error2;
1656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 if (pathlen < byte_cnt) {
1658 byte_cnt = pathlen;
1659 }
1660 pathlen -= byte_cnt;
1661
Chandra Seetharaman62926042011-07-22 23:40:15 +00001662 memcpy(bp->b_addr, cur_chunk, byte_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 cur_chunk += byte_cnt;
1664
1665 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
1666 }
1667 }
1668
1669 /*
1670 * Create the directory entry for the symlink.
1671 */
Barry Naujok556b8b12008-04-10 12:22:07 +10001672 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
1673 &first_block, &free_list, resblks);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001674 if (error)
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001675 goto error2;
Dave Chinnerdcd79a12010-09-28 12:27:25 +10001676 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1678
1679 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 * If this is a synchronous mount, make sure that the
1681 * symlink transaction goes to disk before returning to
1682 * the user.
1683 */
1684 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1685 xfs_trans_set_sync(tp);
1686 }
1687
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001688 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 if (error) {
1690 goto error2;
1691 }
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001692 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwig7d095252009-06-08 15:33:32 +02001693 xfs_qm_dqrele(udqp);
1694 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
Christoph Hellwig288699f2010-06-23 18:11:15 +10001696 *ipp = ip;
1697 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
1699 error2:
1700 IRELE(ip);
1701 error1:
1702 xfs_bmap_cancel(&free_list);
1703 cancel_flags |= XFS_TRANS_ABORT;
1704 error_return:
1705 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig7d095252009-06-08 15:33:32 +02001706 xfs_qm_dqrele(udqp);
1707 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
Christoph Hellwig993386c2007-08-28 16:12:30 +10001709 if (unlock_dp_on_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 xfs_iunlock(dp, XFS_ILOCK_EXCL);
Christoph Hellwig288699f2010-06-23 18:11:15 +10001711 std_return:
1712 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713}
1714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715int
Christoph Hellwig993386c2007-08-28 16:12:30 +10001716xfs_set_dmattrs(
1717 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 u_int evmask,
Christoph Hellwig993386c2007-08-28 16:12:30 +10001719 u_int16_t state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001721 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 int error;
1724
1725 if (!capable(CAP_SYS_ADMIN))
1726 return XFS_ERROR(EPERM);
1727
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 if (XFS_FORCED_SHUTDOWN(mp))
1729 return XFS_ERROR(EIO);
1730
1731 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
1732 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
1733 if (error) {
1734 xfs_trans_cancel(tp, 0);
1735 return error;
1736 }
1737 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +00001738 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
Christoph Hellwig613d7042007-10-11 17:44:08 +10001740 ip->i_d.di_dmevmask = evmask;
1741 ip->i_d.di_dmstate = state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
1743 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001744 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
1746 return error;
1747}
1748
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749/*
1750 * xfs_alloc_file_space()
1751 * This routine allocates disk space for the given file.
1752 *
1753 * If alloc_type == 0, this request is for an ALLOCSP type
1754 * request which will change the file size. In this case, no
1755 * DMAPI event will be generated by the call. A TRUNCATE event
1756 * will be generated later by xfs_setattr.
1757 *
1758 * If alloc_type != 0, this request is for a RESVSP type
1759 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
1760 * lower block boundary byte address is less than the file's
1761 * length.
1762 *
1763 * RETURNS:
1764 * 0 on success
1765 * errno on error
1766 *
1767 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001768STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769xfs_alloc_file_space(
1770 xfs_inode_t *ip,
1771 xfs_off_t offset,
1772 xfs_off_t len,
1773 int alloc_type,
1774 int attr_flags)
1775{
Nathan Scottdd9f4382006-01-11 15:28:28 +11001776 xfs_mount_t *mp = ip->i_mount;
1777 xfs_off_t count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 xfs_filblks_t allocated_fsb;
1779 xfs_filblks_t allocatesize_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001780 xfs_extlen_t extsz, temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 xfs_fileoff_t startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001782 xfs_fsblock_t firstfsb;
1783 int nimaps;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001784 int quota_flag;
1785 int rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 xfs_trans_t *tp;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001787 xfs_bmbt_irec_t imaps[1], *imapp;
1788 xfs_bmap_free_t free_list;
1789 uint qblocks, resblks, resrtextents;
1790 int committed;
1791 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001793 trace_xfs_alloc_file_space(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
1795 if (XFS_FORCED_SHUTDOWN(mp))
1796 return XFS_ERROR(EIO);
1797
Christoph Hellwig7d095252009-06-08 15:33:32 +02001798 error = xfs_qm_dqattach(ip, 0);
1799 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 return error;
1801
1802 if (len <= 0)
1803 return XFS_ERROR(EINVAL);
1804
David Chinner957d0eb2007-06-18 16:50:37 +10001805 rt = XFS_IS_REALTIME_INODE(ip);
1806 extsz = xfs_get_extsz_hint(ip);
1807
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 count = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 imapp = &imaps[0];
Nathan Scottdd9f4382006-01-11 15:28:28 +11001810 nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1812 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11001815 * Allocate file space until done or until there is an error
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 while (allocatesize_fsb && !error) {
Nathan Scottdd9f4382006-01-11 15:28:28 +11001818 xfs_fileoff_t s, e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819
Nathan Scottdd9f4382006-01-11 15:28:28 +11001820 /*
Nathan Scott3ddb8fa2006-01-11 15:33:02 +11001821 * Determine space reservations for data/realtime.
Nathan Scottdd9f4382006-01-11 15:28:28 +11001822 */
1823 if (unlikely(extsz)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 s = startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001825 do_div(s, extsz);
1826 s *= extsz;
1827 e = startoffset_fsb + allocatesize_fsb;
1828 if ((temp = do_mod(startoffset_fsb, extsz)))
1829 e += temp;
1830 if ((temp = do_mod(e, extsz)))
1831 e += extsz - temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 } else {
Nathan Scottdd9f4382006-01-11 15:28:28 +11001833 s = 0;
1834 e = allocatesize_fsb;
1835 }
1836
Dave Chinner72656c42010-09-03 12:19:33 +10001837 /*
1838 * The transaction reservation is limited to a 32-bit block
1839 * count, hence we need to limit the number of blocks we are
1840 * trying to reserve to avoid an overflow. We can't allocate
1841 * more than @nimaps extents, and an extent is limited on disk
1842 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1843 */
1844 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
Nathan Scottdd9f4382006-01-11 15:28:28 +11001845 if (unlikely(rt)) {
Dave Chinner72656c42010-09-03 12:19:33 +10001846 resrtextents = qblocks = resblks;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001847 resrtextents /= mp->m_sb.sb_rextsize;
1848 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1849 quota_flag = XFS_QMOPT_RES_RTBLKS;
1850 } else {
1851 resrtextents = 0;
Dave Chinner72656c42010-09-03 12:19:33 +10001852 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
Nathan Scottdd9f4382006-01-11 15:28:28 +11001853 quota_flag = XFS_QMOPT_RES_REGBLKS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 }
1855
1856 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11001857 * Allocate and setup the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 */
1859 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Nathan Scottdd9f4382006-01-11 15:28:28 +11001860 error = xfs_trans_reserve(tp, resblks,
1861 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 XFS_TRANS_PERM_LOG_RES,
1863 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11001865 * Check for running out of space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 */
1867 if (error) {
1868 /*
1869 * Free the transaction structure.
1870 */
1871 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1872 xfs_trans_cancel(tp, 0);
1873 break;
1874 }
1875 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig7d095252009-06-08 15:33:32 +02001876 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1877 0, quota_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 if (error)
1879 goto error1;
1880
Christoph Hellwigddc34152011-09-19 15:00:54 +00001881 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
Eric Sandeen9d87c312009-01-14 23:22:07 -06001883 xfs_bmap_init(&free_list, &firstfsb);
Dave Chinnerc0dc7822011-09-18 20:40:52 +00001884 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1885 allocatesize_fsb, alloc_type, &firstfsb,
1886 0, imapp, &nimaps, &free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 if (error) {
1888 goto error0;
1889 }
1890
1891 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11001892 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001894 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 if (error) {
1896 goto error0;
1897 }
1898
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001899 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1901 if (error) {
1902 break;
1903 }
1904
1905 allocated_fsb = imapp->br_blockcount;
1906
Nathan Scottdd9f4382006-01-11 15:28:28 +11001907 if (nimaps == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 error = XFS_ERROR(ENOSPC);
1909 break;
1910 }
1911
1912 startoffset_fsb += allocated_fsb;
1913 allocatesize_fsb -= allocated_fsb;
1914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
1916 return error;
1917
Nathan Scottdd9f4382006-01-11 15:28:28 +11001918error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 xfs_bmap_cancel(&free_list);
Christoph Hellwig7d095252009-06-08 15:33:32 +02001920 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
Nathan Scottdd9f4382006-01-11 15:28:28 +11001921
1922error1: /* Just cancel transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1924 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig288699f2010-06-23 18:11:15 +10001925 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926}
1927
1928/*
1929 * Zero file bytes between startoff and endoff inclusive.
1930 * The iolock is held exclusive and no blocks are buffered.
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10001931 *
1932 * This function is used by xfs_free_file_space() to zero
1933 * partial blocks when the range to free is not block aligned.
1934 * When unreserving space with boundaries that are not block
1935 * aligned we round up the start and round down the end
1936 * boundaries and then use this function to zero the parts of
1937 * the blocks that got dropped during the rounding.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 */
1939STATIC int
1940xfs_zero_remaining_bytes(
1941 xfs_inode_t *ip,
1942 xfs_off_t startoff,
1943 xfs_off_t endoff)
1944{
1945 xfs_bmbt_irec_t imap;
1946 xfs_fileoff_t offset_fsb;
1947 xfs_off_t lastoffset;
1948 xfs_off_t offset;
1949 xfs_buf_t *bp;
1950 xfs_mount_t *mp = ip->i_mount;
1951 int nimap;
1952 int error = 0;
1953
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10001954 /*
1955 * Avoid doing I/O beyond eof - it's not necessary
1956 * since nothing can read beyond eof. The space will
1957 * be zeroed when the file is extended anyway.
1958 */
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00001959 if (startoff >= XFS_ISIZE(ip))
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10001960 return 0;
1961
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00001962 if (endoff > XFS_ISIZE(ip))
1963 endoff = XFS_ISIZE(ip);
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10001964
Dave Chinner686865f2010-09-24 20:07:47 +10001965 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1966 mp->m_rtdev_targp : mp->m_ddev_targp,
Dave Chinneraa5c1582012-04-23 15:58:56 +10001967 BTOBB(mp->m_sb.sb_blocksize), 0);
Lachlan McIlroyc6422612008-12-05 13:16:15 +11001968 if (!bp)
1969 return XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +02001971 xfs_buf_unlock(bp);
1972
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1974 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1975 nimap = 1;
Dave Chinner5c8ed202011-09-18 20:40:45 +00001976 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 if (error || nimap < 1)
1978 break;
1979 ASSERT(imap.br_blockcount >= 1);
1980 ASSERT(imap.br_startoff == offset_fsb);
1981 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1982 if (lastoffset > endoff)
1983 lastoffset = endoff;
1984 if (imap.br_startblock == HOLESTARTBLOCK)
1985 continue;
1986 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1987 if (imap.br_state == XFS_EXT_UNWRITTEN)
1988 continue;
1989 XFS_BUF_UNDONE(bp);
1990 XFS_BUF_UNWRITE(bp);
1991 XFS_BUF_READ(bp);
Eric Sandeen9d87c312009-01-14 23:22:07 -06001992 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 xfsbdstrat(mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001994 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10001995 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00001996 xfs_buf_ioerror_alert(bp,
1997 "xfs_zero_remaining_bytes(read)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 break;
1999 }
Chandra Seetharaman62926042011-07-22 23:40:15 +00002000 memset(bp->b_addr +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2002 0, lastoffset - offset + 1);
2003 XFS_BUF_UNDONE(bp);
2004 XFS_BUF_UNREAD(bp);
2005 XFS_BUF_WRITE(bp);
2006 xfsbdstrat(mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00002007 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10002008 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002009 xfs_buf_ioerror_alert(bp,
2010 "xfs_zero_remaining_bytes(write)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 break;
2012 }
2013 }
2014 xfs_buf_free(bp);
2015 return error;
2016}
2017
2018/*
2019 * xfs_free_file_space()
2020 * This routine frees disk space for the given file.
2021 *
2022 * This routine is only called by xfs_change_file_space
2023 * for an UNRESVSP type call.
2024 *
2025 * RETURNS:
2026 * 0 on success
2027 * errno on error
2028 *
2029 */
2030STATIC int
2031xfs_free_file_space(
2032 xfs_inode_t *ip,
2033 xfs_off_t offset,
2034 xfs_off_t len,
2035 int attr_flags)
2036{
2037 int committed;
2038 int done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 xfs_fileoff_t endoffset_fsb;
2040 int error;
2041 xfs_fsblock_t firstfsb;
2042 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 xfs_bmbt_irec_t imap;
2044 xfs_off_t ioffset;
2045 xfs_extlen_t mod=0;
2046 xfs_mount_t *mp;
2047 int nimap;
2048 uint resblks;
Nathan Scott673cdf52006-09-28 10:56:26 +10002049 uint rounding;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 int rt;
2051 xfs_fileoff_t startoffset_fsb;
2052 xfs_trans_t *tp;
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002053 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 mp = ip->i_mount;
2056
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10002057 trace_xfs_free_file_space(ip);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002058
Christoph Hellwig7d095252009-06-08 15:33:32 +02002059 error = xfs_qm_dqattach(ip, 0);
2060 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 return error;
2062
2063 error = 0;
2064 if (len <= 0) /* if nothing being freed */
2065 return error;
Eric Sandeen71ddabb2007-11-23 16:29:42 +11002066 rt = XFS_IS_REALTIME_INODE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
Christoph Hellwig288699f2010-06-23 18:11:15 +10002068 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002070 if (attr_flags & XFS_ATTR_NOLOCK)
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002071 need_iolock = 0;
Yingping Lu9fa80462006-03-22 12:44:35 +11002072 if (need_iolock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwig25e41b32008-12-03 12:20:39 +01002074 /* wait for the completion of any pending DIOs */
Christoph Hellwig4a06fd22011-08-23 08:28:13 +00002075 inode_dio_wait(VFS_I(ip));
Yingping Lu9fa80462006-03-22 12:44:35 +11002076 }
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002077
Tim Shimmine6a4b372007-11-23 16:30:42 +11002078 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 ioffset = offset & ~(rounding - 1);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002080
Christoph Hellwigdf80c932008-08-13 16:22:09 +10002081 if (VN_CACHED(VFS_I(ip)) != 0) {
Tim Shimmine6a4b372007-11-23 16:30:42 +11002082 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10002083 if (error)
2084 goto out_unlock_iolock;
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002085 }
2086
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 /*
2088 * Need to zero the stuff we're not freeing, on disk.
Malcolm Parsons9da096f2009-03-29 09:55:42 +02002089 * If it's a realtime file & can't use unwritten extents then we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2091 * will take care of it for us.
2092 */
Eric Sandeen62118702008-03-06 13:44:28 +11002093 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 nimap = 1;
Dave Chinner5c8ed202011-09-18 20:40:45 +00002095 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
2096 &imap, &nimap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 if (error)
2098 goto out_unlock_iolock;
2099 ASSERT(nimap == 0 || nimap == 1);
2100 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2101 xfs_daddr_t block;
2102
2103 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2104 block = imap.br_startblock;
2105 mod = do_div(block, mp->m_sb.sb_rextsize);
2106 if (mod)
2107 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
2108 }
2109 nimap = 1;
Dave Chinner5c8ed202011-09-18 20:40:45 +00002110 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
2111 &imap, &nimap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 if (error)
2113 goto out_unlock_iolock;
2114 ASSERT(nimap == 0 || nimap == 1);
2115 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2116 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2117 mod++;
2118 if (mod && (mod != mp->m_sb.sb_rextsize))
2119 endoffset_fsb -= mod;
2120 }
2121 }
2122 if ((done = (endoffset_fsb <= startoffset_fsb)))
2123 /*
2124 * One contiguous piece to clear
2125 */
2126 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2127 else {
2128 /*
2129 * Some full blocks, possibly two pieces to clear
2130 */
2131 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2132 error = xfs_zero_remaining_bytes(ip, offset,
2133 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2134 if (!error &&
2135 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2136 error = xfs_zero_remaining_bytes(ip,
2137 XFS_FSB_TO_B(mp, endoffset_fsb),
2138 offset + len - 1);
2139 }
2140
2141 /*
2142 * free file space until done or until there is an error
2143 */
2144 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2145 while (!error && !done) {
2146
2147 /*
David Chinner91ebecc2007-07-19 16:28:30 +10002148 * allocate and setup the transaction. Allow this
2149 * transaction to dip into the reserve blocks to ensure
2150 * the freeing of the space succeeds at ENOSPC.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 */
2152 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
David Chinner91ebecc2007-07-19 16:28:30 +10002153 tp->t_flags |= XFS_TRANS_RESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 error = xfs_trans_reserve(tp,
2155 resblks,
2156 XFS_WRITE_LOG_RES(mp),
2157 0,
2158 XFS_TRANS_PERM_LOG_RES,
2159 XFS_WRITE_LOG_COUNT);
2160
2161 /*
2162 * check for running out of space
2163 */
2164 if (error) {
2165 /*
2166 * Free the transaction structure.
2167 */
2168 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2169 xfs_trans_cancel(tp, 0);
2170 break;
2171 }
2172 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002173 error = xfs_trans_reserve_quota(tp, mp,
2174 ip->i_udquot, ip->i_gdquot,
2175 resblks, 0, XFS_QMOPT_RES_REGBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 if (error)
2177 goto error1;
2178
Christoph Hellwigddc34152011-09-19 15:00:54 +00002179 xfs_trans_ijoin(tp, ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180
2181 /*
2182 * issue the bunmapi() call to free the blocks
2183 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002184 xfs_bmap_init(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002185 error = xfs_bunmapi(tp, ip, startoffset_fsb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 endoffset_fsb - startoffset_fsb,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002187 0, 2, &firstfsb, &free_list, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 if (error) {
2189 goto error0;
2190 }
2191
2192 /*
2193 * complete the transaction
2194 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002195 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 if (error) {
2197 goto error0;
2198 }
2199
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002200 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2202 }
2203
2204 out_unlock_iolock:
2205 if (need_iolock)
2206 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2207 return error;
2208
2209 error0:
2210 xfs_bmap_cancel(&free_list);
2211 error1:
2212 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2213 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
2214 XFS_ILOCK_EXCL);
2215 return error;
2216}
2217
2218/*
2219 * xfs_change_file_space()
2220 * This routine allocates or frees disk space for the given file.
2221 * The user specified parameters are checked for alignment and size
2222 * limitations.
2223 *
2224 * RETURNS:
2225 * 0 on success
2226 * errno on error
2227 *
2228 */
2229int
2230xfs_change_file_space(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002231 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 int cmd,
2233 xfs_flock64_t *bf,
2234 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 int attr_flags)
2236{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002237 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 int clrprealloc;
2239 int error;
2240 xfs_fsize_t fsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 int setprealloc;
2242 xfs_off_t startoffset;
2243 xfs_off_t llen;
2244 xfs_trans_t *tp;
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002245 struct iattr iattr;
Dave Chinner44722352010-08-24 12:02:11 +10002246 int prealloc_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
Christoph Hellwig993386c2007-08-28 16:12:30 +10002248 if (!S_ISREG(ip->i_d.di_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 return XFS_ERROR(EINVAL);
2250
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 switch (bf->l_whence) {
2252 case 0: /*SEEK_SET*/
2253 break;
2254 case 1: /*SEEK_CUR*/
2255 bf->l_start += offset;
2256 break;
2257 case 2: /*SEEK_END*/
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00002258 bf->l_start += XFS_ISIZE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 break;
2260 default:
2261 return XFS_ERROR(EINVAL);
2262 }
2263
2264 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
2265
2266 if ( (bf->l_start < 0)
2267 || (bf->l_start > XFS_MAXIOFFSET(mp))
2268 || (bf->l_start + llen < 0)
2269 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
2270 return XFS_ERROR(EINVAL);
2271
2272 bf->l_whence = 0;
2273
2274 startoffset = bf->l_start;
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00002275 fsize = XFS_ISIZE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
2277 /*
2278 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
2279 * file space.
2280 * These calls do NOT zero the data space allocated to the file,
2281 * nor do they change the file size.
2282 *
2283 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
2284 * space.
2285 * These calls cause the new file data to be zeroed and the file
2286 * size to be changed.
2287 */
2288 setprealloc = clrprealloc = 0;
Dave Chinner44722352010-08-24 12:02:11 +10002289 prealloc_type = XFS_BMAPI_PREALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
2291 switch (cmd) {
Dave Chinner44722352010-08-24 12:02:11 +10002292 case XFS_IOC_ZERO_RANGE:
2293 prealloc_type |= XFS_BMAPI_CONVERT;
2294 xfs_tosspages(ip, startoffset, startoffset + bf->l_len, 0);
2295 /* FALLTHRU */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 case XFS_IOC_RESVSP:
2297 case XFS_IOC_RESVSP64:
2298 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
Dave Chinner44722352010-08-24 12:02:11 +10002299 prealloc_type, attr_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 if (error)
2301 return error;
2302 setprealloc = 1;
2303 break;
2304
2305 case XFS_IOC_UNRESVSP:
2306 case XFS_IOC_UNRESVSP64:
2307 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
2308 attr_flags)))
2309 return error;
2310 break;
2311
2312 case XFS_IOC_ALLOCSP:
2313 case XFS_IOC_ALLOCSP64:
2314 case XFS_IOC_FREESP:
2315 case XFS_IOC_FREESP64:
Dave Chinnerbc4010ec2012-04-23 15:58:57 +10002316 /*
2317 * These operations actually do IO when extending the file, but
2318 * the allocation is done seperately to the zeroing that is
2319 * done. This set of operations need to be serialised against
2320 * other IO operations, such as truncate and buffered IO. We
2321 * need to take the IOLOCK here to serialise the allocation and
2322 * zeroing IO to prevent other IOLOCK holders (e.g. getbmap,
2323 * truncate, direct IO) from racing against the transient
2324 * allocated but not written state we can have here.
2325 */
2326 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 if (startoffset > fsize) {
2328 error = xfs_alloc_file_space(ip, fsize,
Dave Chinnerbc4010ec2012-04-23 15:58:57 +10002329 startoffset - fsize, 0,
2330 attr_flags | XFS_ATTR_NOLOCK);
2331 if (error) {
2332 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 break;
Dave Chinnerbc4010ec2012-04-23 15:58:57 +10002334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 }
2336
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002337 iattr.ia_valid = ATTR_SIZE;
2338 iattr.ia_size = startoffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
Dave Chinnerbc4010ec2012-04-23 15:58:57 +10002340 error = xfs_setattr_size(ip, &iattr,
2341 attr_flags | XFS_ATTR_NOLOCK);
2342 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
2344 if (error)
2345 return error;
2346
2347 clrprealloc = 1;
2348 break;
2349
2350 default:
2351 ASSERT(0);
2352 return XFS_ERROR(EINVAL);
2353 }
2354
2355 /*
2356 * update the inode timestamp, mode, and prealloc flag bits
2357 */
2358 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
2359
2360 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
2361 0, 0, 0))) {
2362 /* ASSERT(0); */
2363 xfs_trans_cancel(tp, 0);
2364 return error;
2365 }
2366
2367 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigddc34152011-09-19 15:00:54 +00002368 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002370 if ((attr_flags & XFS_ATTR_DMI) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 ip->i_d.di_mode &= ~S_ISUID;
2372
2373 /*
2374 * Note that we don't have to worry about mandatory
2375 * file locking being disabled here because we only
2376 * clear the S_ISGID bit if the Group execute bit is
2377 * on, but if it was on then mandatory locking wouldn't
2378 * have been enabled.
2379 */
2380 if (ip->i_d.di_mode & S_IXGRP)
2381 ip->i_d.di_mode &= ~S_ISGID;
2382
Dave Chinnerdcd79a12010-09-28 12:27:25 +10002383 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 }
2385 if (setprealloc)
2386 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
2387 else if (clrprealloc)
2388 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
2389
2390 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Dave Chinner82878892011-03-26 09:13:08 +11002391 if (attr_flags & XFS_ATTR_SYNC)
2392 xfs_trans_set_sync(tp);
Christoph Hellwig23bb0be2011-09-18 20:47:51 +00002393 return xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394}