blob: 45d740df53b73d2166fe92490e68ea004ec83a7e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * 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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
Nathan Scotta844f452005-11-02 14:38:42 +110026#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
30#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_bmap.h"
37#include "xfs_error.h"
38#include "xfs_quota.h"
39#include "xfs_rw.h"
40#include "xfs_itable.h"
41#include "xfs_utils.h"
42
43/*
44 * xfs_get_dir_entry is used to get a reference to an inode given
45 * its parent directory inode and the name of the file. It does
46 * not lock the child inode, and it unlocks the directory before
47 * returning. The directory's generation number is returned for
48 * use by a later call to xfs_lock_dir_and_entry.
49 */
50int
51xfs_get_dir_entry(
Nathan Scott8285fb52006-06-09 17:07:12 +100052 bhv_vname_t *dentry,
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 xfs_inode_t **ipp)
54{
Nathan Scott67fcaa72006-06-09 17:00:52 +100055 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57 vp = VNAME_TO_VNODE(dentry);
Christoph Hellwig75e17b32006-01-11 20:58:44 +110058
59 *ipp = xfs_vtoi(vp);
60 if (!*ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 return XFS_ERROR(ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 VN_HOLD(vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 return 0;
64}
65
66int
67xfs_dir_lookup_int(
Christoph Hellwig993386c2007-08-28 16:12:30 +100068 xfs_inode_t *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 uint lock_mode,
Nathan Scott8285fb52006-06-09 17:07:12 +100070 bhv_vname_t *dentry,
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 xfs_ino_t *inum,
72 xfs_inode_t **ipp)
73{
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 int error;
75
Lachlan McIlroycf441ee2008-02-07 16:42:19 +110076 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Nathan Scottf6c2d1f2006-06-20 13:04:51 +100078 error = xfs_dir_lookup(NULL, dp, VNAME(dentry), VNAMELEN(dentry), inum);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 if (!error) {
80 /*
81 * Unlock the directory. We do this because we can't
82 * hold the directory lock while doing the vn_get()
83 * in xfs_iget(). Doing so could cause us to hold
84 * a lock while waiting for the inode to finish
85 * being inactive while it's waiting for a log
86 * reservation in the inactive routine.
87 */
88 xfs_iunlock(dp, lock_mode);
89 error = xfs_iget(dp->i_mount, NULL, *inum, 0, 0, ipp, 0);
90 xfs_ilock(dp, lock_mode);
91
92 if (error) {
93 *ipp = NULL;
94 } else if ((*ipp)->i_d.di_mode == 0) {
95 /*
96 * The inode has been freed. Something is
97 * wrong so just get out of here.
98 */
99 xfs_iunlock(dp, lock_mode);
100 xfs_iput_new(*ipp, 0);
101 *ipp = NULL;
102 xfs_ilock(dp, lock_mode);
103 error = XFS_ERROR(ENOENT);
104 }
105 }
106 return error;
107}
108
109/*
110 * Allocates a new inode from disk and return a pointer to the
111 * incore copy. This routine will internally commit the current
112 * transaction and allocate a new one if the Space Manager needed
113 * to do an allocation to replenish the inode free-list.
114 *
115 * This routine is designed to be called from xfs_create and
116 * xfs_create_dir.
117 *
118 */
119int
120xfs_dir_ialloc(
121 xfs_trans_t **tpp, /* input: current transaction;
122 output: may be a new transaction. */
123 xfs_inode_t *dp, /* directory within whose allocate
124 the inode. */
125 mode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -0700126 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 xfs_dev_t rdev,
128 cred_t *credp,
129 prid_t prid, /* project id */
130 int okalloc, /* ok to allocate new space */
131 xfs_inode_t **ipp, /* pointer to inode; it will be
132 locked. */
133 int *committed)
134
135{
136 xfs_trans_t *tp;
137 xfs_trans_t *ntp;
138 xfs_inode_t *ip;
139 xfs_buf_t *ialloc_context = NULL;
140 boolean_t call_again = B_FALSE;
141 int code;
142 uint log_res;
143 uint log_count;
144 void *dqinfo;
145 uint tflags;
146
147 tp = *tpp;
148 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
149
150 /*
151 * xfs_ialloc will return a pointer to an incore inode if
152 * the Space Manager has an available inode on the free
153 * list. Otherwise, it will do an allocation and replenish
154 * the freelist. Since we can only do one allocation per
155 * transaction without deadlocks, we will need to commit the
156 * current transaction and start a new one. We will then
157 * need to call xfs_ialloc again to get the inode.
158 *
159 * If xfs_ialloc did an allocation to replenish the freelist,
160 * it returns the bp containing the head of the freelist as
161 * ialloc_context. We will hold a lock on it across the
162 * transaction commit so that no other process can steal
163 * the inode(s) that we've just allocated.
164 */
165 code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid, okalloc,
166 &ialloc_context, &call_again, &ip);
167
168 /*
169 * Return an error if we were unable to allocate a new inode.
170 * This should only happen if we run out of space on disk or
171 * encounter a disk error.
172 */
173 if (code) {
174 *ipp = NULL;
175 return code;
176 }
177 if (!call_again && (ip == NULL)) {
178 *ipp = NULL;
179 return XFS_ERROR(ENOSPC);
180 }
181
182 /*
183 * If call_again is set, then we were unable to get an
184 * inode in one operation. We need to commit the current
185 * transaction and call xfs_ialloc() again. It is guaranteed
186 * to succeed the second time.
187 */
188 if (call_again) {
189
190 /*
191 * Normally, xfs_trans_commit releases all the locks.
192 * We call bhold to hang on to the ialloc_context across
193 * the commit. Holding this buffer prevents any other
194 * processes from doing any allocations in this
195 * allocation group.
196 */
197 xfs_trans_bhold(tp, ialloc_context);
198 /*
199 * Save the log reservation so we can use
200 * them in the next transaction.
201 */
202 log_res = xfs_trans_get_log_res(tp);
203 log_count = xfs_trans_get_log_count(tp);
204
205 /*
206 * We want the quota changes to be associated with the next
207 * transaction, NOT this one. So, detach the dqinfo from this
208 * and attach it to the next transaction.
209 */
210 dqinfo = NULL;
211 tflags = 0;
212 if (tp->t_dqinfo) {
213 dqinfo = (void *)tp->t_dqinfo;
214 tp->t_dqinfo = NULL;
215 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
216 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
217 }
218
219 ntp = xfs_trans_dup(tp);
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000220 code = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 tp = ntp;
222 if (committed != NULL) {
223 *committed = 1;
224 }
225 /*
226 * If we get an error during the commit processing,
227 * release the buffer that is still held and return
228 * to the caller.
229 */
230 if (code) {
231 xfs_buf_relse(ialloc_context);
232 if (dqinfo) {
233 tp->t_dqinfo = dqinfo;
234 XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp);
235 }
236 *tpp = ntp;
237 *ipp = NULL;
238 return code;
239 }
240 code = xfs_trans_reserve(tp, 0, log_res, 0,
241 XFS_TRANS_PERM_LOG_RES, log_count);
242 /*
243 * Re-attach the quota info that we detached from prev trx.
244 */
245 if (dqinfo) {
246 tp->t_dqinfo = dqinfo;
247 tp->t_flags |= tflags;
248 }
249
250 if (code) {
251 xfs_buf_relse(ialloc_context);
252 *tpp = ntp;
253 *ipp = NULL;
254 return code;
255 }
256 xfs_trans_bjoin(tp, ialloc_context);
257
258 /*
259 * Call ialloc again. Since we've locked out all
260 * other allocations in this allocation group,
261 * this call should always succeed.
262 */
263 code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid,
264 okalloc, &ialloc_context, &call_again, &ip);
265
266 /*
267 * If we get an error at this point, return to the caller
268 * so that the current transaction can be aborted.
269 */
270 if (code) {
271 *tpp = tp;
272 *ipp = NULL;
273 return code;
274 }
275 ASSERT ((!call_again) && (ip != NULL));
276
277 } else {
278 if (committed != NULL) {
279 *committed = 0;
280 }
281 }
282
283 *ipp = ip;
284 *tpp = tp;
285
286 return 0;
287}
288
289/*
290 * Decrement the link count on an inode & log the change.
291 * If this causes the link count to go to zero, initiate the
292 * logging activity required to truncate a file.
293 */
294int /* error */
295xfs_droplink(
296 xfs_trans_t *tp,
297 xfs_inode_t *ip)
298{
299 int error;
300
301 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
302
303 ASSERT (ip->i_d.di_nlink > 0);
304 ip->i_d.di_nlink--;
Christoph Hellwig4188c782008-02-05 12:13:53 +1100305 drop_nlink(ip->i_vnode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
307
308 error = 0;
309 if (ip->i_d.di_nlink == 0) {
310 /*
311 * We're dropping the last link to this file.
312 * Move the on-disk inode to the AGI unlinked list.
313 * From xfs_inactive() we will pull the inode from
314 * the list and free it.
315 */
316 error = xfs_iunlink(tp, ip);
317 }
318 return error;
319}
320
321/*
322 * This gets called when the inode's version needs to be changed from 1 to 2.
323 * Currently this happens when the nlink field overflows the old 16-bit value
324 * or when chproj is called to change the project for the first time.
325 * As a side effect the superblock version will also get rev'd
326 * to contain the NLINK bit.
327 */
328void
329xfs_bump_ino_vers2(
330 xfs_trans_t *tp,
331 xfs_inode_t *ip)
332{
333 xfs_mount_t *mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
336 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1);
337
338 ip->i_d.di_version = XFS_DINODE_VERSION_2;
339 ip->i_d.di_onlink = 0;
340 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
341 mp = tp->t_mountp;
342 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000343 spin_lock(&mp->m_sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
345 XFS_SB_VERSION_ADDNLINK(&mp->m_sb);
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000346 spin_unlock(&mp->m_sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
348 } else {
Eric Sandeen3685c2a2007-10-11 17:42:32 +1000349 spin_unlock(&mp->m_sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 }
351 }
352 /* Caller must log the inode */
353}
354
355/*
356 * Increment the link count on an inode & log the change.
357 */
358int
359xfs_bumplink(
360 xfs_trans_t *tp,
361 xfs_inode_t *ip)
362{
363 if (ip->i_d.di_nlink >= XFS_MAXLINK)
364 return XFS_ERROR(EMLINK);
365 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
366
367 ASSERT(ip->i_d.di_nlink > 0);
368 ip->i_d.di_nlink++;
Christoph Hellwig4188c782008-02-05 12:13:53 +1100369 inc_nlink(ip->i_vnode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 if ((ip->i_d.di_version == XFS_DINODE_VERSION_1) &&
371 (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
372 /*
373 * The inode has increased its number of links beyond
374 * what can fit in an old format inode. It now needs
375 * to be converted to a version 2 inode with a 32 bit
376 * link count. If this is the first inode in the file
377 * system to do this, then we need to bump the superblock
378 * version number as well.
379 */
380 xfs_bump_ino_vers2(tp, ip);
381 }
382
383 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
384 return 0;
385}
386
387/*
388 * Try to truncate the given file to 0 length. Currently called
389 * only out of xfs_remove when it has to truncate a file to free
390 * up space for the remove to proceed.
391 */
392int
393xfs_truncate_file(
394 xfs_mount_t *mp,
395 xfs_inode_t *ip)
396{
397 xfs_trans_t *tp;
398 int error;
399
400#ifdef QUOTADEBUG
401 /*
402 * This is called to truncate the quotainodes too.
403 */
404 if (XFS_IS_UQUOTA_ON(mp)) {
405 if (ip->i_ino != mp->m_sb.sb_uquotino)
406 ASSERT(ip->i_udquot);
407 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000408 if (XFS_IS_OQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 if (ip->i_ino != mp->m_sb.sb_gquotino)
410 ASSERT(ip->i_gdquot);
411 }
412#endif
413 /*
414 * Make the call to xfs_itruncate_start before starting the
415 * transaction, because we cannot make the call while we're
416 * in a transaction.
417 */
418 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000419 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, (xfs_fsize_t)0);
420 if (error) {
421 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
422 return error;
423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
425 tp = xfs_trans_alloc(mp, XFS_TRANS_TRUNCATE_FILE);
426 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
427 XFS_TRANS_PERM_LOG_RES,
428 XFS_ITRUNCATE_LOG_COUNT))) {
429 xfs_trans_cancel(tp, 0);
430 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
431 return error;
432 }
433
434 /*
435 * Follow the normal truncate locking protocol. Since we
436 * hold the inode in the transaction, we know that it's number
437 * of references will stay constant.
438 */
439 xfs_ilock(ip, XFS_ILOCK_EXCL);
440 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
441 xfs_trans_ihold(tp, ip);
442 /*
443 * Signal a sync xaction. The only case where that isn't
444 * the case is if we're truncating an already unlinked file
445 * on a wsync fs. In that case, we know the blocks can't
446 * reappear in the file because the links to file are
447 * permanently toast. Currently, we're always going to
448 * want a sync transaction because this code is being
449 * called from places where nlink is guaranteed to be 1
450 * but I'm leaving the tests in to protect against future
451 * changes -- rcc.
452 */
453 error = xfs_itruncate_finish(&tp, ip, (xfs_fsize_t)0,
454 XFS_DATA_FORK,
455 ((ip->i_d.di_nlink != 0 ||
456 !(mp->m_flags & XFS_MOUNT_WSYNC))
457 ? 1 : 0));
458 if (error) {
459 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
460 XFS_TRANS_ABORT);
461 } else {
462 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000463 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 }
465 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
466
467 return error;
468}