blob: 20ffec308e1e906cd09d661c3a9d409f97b5a8c3 [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(
68 bhv_desc_t *dir_bdp,
69 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{
Nathan Scott67fcaa72006-06-09 17:00:52 +100074 bhv_vnode_t *dir_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 xfs_inode_t *dp;
76 int error;
77
78 dir_vp = BHV_TO_VNODE(dir_bdp);
79 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
80
81 dp = XFS_BHVTOI(dir_bdp);
82
Nathan Scottf6c2d1f2006-06-20 13:04:51 +100083 error = xfs_dir_lookup(NULL, dp, VNAME(dentry), VNAMELEN(dentry), inum);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 if (!error) {
85 /*
86 * Unlock the directory. We do this because we can't
87 * hold the directory lock while doing the vn_get()
88 * in xfs_iget(). Doing so could cause us to hold
89 * a lock while waiting for the inode to finish
90 * being inactive while it's waiting for a log
91 * reservation in the inactive routine.
92 */
93 xfs_iunlock(dp, lock_mode);
94 error = xfs_iget(dp->i_mount, NULL, *inum, 0, 0, ipp, 0);
95 xfs_ilock(dp, lock_mode);
96
97 if (error) {
98 *ipp = NULL;
99 } else if ((*ipp)->i_d.di_mode == 0) {
100 /*
101 * The inode has been freed. Something is
102 * wrong so just get out of here.
103 */
104 xfs_iunlock(dp, lock_mode);
105 xfs_iput_new(*ipp, 0);
106 *ipp = NULL;
107 xfs_ilock(dp, lock_mode);
108 error = XFS_ERROR(ENOENT);
109 }
110 }
111 return error;
112}
113
114/*
115 * Allocates a new inode from disk and return a pointer to the
116 * incore copy. This routine will internally commit the current
117 * transaction and allocate a new one if the Space Manager needed
118 * to do an allocation to replenish the inode free-list.
119 *
120 * This routine is designed to be called from xfs_create and
121 * xfs_create_dir.
122 *
123 */
124int
125xfs_dir_ialloc(
126 xfs_trans_t **tpp, /* input: current transaction;
127 output: may be a new transaction. */
128 xfs_inode_t *dp, /* directory within whose allocate
129 the inode. */
130 mode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -0700131 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 xfs_dev_t rdev,
133 cred_t *credp,
134 prid_t prid, /* project id */
135 int okalloc, /* ok to allocate new space */
136 xfs_inode_t **ipp, /* pointer to inode; it will be
137 locked. */
138 int *committed)
139
140{
141 xfs_trans_t *tp;
142 xfs_trans_t *ntp;
143 xfs_inode_t *ip;
144 xfs_buf_t *ialloc_context = NULL;
145 boolean_t call_again = B_FALSE;
146 int code;
147 uint log_res;
148 uint log_count;
149 void *dqinfo;
150 uint tflags;
151
152 tp = *tpp;
153 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
154
155 /*
156 * xfs_ialloc will return a pointer to an incore inode if
157 * the Space Manager has an available inode on the free
158 * list. Otherwise, it will do an allocation and replenish
159 * the freelist. Since we can only do one allocation per
160 * transaction without deadlocks, we will need to commit the
161 * current transaction and start a new one. We will then
162 * need to call xfs_ialloc again to get the inode.
163 *
164 * If xfs_ialloc did an allocation to replenish the freelist,
165 * it returns the bp containing the head of the freelist as
166 * ialloc_context. We will hold a lock on it across the
167 * transaction commit so that no other process can steal
168 * the inode(s) that we've just allocated.
169 */
170 code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid, okalloc,
171 &ialloc_context, &call_again, &ip);
172
173 /*
174 * Return an error if we were unable to allocate a new inode.
175 * This should only happen if we run out of space on disk or
176 * encounter a disk error.
177 */
178 if (code) {
179 *ipp = NULL;
180 return code;
181 }
182 if (!call_again && (ip == NULL)) {
183 *ipp = NULL;
184 return XFS_ERROR(ENOSPC);
185 }
186
187 /*
188 * If call_again is set, then we were unable to get an
189 * inode in one operation. We need to commit the current
190 * transaction and call xfs_ialloc() again. It is guaranteed
191 * to succeed the second time.
192 */
193 if (call_again) {
194
195 /*
196 * Normally, xfs_trans_commit releases all the locks.
197 * We call bhold to hang on to the ialloc_context across
198 * the commit. Holding this buffer prevents any other
199 * processes from doing any allocations in this
200 * allocation group.
201 */
202 xfs_trans_bhold(tp, ialloc_context);
203 /*
204 * Save the log reservation so we can use
205 * them in the next transaction.
206 */
207 log_res = xfs_trans_get_log_res(tp);
208 log_count = xfs_trans_get_log_count(tp);
209
210 /*
211 * We want the quota changes to be associated with the next
212 * transaction, NOT this one. So, detach the dqinfo from this
213 * and attach it to the next transaction.
214 */
215 dqinfo = NULL;
216 tflags = 0;
217 if (tp->t_dqinfo) {
218 dqinfo = (void *)tp->t_dqinfo;
219 tp->t_dqinfo = NULL;
220 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
221 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
222 }
223
224 ntp = xfs_trans_dup(tp);
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000225 code = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 tp = ntp;
227 if (committed != NULL) {
228 *committed = 1;
229 }
230 /*
231 * If we get an error during the commit processing,
232 * release the buffer that is still held and return
233 * to the caller.
234 */
235 if (code) {
236 xfs_buf_relse(ialloc_context);
237 if (dqinfo) {
238 tp->t_dqinfo = dqinfo;
239 XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp);
240 }
241 *tpp = ntp;
242 *ipp = NULL;
243 return code;
244 }
245 code = xfs_trans_reserve(tp, 0, log_res, 0,
246 XFS_TRANS_PERM_LOG_RES, log_count);
247 /*
248 * Re-attach the quota info that we detached from prev trx.
249 */
250 if (dqinfo) {
251 tp->t_dqinfo = dqinfo;
252 tp->t_flags |= tflags;
253 }
254
255 if (code) {
256 xfs_buf_relse(ialloc_context);
257 *tpp = ntp;
258 *ipp = NULL;
259 return code;
260 }
261 xfs_trans_bjoin(tp, ialloc_context);
262
263 /*
264 * Call ialloc again. Since we've locked out all
265 * other allocations in this allocation group,
266 * this call should always succeed.
267 */
268 code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid,
269 okalloc, &ialloc_context, &call_again, &ip);
270
271 /*
272 * If we get an error at this point, return to the caller
273 * so that the current transaction can be aborted.
274 */
275 if (code) {
276 *tpp = tp;
277 *ipp = NULL;
278 return code;
279 }
280 ASSERT ((!call_again) && (ip != NULL));
281
282 } else {
283 if (committed != NULL) {
284 *committed = 0;
285 }
286 }
287
288 *ipp = ip;
289 *tpp = tp;
290
291 return 0;
292}
293
294/*
295 * Decrement the link count on an inode & log the change.
296 * If this causes the link count to go to zero, initiate the
297 * logging activity required to truncate a file.
298 */
299int /* error */
300xfs_droplink(
301 xfs_trans_t *tp,
302 xfs_inode_t *ip)
303{
304 int error;
305
306 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
307
308 ASSERT (ip->i_d.di_nlink > 0);
309 ip->i_d.di_nlink--;
310 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
311
312 error = 0;
313 if (ip->i_d.di_nlink == 0) {
314 /*
315 * We're dropping the last link to this file.
316 * Move the on-disk inode to the AGI unlinked list.
317 * From xfs_inactive() we will pull the inode from
318 * the list and free it.
319 */
320 error = xfs_iunlink(tp, ip);
321 }
322 return error;
323}
324
325/*
326 * This gets called when the inode's version needs to be changed from 1 to 2.
327 * Currently this happens when the nlink field overflows the old 16-bit value
328 * or when chproj is called to change the project for the first time.
329 * As a side effect the superblock version will also get rev'd
330 * to contain the NLINK bit.
331 */
332void
333xfs_bump_ino_vers2(
334 xfs_trans_t *tp,
335 xfs_inode_t *ip)
336{
337 xfs_mount_t *mp;
338 unsigned long s;
339
340 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
341 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1);
342
343 ip->i_d.di_version = XFS_DINODE_VERSION_2;
344 ip->i_d.di_onlink = 0;
345 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
346 mp = tp->t_mountp;
347 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
348 s = XFS_SB_LOCK(mp);
349 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
350 XFS_SB_VERSION_ADDNLINK(&mp->m_sb);
351 XFS_SB_UNLOCK(mp, s);
352 xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
353 } else {
354 XFS_SB_UNLOCK(mp, s);
355 }
356 }
357 /* Caller must log the inode */
358}
359
360/*
361 * Increment the link count on an inode & log the change.
362 */
363int
364xfs_bumplink(
365 xfs_trans_t *tp,
366 xfs_inode_t *ip)
367{
368 if (ip->i_d.di_nlink >= XFS_MAXLINK)
369 return XFS_ERROR(EMLINK);
370 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
371
372 ASSERT(ip->i_d.di_nlink > 0);
373 ip->i_d.di_nlink++;
374 if ((ip->i_d.di_version == XFS_DINODE_VERSION_1) &&
375 (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
376 /*
377 * The inode has increased its number of links beyond
378 * what can fit in an old format inode. It now needs
379 * to be converted to a version 2 inode with a 32 bit
380 * link count. If this is the first inode in the file
381 * system to do this, then we need to bump the superblock
382 * version number as well.
383 */
384 xfs_bump_ino_vers2(tp, ip);
385 }
386
387 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
388 return 0;
389}
390
391/*
392 * Try to truncate the given file to 0 length. Currently called
393 * only out of xfs_remove when it has to truncate a file to free
394 * up space for the remove to proceed.
395 */
396int
397xfs_truncate_file(
398 xfs_mount_t *mp,
399 xfs_inode_t *ip)
400{
401 xfs_trans_t *tp;
402 int error;
403
404#ifdef QUOTADEBUG
405 /*
406 * This is called to truncate the quotainodes too.
407 */
408 if (XFS_IS_UQUOTA_ON(mp)) {
409 if (ip->i_ino != mp->m_sb.sb_uquotino)
410 ASSERT(ip->i_udquot);
411 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000412 if (XFS_IS_OQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 if (ip->i_ino != mp->m_sb.sb_gquotino)
414 ASSERT(ip->i_gdquot);
415 }
416#endif
417 /*
418 * Make the call to xfs_itruncate_start before starting the
419 * transaction, because we cannot make the call while we're
420 * in a transaction.
421 */
422 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000423 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, (xfs_fsize_t)0);
424 if (error) {
425 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
426 return error;
427 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
429 tp = xfs_trans_alloc(mp, XFS_TRANS_TRUNCATE_FILE);
430 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
431 XFS_TRANS_PERM_LOG_RES,
432 XFS_ITRUNCATE_LOG_COUNT))) {
433 xfs_trans_cancel(tp, 0);
434 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
435 return error;
436 }
437
438 /*
439 * Follow the normal truncate locking protocol. Since we
440 * hold the inode in the transaction, we know that it's number
441 * of references will stay constant.
442 */
443 xfs_ilock(ip, XFS_ILOCK_EXCL);
444 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
445 xfs_trans_ihold(tp, ip);
446 /*
447 * Signal a sync xaction. The only case where that isn't
448 * the case is if we're truncating an already unlinked file
449 * on a wsync fs. In that case, we know the blocks can't
450 * reappear in the file because the links to file are
451 * permanently toast. Currently, we're always going to
452 * want a sync transaction because this code is being
453 * called from places where nlink is guaranteed to be 1
454 * but I'm leaving the tests in to protect against future
455 * changes -- rcc.
456 */
457 error = xfs_itruncate_finish(&tp, ip, (xfs_fsize_t)0,
458 XFS_DATA_FORK,
459 ((ip->i_d.di_nlink != 0 ||
460 !(mp->m_flags & XFS_MOUNT_WSYNC))
461 ? 1 : 0));
462 if (error) {
463 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
464 XFS_TRANS_ABORT);
465 } else {
466 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000467 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 }
469 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
470
471 return error;
472}