blob: 514eac913f1cc0c3b9759e488f80759a06a7a416 [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 */
Robert P. J. Day40ebd812007-11-23 16:30:51 +110018#include <linux/log2.h>
19
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_types.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_trans.h"
26#include "xfs_trans_priv.h"
27#include "xfs_sb.h"
28#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_inode_item.h"
38#include "xfs_btree.h"
39#include "xfs_alloc.h"
40#include "xfs_ialloc.h"
41#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "xfs_utils.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_quota.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100045#include "xfs_filestream.h"
Christoph Hellwig739bfb22007-08-29 10:58:01 +100046#include "xfs_vnodeops.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000047#include "xfs_trace.h"
Dave Chinner33479e02012-10-08 21:56:11 +110048#include "xfs_icache.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Linus Torvalds1da177e2005-04-16 15:20:36 -070050kmem_zone_t *xfs_ifork_zone;
51kmem_zone_t *xfs_inode_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +020054 * Used in xfs_itruncate_extents(). This is the maximum number of extents
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 * freed from a file in a single transaction.
56 */
57#define XFS_ITRUNC_MAX_EXTENTS 2
58
59STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
60STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
61STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
62STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
63
Dave Chinner2a0ec1d2012-04-23 15:59:02 +100064/*
65 * helper function to extract extent size hint from inode
66 */
67xfs_extlen_t
68xfs_get_extsz_hint(
69 struct xfs_inode *ip)
70{
71 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
72 return ip->i_d.di_extsize;
73 if (XFS_IS_REALTIME_INODE(ip))
74 return ip->i_mount->m_sb.sb_rextsize;
75 return 0;
76}
77
Dave Chinnerfa96aca2012-10-08 21:56:10 +110078/*
79 * This is a wrapper routine around the xfs_ilock() routine used to centralize
80 * some grungy code. It is used in places that wish to lock the inode solely
81 * for reading the extents. The reason these places can't just call
82 * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the
83 * extents from disk for a file in b-tree format. If the inode is in b-tree
84 * format, then we need to lock the inode exclusively until the extents are read
85 * in. Locking it exclusively all the time would limit our parallelism
86 * unnecessarily, though. What we do instead is check to see if the extents
87 * have been read in yet, and only lock the inode exclusively if they have not.
88 *
89 * The function returns a value which should be given to the corresponding
90 * xfs_iunlock_map_shared(). This value is the mode in which the lock was
91 * actually taken.
92 */
93uint
94xfs_ilock_map_shared(
95 xfs_inode_t *ip)
96{
97 uint lock_mode;
98
99 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
100 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
101 lock_mode = XFS_ILOCK_EXCL;
102 } else {
103 lock_mode = XFS_ILOCK_SHARED;
104 }
105
106 xfs_ilock(ip, lock_mode);
107
108 return lock_mode;
109}
110
111/*
112 * This is simply the unlock routine to go with xfs_ilock_map_shared().
113 * All it does is call xfs_iunlock() with the given lock_mode.
114 */
115void
116xfs_iunlock_map_shared(
117 xfs_inode_t *ip,
118 unsigned int lock_mode)
119{
120 xfs_iunlock(ip, lock_mode);
121}
122
123/*
124 * The xfs inode contains 2 locks: a multi-reader lock called the
125 * i_iolock and a multi-reader lock called the i_lock. This routine
126 * allows either or both of the locks to be obtained.
127 *
128 * The 2 locks should always be ordered so that the IO lock is
129 * obtained first in order to prevent deadlock.
130 *
131 * ip -- the inode being locked
132 * lock_flags -- this parameter indicates the inode's locks
133 * to be locked. It can be:
134 * XFS_IOLOCK_SHARED,
135 * XFS_IOLOCK_EXCL,
136 * XFS_ILOCK_SHARED,
137 * XFS_ILOCK_EXCL,
138 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
139 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
140 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
141 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
142 */
143void
144xfs_ilock(
145 xfs_inode_t *ip,
146 uint lock_flags)
147{
148 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
149
150 /*
151 * You can't set both SHARED and EXCL for the same lock,
152 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
153 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
154 */
155 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
156 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
157 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
158 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
159 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
160
161 if (lock_flags & XFS_IOLOCK_EXCL)
162 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
163 else if (lock_flags & XFS_IOLOCK_SHARED)
164 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
165
166 if (lock_flags & XFS_ILOCK_EXCL)
167 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
168 else if (lock_flags & XFS_ILOCK_SHARED)
169 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
170}
171
172/*
173 * This is just like xfs_ilock(), except that the caller
174 * is guaranteed not to sleep. It returns 1 if it gets
175 * the requested locks and 0 otherwise. If the IO lock is
176 * obtained but the inode lock cannot be, then the IO lock
177 * is dropped before returning.
178 *
179 * ip -- the inode being locked
180 * lock_flags -- this parameter indicates the inode's locks to be
181 * to be locked. See the comment for xfs_ilock() for a list
182 * of valid values.
183 */
184int
185xfs_ilock_nowait(
186 xfs_inode_t *ip,
187 uint lock_flags)
188{
189 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
190
191 /*
192 * You can't set both SHARED and EXCL for the same lock,
193 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
194 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
195 */
196 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
197 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
198 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
199 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
200 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
201
202 if (lock_flags & XFS_IOLOCK_EXCL) {
203 if (!mrtryupdate(&ip->i_iolock))
204 goto out;
205 } else if (lock_flags & XFS_IOLOCK_SHARED) {
206 if (!mrtryaccess(&ip->i_iolock))
207 goto out;
208 }
209 if (lock_flags & XFS_ILOCK_EXCL) {
210 if (!mrtryupdate(&ip->i_lock))
211 goto out_undo_iolock;
212 } else if (lock_flags & XFS_ILOCK_SHARED) {
213 if (!mrtryaccess(&ip->i_lock))
214 goto out_undo_iolock;
215 }
216 return 1;
217
218 out_undo_iolock:
219 if (lock_flags & XFS_IOLOCK_EXCL)
220 mrunlock_excl(&ip->i_iolock);
221 else if (lock_flags & XFS_IOLOCK_SHARED)
222 mrunlock_shared(&ip->i_iolock);
223 out:
224 return 0;
225}
226
227/*
228 * xfs_iunlock() is used to drop the inode locks acquired with
229 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
230 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
231 * that we know which locks to drop.
232 *
233 * ip -- the inode being unlocked
234 * lock_flags -- this parameter indicates the inode's locks to be
235 * to be unlocked. See the comment for xfs_ilock() for a list
236 * of valid values for this parameter.
237 *
238 */
239void
240xfs_iunlock(
241 xfs_inode_t *ip,
242 uint lock_flags)
243{
244 /*
245 * You can't set both SHARED and EXCL for the same lock,
246 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
247 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
248 */
249 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
250 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
251 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
252 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
253 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
254 ASSERT(lock_flags != 0);
255
256 if (lock_flags & XFS_IOLOCK_EXCL)
257 mrunlock_excl(&ip->i_iolock);
258 else if (lock_flags & XFS_IOLOCK_SHARED)
259 mrunlock_shared(&ip->i_iolock);
260
261 if (lock_flags & XFS_ILOCK_EXCL)
262 mrunlock_excl(&ip->i_lock);
263 else if (lock_flags & XFS_ILOCK_SHARED)
264 mrunlock_shared(&ip->i_lock);
265
266 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
267}
268
269/*
270 * give up write locks. the i/o lock cannot be held nested
271 * if it is being demoted.
272 */
273void
274xfs_ilock_demote(
275 xfs_inode_t *ip,
276 uint lock_flags)
277{
278 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
279 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
280
281 if (lock_flags & XFS_ILOCK_EXCL)
282 mrdemote(&ip->i_lock);
283 if (lock_flags & XFS_IOLOCK_EXCL)
284 mrdemote(&ip->i_iolock);
285
286 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
287}
288
289#ifdef DEBUG
290int
291xfs_isilocked(
292 xfs_inode_t *ip,
293 uint lock_flags)
294{
295 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
296 if (!(lock_flags & XFS_ILOCK_SHARED))
297 return !!ip->i_lock.mr_writer;
298 return rwsem_is_locked(&ip->i_lock.mr_lock);
299 }
300
301 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
302 if (!(lock_flags & XFS_IOLOCK_SHARED))
303 return !!ip->i_iolock.mr_writer;
304 return rwsem_is_locked(&ip->i_iolock.mr_lock);
305 }
306
307 ASSERT(0);
308 return 0;
309}
310#endif
311
312void
313__xfs_iflock(
314 struct xfs_inode *ip)
315{
316 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
317 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
318
319 do {
320 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
321 if (xfs_isiflocked(ip))
322 io_schedule();
323 } while (!xfs_iflock_nowait(ip));
324
325 finish_wait(wq, &wait.wait);
326}
327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328#ifdef DEBUG
329/*
330 * Make sure that the extents in the given memory buffer
331 * are valid.
332 */
333STATIC void
334xfs_validate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100335 xfs_ifork_t *ifp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 int nrecs,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 xfs_exntfmt_t fmt)
338{
339 xfs_bmbt_irec_t irec;
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000340 xfs_bmbt_rec_host_t rec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 int i;
342
343 for (i = 0; i < nrecs; i++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000344 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
345 rec.l0 = get_unaligned(&ep->l0);
346 rec.l1 = get_unaligned(&ep->l1);
347 xfs_bmbt_get_all(&rec, &irec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 if (fmt == XFS_EXTFMT_NOSTATE)
349 ASSERT(irec.br_state == XFS_EXT_NORM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 }
351}
352#else /* DEBUG */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000353#define xfs_validate_extents(ifp, nrecs, fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354#endif /* DEBUG */
355
356/*
357 * Check that none of the inode's in the buffer have a next
358 * unlinked field of 0.
359 */
360#if defined(DEBUG)
361void
362xfs_inobp_check(
363 xfs_mount_t *mp,
364 xfs_buf_t *bp)
365{
366 int i;
367 int j;
368 xfs_dinode_t *dip;
369
370 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
371
372 for (i = 0; i < j; i++) {
373 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
374 i * mp->m_sb.sb_inodesize);
375 if (!dip->di_next_unlinked) {
Dave Chinner53487782011-03-07 10:05:35 +1100376 xfs_alert(mp,
377 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 bp);
379 ASSERT(dip->di_next_unlinked);
380 }
381 }
382}
383#endif
384
Dave Chinneraf133e82012-11-12 22:54:07 +1100385static void
386xfs_inode_buf_verify(
387 struct xfs_buf *bp)
388{
389 struct xfs_mount *mp = bp->b_target->bt_mount;
390 int i;
391 int ni;
392
393 /*
394 * Validate the magic number and version of every inode in the buffer
395 */
396 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
397 for (i = 0; i < ni; i++) {
398 int di_ok;
399 xfs_dinode_t *dip;
400
401 dip = (struct xfs_dinode *)xfs_buf_offset(bp,
402 (i << mp->m_sb.sb_inodelog));
403 di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) &&
404 XFS_DINODE_GOOD_VERSION(dip->di_version);
405 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
406 XFS_ERRTAG_ITOBP_INOTOBP,
407 XFS_RANDOM_ITOBP_INOTOBP))) {
408 xfs_buf_ioerror(bp, EFSCORRUPTED);
409 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_HIGH,
410 mp, dip);
411#ifdef DEBUG
412 xfs_emerg(mp,
413 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
414 (unsigned long long)bp->b_bn, i,
415 be16_to_cpu(dip->di_magic));
416 ASSERT(0);
417#endif
418 }
419 }
420 xfs_inobp_check(mp, bp);
421 bp->b_iodone = NULL;
422 xfs_buf_ioend(bp, 0);
423}
424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425/*
Christoph Hellwig475ee412012-07-03 12:21:22 -0400426 * This routine is called to map an inode to the buffer containing the on-disk
427 * version of the inode. It returns a pointer to the buffer containing the
428 * on-disk inode in the bpp parameter, and in the dipp parameter it returns a
429 * pointer to the on-disk inode within that buffer.
430 *
431 * If a non-zero error is returned, then the contents of bpp and dipp are
432 * undefined.
David Chinner4ae29b42008-03-06 13:43:34 +1100433 */
Christoph Hellwig475ee412012-07-03 12:21:22 -0400434int
David Chinner4ae29b42008-03-06 13:43:34 +1100435xfs_imap_to_bp(
Christoph Hellwig475ee412012-07-03 12:21:22 -0400436 struct xfs_mount *mp,
437 struct xfs_trans *tp,
438 struct xfs_imap *imap,
Dave Chinneraf133e82012-11-12 22:54:07 +1100439 struct xfs_dinode **dipp,
Christoph Hellwig475ee412012-07-03 12:21:22 -0400440 struct xfs_buf **bpp,
441 uint buf_flags,
442 uint iget_flags)
David Chinner4ae29b42008-03-06 13:43:34 +1100443{
Christoph Hellwig475ee412012-07-03 12:21:22 -0400444 struct xfs_buf *bp;
445 int error;
David Chinner4ae29b42008-03-06 13:43:34 +1100446
Dave Chinner611c9942012-04-23 15:59:07 +1000447 buf_flags |= XBF_UNMAPPED;
David Chinner4ae29b42008-03-06 13:43:34 +1100448 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
Dave Chinneraf133e82012-11-12 22:54:07 +1100449 (int)imap->im_len, buf_flags, &bp,
450 xfs_inode_buf_verify);
David Chinner4ae29b42008-03-06 13:43:34 +1100451 if (error) {
Dave Chinneraf133e82012-11-12 22:54:07 +1100452 if (error == EAGAIN) {
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000453 ASSERT(buf_flags & XBF_TRYLOCK);
Dave Chinneraf133e82012-11-12 22:54:07 +1100454 return error;
David Chinnera3f74ff2008-03-06 13:43:42 +1100455 }
Dave Chinneraf133e82012-11-12 22:54:07 +1100456
457 if (error == EFSCORRUPTED &&
458 (iget_flags & XFS_IGET_UNTRUSTED))
459 return XFS_ERROR(EINVAL);
460
461 xfs_warn(mp, "%s: xfs_trans_read_buf() returned error %d.",
462 __func__, error);
David Chinner4ae29b42008-03-06 13:43:34 +1100463 return error;
464 }
465
David Chinner4ae29b42008-03-06 13:43:34 +1100466 *bpp = bp;
Christoph Hellwig475ee412012-07-03 12:21:22 -0400467 *dipp = (struct xfs_dinode *)xfs_buf_offset(bp, imap->im_boffset);
David Chinner4ae29b42008-03-06 13:43:34 +1100468 return 0;
469}
470
471/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 * Move inode type and inode format specific information from the
473 * on-disk inode to the in-core inode. For fifos, devs, and sockets
474 * this means set if_rdev to the proper value. For files, directories,
475 * and symlinks this means to bring in the in-line data or extent
476 * pointers. For a file in B-tree format, only the root is immediately
477 * brought in-core. The rest will be in-lined in if_extents when it
478 * is first referenced (see xfs_iread_extents()).
479 */
480STATIC int
481xfs_iformat(
482 xfs_inode_t *ip,
483 xfs_dinode_t *dip)
484{
485 xfs_attr_shortform_t *atp;
486 int size;
Christoph Hellwig8096b1e2011-12-18 20:00:07 +0000487 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 xfs_fsize_t di_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100490 if (unlikely(be32_to_cpu(dip->di_nextents) +
491 be16_to_cpu(dip->di_anextents) >
492 be64_to_cpu(dip->di_nblocks))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100493 xfs_warn(ip->i_mount,
Nathan Scott3762ec62006-01-12 10:29:53 +1100494 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 (unsigned long long)ip->i_ino,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100496 (int)(be32_to_cpu(dip->di_nextents) +
497 be16_to_cpu(dip->di_anextents)),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 (unsigned long long)
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100499 be64_to_cpu(dip->di_nblocks));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
501 ip->i_mount, dip);
502 return XFS_ERROR(EFSCORRUPTED);
503 }
504
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100505 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
Dave Chinner65333b42011-03-07 10:03:35 +1100506 xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 (unsigned long long)ip->i_ino,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100508 dip->di_forkoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
510 ip->i_mount, dip);
511 return XFS_ERROR(EFSCORRUPTED);
512 }
513
Christoph Hellwigb89d4202009-08-10 11:32:18 -0300514 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
515 !ip->i_mount->m_rtdev_targp)) {
Dave Chinner65333b42011-03-07 10:03:35 +1100516 xfs_warn(ip->i_mount,
Christoph Hellwigb89d4202009-08-10 11:32:18 -0300517 "corrupt dinode %Lu, has realtime flag set.",
518 ip->i_ino);
519 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
520 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
521 return XFS_ERROR(EFSCORRUPTED);
522 }
523
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 switch (ip->i_d.di_mode & S_IFMT) {
525 case S_IFIFO:
526 case S_IFCHR:
527 case S_IFBLK:
528 case S_IFSOCK:
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100529 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
531 ip->i_mount, dip);
532 return XFS_ERROR(EFSCORRUPTED);
533 }
534 ip->i_d.di_size = 0;
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100535 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 break;
537
538 case S_IFREG:
539 case S_IFLNK:
540 case S_IFDIR:
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100541 switch (dip->di_format) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 case XFS_DINODE_FMT_LOCAL:
543 /*
544 * no local regular files yet
545 */
Al Viroabbede12011-07-26 02:31:30 -0400546 if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100547 xfs_warn(ip->i_mount,
548 "corrupt inode %Lu (local format for regular file).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 (unsigned long long) ip->i_ino);
550 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
551 XFS_ERRLEVEL_LOW,
552 ip->i_mount, dip);
553 return XFS_ERROR(EFSCORRUPTED);
554 }
555
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100556 di_size = be64_to_cpu(dip->di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100558 xfs_warn(ip->i_mount,
559 "corrupt inode %Lu (bad size %Ld for local inode).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 (unsigned long long) ip->i_ino,
561 (long long) di_size);
562 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
563 XFS_ERRLEVEL_LOW,
564 ip->i_mount, dip);
565 return XFS_ERROR(EFSCORRUPTED);
566 }
567
568 size = (int)di_size;
569 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
570 break;
571 case XFS_DINODE_FMT_EXTENTS:
572 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
573 break;
574 case XFS_DINODE_FMT_BTREE:
575 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
576 break;
577 default:
578 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
579 ip->i_mount);
580 return XFS_ERROR(EFSCORRUPTED);
581 }
582 break;
583
584 default:
585 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
586 return XFS_ERROR(EFSCORRUPTED);
587 }
588 if (error) {
589 return error;
590 }
591 if (!XFS_DFORK_Q(dip))
592 return 0;
Christoph Hellwig8096b1e2011-12-18 20:00:07 +0000593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 ASSERT(ip->i_afp == NULL);
Dave Chinner4a7eddd2010-07-20 17:53:59 +1000595 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
Christoph Hellwig8096b1e2011-12-18 20:00:07 +0000596
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100597 switch (dip->di_aformat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 case XFS_DINODE_FMT_LOCAL:
599 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
Nathan Scott3b244aa2006-03-17 17:29:25 +1100600 size = be16_to_cpu(atp->hdr.totsize);
Christoph Hellwig2809f762009-01-19 02:04:16 +0100601
602 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100603 xfs_warn(ip->i_mount,
604 "corrupt inode %Lu (bad attr fork size %Ld).",
Christoph Hellwig2809f762009-01-19 02:04:16 +0100605 (unsigned long long) ip->i_ino,
606 (long long) size);
607 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
608 XFS_ERRLEVEL_LOW,
609 ip->i_mount, dip);
610 return XFS_ERROR(EFSCORRUPTED);
611 }
612
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
614 break;
615 case XFS_DINODE_FMT_EXTENTS:
616 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
617 break;
618 case XFS_DINODE_FMT_BTREE:
619 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
620 break;
621 default:
622 error = XFS_ERROR(EFSCORRUPTED);
623 break;
624 }
625 if (error) {
626 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
627 ip->i_afp = NULL;
628 xfs_idestroy_fork(ip, XFS_DATA_FORK);
629 }
630 return error;
631}
632
633/*
634 * The file is in-lined in the on-disk inode.
635 * If it fits into if_inline_data, then copy
636 * it there, otherwise allocate a buffer for it
637 * and copy the data there. Either way, set
638 * if_data to point at the data.
639 * If we allocate a buffer for the data, make
640 * sure that its size is a multiple of 4 and
641 * record the real size in i_real_bytes.
642 */
643STATIC int
644xfs_iformat_local(
645 xfs_inode_t *ip,
646 xfs_dinode_t *dip,
647 int whichfork,
648 int size)
649{
650 xfs_ifork_t *ifp;
651 int real_size;
652
653 /*
654 * If the size is unreasonable, then something
655 * is wrong and we just bail out rather than crash in
656 * kmem_alloc() or memcpy() below.
657 */
658 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100659 xfs_warn(ip->i_mount,
660 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 (unsigned long long) ip->i_ino, size,
662 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
663 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
664 ip->i_mount, dip);
665 return XFS_ERROR(EFSCORRUPTED);
666 }
667 ifp = XFS_IFORK_PTR(ip, whichfork);
668 real_size = 0;
669 if (size == 0)
670 ifp->if_u1.if_data = NULL;
671 else if (size <= sizeof(ifp->if_u2.if_inline_data))
672 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
673 else {
674 real_size = roundup(size, 4);
Dave Chinner4a7eddd2010-07-20 17:53:59 +1000675 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
677 ifp->if_bytes = size;
678 ifp->if_real_bytes = real_size;
679 if (size)
680 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
681 ifp->if_flags &= ~XFS_IFEXTENTS;
682 ifp->if_flags |= XFS_IFINLINE;
683 return 0;
684}
685
686/*
687 * The file consists of a set of extents all
688 * of which fit into the on-disk inode.
689 * If there are few enough extents to fit into
690 * the if_inline_ext, then copy them there.
691 * Otherwise allocate a buffer for them and copy
692 * them into it. Either way, set if_extents
693 * to point at the extents.
694 */
695STATIC int
696xfs_iformat_extents(
697 xfs_inode_t *ip,
698 xfs_dinode_t *dip,
699 int whichfork)
700{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000701 xfs_bmbt_rec_t *dp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 xfs_ifork_t *ifp;
703 int nex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 int size;
705 int i;
706
707 ifp = XFS_IFORK_PTR(ip, whichfork);
708 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
709 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
710
711 /*
712 * If the number of extents is unreasonable, then something
713 * is wrong and we just bail out rather than crash in
714 * kmem_alloc() or memcpy() below.
715 */
716 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100717 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 (unsigned long long) ip->i_ino, nex);
719 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
720 ip->i_mount, dip);
721 return XFS_ERROR(EFSCORRUPTED);
722 }
723
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100724 ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 if (nex == 0)
726 ifp->if_u1.if_extents = NULL;
727 else if (nex <= XFS_INLINE_EXTS)
728 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100729 else
730 xfs_iext_add(ifp, 0, nex);
731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 ifp->if_bytes = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 if (size) {
734 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000735 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100736 for (i = 0; i < nex; i++, dp++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000737 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Harvey Harrison597bca62008-08-13 16:29:21 +1000738 ep->l0 = get_unaligned_be64(&dp->l0);
739 ep->l1 = get_unaligned_be64(&dp->l1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 }
Eric Sandeen3a59c942007-07-11 11:09:47 +1000741 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 if (whichfork != XFS_DATA_FORK ||
743 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
744 if (unlikely(xfs_check_nostate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100745 ifp, 0, nex))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
747 XFS_ERRLEVEL_LOW,
748 ip->i_mount);
749 return XFS_ERROR(EFSCORRUPTED);
750 }
751 }
752 ifp->if_flags |= XFS_IFEXTENTS;
753 return 0;
754}
755
756/*
757 * The file has too many extents to fit into
758 * the inode, so they are in B-tree format.
759 * Allocate a buffer for the root of the B-tree
760 * and copy the root into it. The i_extents
761 * field will remain NULL until all of the
762 * extents are read in (when they are needed).
763 */
764STATIC int
765xfs_iformat_btree(
766 xfs_inode_t *ip,
767 xfs_dinode_t *dip,
768 int whichfork)
769{
770 xfs_bmdr_block_t *dfp;
771 xfs_ifork_t *ifp;
772 /* REFERENCED */
773 int nrecs;
774 int size;
775
776 ifp = XFS_IFORK_PTR(ip, whichfork);
777 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
778 size = XFS_BMAP_BROOT_SPACE(dfp);
Christoph Hellwig60197e82008-10-30 17:11:19 +1100779 nrecs = be16_to_cpu(dfp->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
781 /*
782 * blow out if -- fork has less extents than can fit in
783 * fork (fork shouldn't be a btree format), root btree
784 * block has more records than can fit into the fork,
785 * or the number of extents is greater than the number of
786 * blocks.
787 */
Christoph Hellwig8096b1e2011-12-18 20:00:07 +0000788 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <=
789 XFS_IFORK_MAXEXT(ip, whichfork) ||
790 XFS_BMDR_SPACE_CALC(nrecs) >
791 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) ||
792 XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
Dave Chinner65333b42011-03-07 10:03:35 +1100793 xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 (unsigned long long) ip->i_ino);
Dave Chinner65333b42011-03-07 10:03:35 +1100795 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
796 ip->i_mount, dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 return XFS_ERROR(EFSCORRUPTED);
798 }
799
800 ifp->if_broot_bytes = size;
Dave Chinner4a7eddd2010-07-20 17:53:59 +1000801 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 ASSERT(ifp->if_broot != NULL);
803 /*
804 * Copy and convert from the on-disk structure
805 * to the in-memory structure.
806 */
Christoph Hellwig60197e82008-10-30 17:11:19 +1100807 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
808 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
809 ifp->if_broot, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 ifp->if_flags &= ~XFS_IFEXTENTS;
811 ifp->if_flags |= XFS_IFBROOT;
812
813 return 0;
814}
815
Eric Sandeend96f8f82009-07-02 00:09:33 -0500816STATIC void
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000817xfs_dinode_from_disk(
818 xfs_icdinode_t *to,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100819 xfs_dinode_t *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000821 to->di_magic = be16_to_cpu(from->di_magic);
822 to->di_mode = be16_to_cpu(from->di_mode);
823 to->di_version = from ->di_version;
824 to->di_format = from->di_format;
825 to->di_onlink = be16_to_cpu(from->di_onlink);
826 to->di_uid = be32_to_cpu(from->di_uid);
827 to->di_gid = be32_to_cpu(from->di_gid);
828 to->di_nlink = be32_to_cpu(from->di_nlink);
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000829 to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
830 to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000831 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
832 to->di_flushiter = be16_to_cpu(from->di_flushiter);
833 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
834 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
835 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
836 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
837 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
838 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
839 to->di_size = be64_to_cpu(from->di_size);
840 to->di_nblocks = be64_to_cpu(from->di_nblocks);
841 to->di_extsize = be32_to_cpu(from->di_extsize);
842 to->di_nextents = be32_to_cpu(from->di_nextents);
843 to->di_anextents = be16_to_cpu(from->di_anextents);
844 to->di_forkoff = from->di_forkoff;
845 to->di_aformat = from->di_aformat;
846 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
847 to->di_dmstate = be16_to_cpu(from->di_dmstate);
848 to->di_flags = be16_to_cpu(from->di_flags);
849 to->di_gen = be32_to_cpu(from->di_gen);
850}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000852void
853xfs_dinode_to_disk(
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100854 xfs_dinode_t *to,
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000855 xfs_icdinode_t *from)
856{
857 to->di_magic = cpu_to_be16(from->di_magic);
858 to->di_mode = cpu_to_be16(from->di_mode);
859 to->di_version = from ->di_version;
860 to->di_format = from->di_format;
861 to->di_onlink = cpu_to_be16(from->di_onlink);
862 to->di_uid = cpu_to_be32(from->di_uid);
863 to->di_gid = cpu_to_be32(from->di_gid);
864 to->di_nlink = cpu_to_be32(from->di_nlink);
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000865 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
866 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000867 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
868 to->di_flushiter = cpu_to_be16(from->di_flushiter);
869 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
870 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
871 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
872 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
873 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
874 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
875 to->di_size = cpu_to_be64(from->di_size);
876 to->di_nblocks = cpu_to_be64(from->di_nblocks);
877 to->di_extsize = cpu_to_be32(from->di_extsize);
878 to->di_nextents = cpu_to_be32(from->di_nextents);
879 to->di_anextents = cpu_to_be16(from->di_anextents);
880 to->di_forkoff = from->di_forkoff;
881 to->di_aformat = from->di_aformat;
882 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
883 to->di_dmstate = cpu_to_be16(from->di_dmstate);
884 to->di_flags = cpu_to_be16(from->di_flags);
885 to->di_gen = cpu_to_be32(from->di_gen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
887
888STATIC uint
889_xfs_dic2xflags(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 __uint16_t di_flags)
891{
892 uint flags = 0;
893
894 if (di_flags & XFS_DIFLAG_ANY) {
895 if (di_flags & XFS_DIFLAG_REALTIME)
896 flags |= XFS_XFLAG_REALTIME;
897 if (di_flags & XFS_DIFLAG_PREALLOC)
898 flags |= XFS_XFLAG_PREALLOC;
899 if (di_flags & XFS_DIFLAG_IMMUTABLE)
900 flags |= XFS_XFLAG_IMMUTABLE;
901 if (di_flags & XFS_DIFLAG_APPEND)
902 flags |= XFS_XFLAG_APPEND;
903 if (di_flags & XFS_DIFLAG_SYNC)
904 flags |= XFS_XFLAG_SYNC;
905 if (di_flags & XFS_DIFLAG_NOATIME)
906 flags |= XFS_XFLAG_NOATIME;
907 if (di_flags & XFS_DIFLAG_NODUMP)
908 flags |= XFS_XFLAG_NODUMP;
909 if (di_flags & XFS_DIFLAG_RTINHERIT)
910 flags |= XFS_XFLAG_RTINHERIT;
911 if (di_flags & XFS_DIFLAG_PROJINHERIT)
912 flags |= XFS_XFLAG_PROJINHERIT;
913 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
914 flags |= XFS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100915 if (di_flags & XFS_DIFLAG_EXTSIZE)
916 flags |= XFS_XFLAG_EXTSIZE;
917 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
918 flags |= XFS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000919 if (di_flags & XFS_DIFLAG_NODEFRAG)
920 flags |= XFS_XFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000921 if (di_flags & XFS_DIFLAG_FILESTREAM)
922 flags |= XFS_XFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 }
924
925 return flags;
926}
927
928uint
929xfs_ip2xflags(
930 xfs_inode_t *ip)
931{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000932 xfs_icdinode_t *dic = &ip->i_d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Nathan Scotta916e2b2006-06-09 17:12:17 +1000934 return _xfs_dic2xflags(dic->di_flags) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100935 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936}
937
938uint
939xfs_dic2xflags(
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100940 xfs_dinode_t *dip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941{
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100942 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100943 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944}
945
946/*
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100947 * Read the disk inode attributes into the in-core inode structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 */
949int
950xfs_iread(
951 xfs_mount_t *mp,
952 xfs_trans_t *tp,
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100953 xfs_inode_t *ip,
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100954 uint iget_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955{
956 xfs_buf_t *bp;
957 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 int error;
959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 /*
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100961 * Fill in the location information in the in-core inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 */
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100963 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100964 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100965 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
967 /*
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100968 * Get pointers to the on-disk inode and the buffer containing it.
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100969 */
Christoph Hellwig475ee412012-07-03 12:21:22 -0400970 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0, iget_flags);
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100971 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100972 return error;
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100973
974 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 * If we got something that isn't an inode it means someone
976 * (nfs or dmi) has a stale handle.
977 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200978 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979#ifdef DEBUG
Dave Chinner53487782011-03-07 10:05:35 +1100980 xfs_alert(mp,
981 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
982 __func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983#endif /* DEBUG */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100984 error = XFS_ERROR(EINVAL);
985 goto out_brelse;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 }
987
988 /*
989 * If the on-disk inode is already linked to a directory
990 * entry, copy all of the inode into the in-core inode.
991 * xfs_iformat() handles copying in the inode format
992 * specific information.
993 * Otherwise, just get the truly permanent information.
994 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100995 if (dip->di_mode) {
996 xfs_dinode_from_disk(&ip->i_d, dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 error = xfs_iformat(ip, dip);
998 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999#ifdef DEBUG
Dave Chinner53487782011-03-07 10:05:35 +11001000 xfs_alert(mp, "%s: xfs_iformat() returned error %d",
1001 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002#endif /* DEBUG */
Christoph Hellwig9ed04512008-10-30 18:26:04 +11001003 goto out_brelse;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005 } else {
Christoph Hellwig81591fe2008-11-28 14:23:39 +11001006 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
1007 ip->i_d.di_version = dip->di_version;
1008 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
1009 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 /*
1011 * Make sure to pull in the mode here as well in
1012 * case the inode is released without being used.
1013 * This ensures that xfs_inactive() will see that
1014 * the inode is already free and not try to mess
1015 * with the uninitialized part of it.
1016 */
1017 ip->i_d.di_mode = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
1019
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 /*
1021 * The inode format changed when we moved the link count and
1022 * made it 32 bits long. If this is an old format inode,
1023 * convert it in memory to look like a new one. If it gets
1024 * flushed to disk we will convert back before flushing or
1025 * logging it. We zero out the new projid field and the old link
1026 * count field. We'll handle clearing the pad field (the remains
1027 * of the old uuid field) when we actually convert the inode to
1028 * the new format. We don't change the version number so that we
1029 * can distinguish this from a real new format inode.
1030 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11001031 if (ip->i_d.di_version == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 ip->i_d.di_nlink = ip->i_d.di_onlink;
1033 ip->i_d.di_onlink = 0;
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001034 xfs_set_projid(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 }
1036
1037 ip->i_delayed_blks = 0;
1038
1039 /*
1040 * Mark the buffer containing the inode as something to keep
1041 * around for a while. This helps to keep recently accessed
1042 * meta-data in-core longer.
1043 */
Dave Chinner821eb212010-12-02 16:31:13 +11001044 xfs_buf_set_ref(bp, XFS_INO_REF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
1046 /*
1047 * Use xfs_trans_brelse() to release the buffer containing the
1048 * on-disk inode, because it was acquired with xfs_trans_read_buf()
Christoph Hellwig475ee412012-07-03 12:21:22 -04001049 * in xfs_imap_to_bp() above. If tp is NULL, this is just a normal
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1051 * will only release the buffer if it is not dirty within the
1052 * transaction. It will be OK to release the buffer in this case,
1053 * because inodes on disk are never destroyed and we will be
1054 * locking the new in-core inode before putting it in the hash
1055 * table where other processes can find it. Thus we don't have
1056 * to worry about the inode being changed just because we released
1057 * the buffer.
1058 */
Christoph Hellwig9ed04512008-10-30 18:26:04 +11001059 out_brelse:
1060 xfs_trans_brelse(tp, bp);
Christoph Hellwig9ed04512008-10-30 18:26:04 +11001061 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062}
1063
1064/*
1065 * Read in extents from a btree-format inode.
1066 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1067 */
1068int
1069xfs_iread_extents(
1070 xfs_trans_t *tp,
1071 xfs_inode_t *ip,
1072 int whichfork)
1073{
1074 int error;
1075 xfs_ifork_t *ifp;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001076 xfs_extnum_t nextents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1079 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1080 ip->i_mount);
1081 return XFS_ERROR(EFSCORRUPTED);
1082 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001083 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 ifp = XFS_IFORK_PTR(ip, whichfork);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 /*
1087 * We know that the size is valid (it's checked in iformat_btree)
1088 */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001089 ifp->if_bytes = ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 ifp->if_flags |= XFS_IFEXTENTS;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001091 xfs_iext_add(ifp, 0, nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 error = xfs_bmap_read_extents(tp, ip, whichfork);
1093 if (error) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001094 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 ifp->if_flags &= ~XFS_IFEXTENTS;
1096 return error;
1097 }
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10001098 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 return 0;
1100}
1101
1102/*
1103 * Allocate an inode on disk and return a copy of its in-core version.
1104 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1105 * appropriately within the inode. The uid and gid for the inode are
1106 * set according to the contents of the given cred structure.
1107 *
1108 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
Carlos Maiolinocd856db2012-10-20 11:08:19 -03001109 * has a free inode available, call xfs_iget() to obtain the in-core
1110 * version of the allocated inode. Finally, fill in the inode and
1111 * log its initial contents. In this case, ialloc_context would be
1112 * set to NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 *
Carlos Maiolinocd856db2012-10-20 11:08:19 -03001114 * If xfs_dialloc() does not have an available inode, it will replenish
1115 * its supply by doing an allocation. Since we can only do one
1116 * allocation within a transaction without deadlocks, we must commit
1117 * the current transaction before returning the inode itself.
1118 * In this case, therefore, we will set ialloc_context and return.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 * The caller should then commit the current transaction, start a new
1120 * transaction, and call xfs_ialloc() again to actually get the inode.
1121 *
1122 * To ensure that some other process does not grab the inode that
1123 * was allocated during the first call to xfs_ialloc(), this routine
1124 * also returns the [locked] bp pointing to the head of the freelist
1125 * as ialloc_context. The caller should hold this buffer across
1126 * the commit and pass it back into this routine on the second call.
David Chinnerb11f94d2007-07-11 11:09:33 +10001127 *
1128 * If we are allocating quota inodes, we do not have a parent inode
1129 * to attach to or associate with (i.e. pip == NULL) because they
1130 * are not linked into the directory structure - they are attached
1131 * directly to the superblock - and so have no parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 */
1133int
1134xfs_ialloc(
1135 xfs_trans_t *tp,
1136 xfs_inode_t *pip,
Al Viro576b1d62011-07-26 02:50:15 -04001137 umode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -07001138 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 xfs_dev_t rdev,
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001140 prid_t prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 int okalloc,
1142 xfs_buf_t **ialloc_context,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 xfs_inode_t **ipp)
1144{
1145 xfs_ino_t ino;
1146 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 uint flags;
1148 int error;
Christoph Hellwigdff35fd2008-08-13 16:44:15 +10001149 timespec_t tv;
David Chinnerbf904242008-10-30 17:36:14 +11001150 int filestreams = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
1152 /*
1153 * Call the space management code to pick
1154 * the on-disk inode to be allocated.
1155 */
David Chinnerb11f94d2007-07-11 11:09:33 +10001156 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
Christoph Hellwig08358902012-07-04 10:54:47 -04001157 ialloc_context, &ino);
David Chinnerbf904242008-10-30 17:36:14 +11001158 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 return error;
Christoph Hellwig08358902012-07-04 10:54:47 -04001160 if (*ialloc_context || ino == NULLFSINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 *ipp = NULL;
1162 return 0;
1163 }
1164 ASSERT(*ialloc_context == NULL);
1165
1166 /*
1167 * Get the in-core inode with the lock held exclusively.
1168 * This is because we're setting fields here we need
1169 * to prevent others from looking at until we're done.
1170 */
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001171 error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE,
1172 XFS_ILOCK_EXCL, &ip);
David Chinnerbf904242008-10-30 17:36:14 +11001173 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 ASSERT(ip != NULL);
1176
Al Viro576b1d62011-07-26 02:50:15 -04001177 ip->i_d.di_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 ip->i_d.di_onlink = 0;
1179 ip->i_d.di_nlink = nlink;
1180 ASSERT(ip->i_d.di_nlink == nlink);
David Howells9e2b2dc2008-08-13 16:20:04 +01001181 ip->i_d.di_uid = current_fsuid();
1182 ip->i_d.di_gid = current_fsgid();
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001183 xfs_set_projid(ip, prid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1185
1186 /*
1187 * If the superblock version is up to where we support new format
1188 * inodes and this is currently an old format inode, then change
1189 * the inode version number now. This way we only do the conversion
1190 * here rather than here and in the flush/logging code.
1191 */
Eric Sandeen62118702008-03-06 13:44:28 +11001192 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11001193 ip->i_d.di_version == 1) {
1194 ip->i_d.di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 /*
1196 * We've already zeroed the old link count, the projid field,
1197 * and the pad field.
1198 */
1199 }
1200
1201 /*
1202 * Project ids won't be stored on disk if we are using a version 1 inode.
1203 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11001204 if ((prid != 0) && (ip->i_d.di_version == 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 xfs_bump_ino_vers2(tp, ip);
1206
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001207 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 ip->i_d.di_gid = pip->i_d.di_gid;
Al Viroabbede12011-07-26 02:31:30 -04001209 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 ip->i_d.di_mode |= S_ISGID;
1211 }
1212 }
1213
1214 /*
1215 * If the group ID of the new file does not match the effective group
1216 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1217 * (and only if the irix_sgid_inherit compatibility variable is set).
1218 */
1219 if ((irix_sgid_inherit) &&
1220 (ip->i_d.di_mode & S_ISGID) &&
1221 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1222 ip->i_d.di_mode &= ~S_ISGID;
1223 }
1224
1225 ip->i_d.di_size = 0;
1226 ip->i_d.di_nextents = 0;
1227 ASSERT(ip->i_d.di_nblocks == 0);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +10001228
1229 nanotime(&tv);
1230 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1231 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1232 ip->i_d.di_atime = ip->i_d.di_mtime;
1233 ip->i_d.di_ctime = ip->i_d.di_mtime;
1234
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 /*
1236 * di_gen will have been taken care of in xfs_iread.
1237 */
1238 ip->i_d.di_extsize = 0;
1239 ip->i_d.di_dmevmask = 0;
1240 ip->i_d.di_dmstate = 0;
1241 ip->i_d.di_flags = 0;
1242 flags = XFS_ILOG_CORE;
1243 switch (mode & S_IFMT) {
1244 case S_IFIFO:
1245 case S_IFCHR:
1246 case S_IFBLK:
1247 case S_IFSOCK:
1248 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1249 ip->i_df.if_u2.if_rdev = rdev;
1250 ip->i_df.if_flags = 0;
1251 flags |= XFS_ILOG_DEV;
1252 break;
1253 case S_IFREG:
David Chinnerbf904242008-10-30 17:36:14 +11001254 /*
1255 * we can't set up filestreams until after the VFS inode
1256 * is set up properly.
1257 */
1258 if (pip && xfs_inode_is_filestream(pip))
1259 filestreams = 1;
David Chinner2a82b8b2007-07-11 11:09:12 +10001260 /* fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +10001262 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Nathan Scott365ca832005-06-21 15:39:12 +10001263 uint di_flags = 0;
1264
Al Viroabbede12011-07-26 02:31:30 -04001265 if (S_ISDIR(mode)) {
Nathan Scott365ca832005-06-21 15:39:12 +10001266 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1267 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001268 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1269 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1270 ip->i_d.di_extsize = pip->i_d.di_extsize;
1271 }
Al Viroabbede12011-07-26 02:31:30 -04001272 } else if (S_ISREG(mode)) {
Christoph Hellwig613d7042007-10-11 17:44:08 +10001273 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +10001274 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001275 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1276 di_flags |= XFS_DIFLAG_EXTSIZE;
1277 ip->i_d.di_extsize = pip->i_d.di_extsize;
1278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 }
1280 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1281 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +10001282 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1284 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +10001285 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1287 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +10001288 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1290 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +10001291 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1292 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1293 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +10001294 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1295 xfs_inherit_nodefrag)
1296 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +10001297 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1298 di_flags |= XFS_DIFLAG_FILESTREAM;
Nathan Scott365ca832005-06-21 15:39:12 +10001299 ip->i_d.di_flags |= di_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 }
1301 /* FALLTHROUGH */
1302 case S_IFLNK:
1303 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1304 ip->i_df.if_flags = XFS_IFEXTENTS;
1305 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1306 ip->i_df.if_u1.if_extents = NULL;
1307 break;
1308 default:
1309 ASSERT(0);
1310 }
1311 /*
1312 * Attribute fork settings for new inode.
1313 */
1314 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1315 ip->i_d.di_anextents = 0;
1316
1317 /*
1318 * Log the new values stuffed into the inode.
1319 */
Christoph Hellwigddc34152011-09-19 15:00:54 +00001320 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 xfs_trans_log_inode(tp, ip, flags);
1322
Nathan Scottb83bd132006-06-09 16:48:30 +10001323 /* now that we have an i_mode we can setup inode ops and unlock */
Christoph Hellwig41be8be2008-08-13 16:23:13 +10001324 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
David Chinnerbf904242008-10-30 17:36:14 +11001326 /* now we have set up the vfs inode we can associate the filestream */
1327 if (filestreams) {
1328 error = xfs_filestream_associate(pip, ip);
1329 if (error < 0)
1330 return -error;
1331 if (!error)
1332 xfs_iflags_set(ip, XFS_IFILESTREAM);
1333 }
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 *ipp = ip;
1336 return 0;
1337}
1338
1339/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001340 * Free up the underlying blocks past new_size. The new size must be smaller
1341 * than the current size. This routine can be used both for the attribute and
1342 * data fork, and does not modify the inode size, which is left to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 *
David Chinnerf6485052008-04-17 16:50:04 +10001344 * The transaction passed to this routine must have made a permanent log
1345 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1346 * given transaction and start new ones, so make sure everything involved in
1347 * the transaction is tidy before calling here. Some transaction will be
1348 * returned to the caller to be committed. The incoming transaction must
1349 * already include the inode, and both inode locks must be held exclusively.
1350 * The inode must also be "held" within the transaction. On return the inode
1351 * will be "held" within the returned transaction. This routine does NOT
1352 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 *
David Chinnerf6485052008-04-17 16:50:04 +10001354 * If we get an error, we must return with the inode locked and linked into the
1355 * current transaction. This keeps things simple for the higher level code,
1356 * because it always knows that the inode is locked and held in the transaction
1357 * that returns to it whether errors occur or not. We don't mark the inode
1358 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 */
1360int
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001361xfs_itruncate_extents(
1362 struct xfs_trans **tpp,
1363 struct xfs_inode *ip,
1364 int whichfork,
1365 xfs_fsize_t new_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001367 struct xfs_mount *mp = ip->i_mount;
1368 struct xfs_trans *tp = *tpp;
1369 struct xfs_trans *ntp;
1370 xfs_bmap_free_t free_list;
1371 xfs_fsblock_t first_block;
1372 xfs_fileoff_t first_unmap_block;
1373 xfs_fileoff_t last_block;
1374 xfs_filblks_t unmap_len;
1375 int committed;
1376 int error = 0;
1377 int done = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
Christoph Hellwig0b561852012-07-04 11:13:31 -04001379 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1380 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1381 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00001382 ASSERT(new_size <= XFS_ISIZE(ip));
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001383 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 ASSERT(ip->i_itemp != NULL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001385 ASSERT(ip->i_itemp->ili_lock_flags == 0);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001386 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001388 trace_xfs_itruncate_extents_start(ip, new_size);
1389
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 /*
1391 * Since it is possible for space to become allocated beyond
1392 * the end of the file (in a crash where the space is allocated
1393 * but the inode size is not yet updated), simply remove any
1394 * blocks which show up between the new EOF and the maximum
1395 * possible file size. If the first block to be removed is
1396 * beyond the maximum file size (ie it is the same as last_block),
1397 * then there is nothing to do.
1398 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001399 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
Dave Chinner32972382012-06-08 15:44:54 +10001400 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001401 if (first_unmap_block == last_block)
1402 return 0;
1403
1404 ASSERT(first_unmap_block < last_block);
1405 unmap_len = last_block - first_unmap_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 while (!done) {
Eric Sandeen9d87c312009-01-14 23:22:07 -06001407 xfs_bmap_init(&free_list, &first_block);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001408 error = xfs_bunmapi(tp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001409 first_unmap_block, unmap_len,
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001410 xfs_bmapi_aflag(whichfork),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001412 &first_block, &free_list,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001413 &done);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001414 if (error)
1415 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
1417 /*
1418 * Duplicate the transaction that has the permanent
1419 * reservation and commit the old transaction.
1420 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001421 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001422 if (committed)
Christoph Hellwigddc34152011-09-19 15:00:54 +00001423 xfs_trans_ijoin(tp, ip, 0);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001424 if (error)
1425 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
1427 if (committed) {
1428 /*
David Chinnerf6485052008-04-17 16:50:04 +10001429 * Mark the inode dirty so it will be logged and
David Chinnere5720ee2008-04-10 12:21:18 +10001430 * moved forward in the log as part of every commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001432 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 }
David Chinnerf6485052008-04-17 16:50:04 +10001434
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001435 ntp = xfs_trans_dup(tp);
1436 error = xfs_trans_commit(tp, 0);
1437 tp = ntp;
David Chinnere5720ee2008-04-10 12:21:18 +10001438
Christoph Hellwigddc34152011-09-19 15:00:54 +00001439 xfs_trans_ijoin(tp, ip, 0);
David Chinnerf6485052008-04-17 16:50:04 +10001440
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001441 if (error)
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001442 goto out;
1443
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001444 /*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001445 * Transaction commit worked ok so we can drop the extra ticket
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001446 * reference that we gained in xfs_trans_dup()
1447 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001448 xfs_log_ticket_put(tp->t_ticket);
1449 error = xfs_trans_reserve(tp, 0,
David Chinnerf6485052008-04-17 16:50:04 +10001450 XFS_ITRUNCATE_LOG_RES(mp), 0,
1451 XFS_TRANS_PERM_LOG_RES,
1452 XFS_ITRUNCATE_LOG_COUNT);
1453 if (error)
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001454 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 }
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001456
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001457 /*
1458 * Always re-log the inode so that our permanent transaction can keep
1459 * on rolling it forward in the log.
1460 */
1461 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1462
1463 trace_xfs_itruncate_extents_end(ip, new_size);
1464
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001465out:
1466 *tpp = tp;
1467 return error;
1468out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 /*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001470 * If the bunmapi call encounters an error, return to the caller where
1471 * the transaction can be properly aborted. We just need to make sure
1472 * we're not holding any resources that we were not when we came in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001474 xfs_bmap_cancel(&free_list);
1475 goto out;
1476}
1477
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478/*
1479 * This is called when the inode's link count goes to 0.
1480 * We place the on-disk inode on a list in the AGI. It
1481 * will be pulled from this list when the inode is freed.
1482 */
1483int
1484xfs_iunlink(
1485 xfs_trans_t *tp,
1486 xfs_inode_t *ip)
1487{
1488 xfs_mount_t *mp;
1489 xfs_agi_t *agi;
1490 xfs_dinode_t *dip;
1491 xfs_buf_t *agibp;
1492 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 xfs_agino_t agino;
1494 short bucket_index;
1495 int offset;
1496 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498 ASSERT(ip->i_d.di_nlink == 0);
1499 ASSERT(ip->i_d.di_mode != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
1501 mp = tp->t_mountp;
1502
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 /*
1504 * Get the agi buffer first. It ensures lock ordering
1505 * on the list.
1506 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001507 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001508 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 /*
1513 * Get the index into the agi hash table for the
1514 * list this inode will go on.
1515 */
1516 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1517 ASSERT(agino != 0);
1518 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1519 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001520 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001522 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 /*
1524 * There is already another inode in the bucket we need
1525 * to add ourselves to. Add us at the front of the list.
1526 * Here we put the head pointer into our next pointer,
1527 * and then we fall through to point the head at us.
1528 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04001529 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1530 0, 0);
Vlad Apostolovc319b582007-11-23 16:27:51 +11001531 if (error)
1532 return error;
1533
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001534 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001536 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 offsetof(xfs_dinode_t, di_next_unlinked);
1538 xfs_trans_inode_buf(tp, ibp);
1539 xfs_trans_log_buf(tp, ibp, offset,
1540 (offset + sizeof(xfs_agino_t) - 1));
1541 xfs_inobp_check(mp, ibp);
1542 }
1543
1544 /*
1545 * Point the bucket head pointer at the inode being inserted.
1546 */
1547 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001548 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 offset = offsetof(xfs_agi_t, agi_unlinked) +
1550 (sizeof(xfs_agino_t) * bucket_index);
1551 xfs_trans_log_buf(tp, agibp, offset,
1552 (offset + sizeof(xfs_agino_t) - 1));
1553 return 0;
1554}
1555
1556/*
1557 * Pull the on-disk inode from the AGI unlinked list.
1558 */
1559STATIC int
1560xfs_iunlink_remove(
1561 xfs_trans_t *tp,
1562 xfs_inode_t *ip)
1563{
1564 xfs_ino_t next_ino;
1565 xfs_mount_t *mp;
1566 xfs_agi_t *agi;
1567 xfs_dinode_t *dip;
1568 xfs_buf_t *agibp;
1569 xfs_buf_t *ibp;
1570 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 xfs_agino_t agino;
1572 xfs_agino_t next_agino;
1573 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001574 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001576 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
1582 /*
1583 * Get the agi buffer first. It ensures lock ordering
1584 * on the list.
1585 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001586 error = xfs_read_agi(mp, tp, agno, &agibp);
1587 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001589
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 /*
1593 * Get the index into the agi hash table for the
1594 * list this inode will go on.
1595 */
1596 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1597 ASSERT(agino != 0);
1598 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001599 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 ASSERT(agi->agi_unlinked[bucket_index]);
1601
Christoph Hellwig16259e72005-11-02 15:11:25 +11001602 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04001604 * We're at the head of the list. Get the inode's on-disk
1605 * buffer to see if there is anyone after us on the list.
1606 * Only modify our next pointer if it is not already NULLAGINO.
1607 * This saves us the overhead of dealing with the buffer when
1608 * there is no need to change it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04001610 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1611 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04001613 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11001614 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 return error;
1616 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001617 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 ASSERT(next_agino != 0);
1619 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001620 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001621 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 offsetof(xfs_dinode_t, di_next_unlinked);
1623 xfs_trans_inode_buf(tp, ibp);
1624 xfs_trans_log_buf(tp, ibp, offset,
1625 (offset + sizeof(xfs_agino_t) - 1));
1626 xfs_inobp_check(mp, ibp);
1627 } else {
1628 xfs_trans_brelse(tp, ibp);
1629 }
1630 /*
1631 * Point the bucket head pointer at the next inode.
1632 */
1633 ASSERT(next_agino != 0);
1634 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001635 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 offset = offsetof(xfs_agi_t, agi_unlinked) +
1637 (sizeof(xfs_agino_t) * bucket_index);
1638 xfs_trans_log_buf(tp, agibp, offset,
1639 (offset + sizeof(xfs_agino_t) - 1));
1640 } else {
1641 /*
1642 * We need to search the list for the inode being freed.
1643 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001644 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 last_ibp = NULL;
1646 while (next_agino != agino) {
Christoph Hellwig129dbc92012-07-03 12:21:51 -04001647 struct xfs_imap imap;
1648
1649 if (last_ibp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 xfs_trans_brelse(tp, last_ibp);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04001651
1652 imap.im_blkno = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04001654
1655 error = xfs_imap(mp, tp, next_ino, &imap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001657 xfs_warn(mp,
Christoph Hellwig129dbc92012-07-03 12:21:51 -04001658 "%s: xfs_imap returned error %d.",
1659 __func__, error);
1660 return error;
1661 }
1662
1663 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
1664 &last_ibp, 0, 0);
1665 if (error) {
1666 xfs_warn(mp,
1667 "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11001668 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 return error;
1670 }
Christoph Hellwig129dbc92012-07-03 12:21:51 -04001671
1672 last_offset = imap.im_boffset;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001673 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 ASSERT(next_agino != NULLAGINO);
1675 ASSERT(next_agino != 0);
1676 }
Christoph Hellwig475ee412012-07-03 12:21:22 -04001677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04001679 * Now last_ibp points to the buffer previous to us on the
1680 * unlinked list. Pull us from the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04001682 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1683 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04001685 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11001686 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 return error;
1688 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001689 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 ASSERT(next_agino != 0);
1691 ASSERT(next_agino != agino);
1692 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001693 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001694 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 offsetof(xfs_dinode_t, di_next_unlinked);
1696 xfs_trans_inode_buf(tp, ibp);
1697 xfs_trans_log_buf(tp, ibp, offset,
1698 (offset + sizeof(xfs_agino_t) - 1));
1699 xfs_inobp_check(mp, ibp);
1700 } else {
1701 xfs_trans_brelse(tp, ibp);
1702 }
1703 /*
1704 * Point the previous inode on the list to the next inode.
1705 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001706 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 ASSERT(next_agino != 0);
1708 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1709 xfs_trans_inode_buf(tp, last_ibp);
1710 xfs_trans_log_buf(tp, last_ibp, offset,
1711 (offset + sizeof(xfs_agino_t) - 1));
1712 xfs_inobp_check(mp, last_ibp);
1713 }
1714 return 0;
1715}
1716
Dave Chinner5b3eed72010-08-24 11:42:41 +10001717/*
1718 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1719 * inodes that are in memory - they all must be marked stale and attached to
1720 * the cluster buffer.
1721 */
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00001722STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723xfs_ifree_cluster(
1724 xfs_inode_t *free_ip,
1725 xfs_trans_t *tp,
1726 xfs_ino_t inum)
1727{
1728 xfs_mount_t *mp = free_ip->i_mount;
1729 int blks_per_cluster;
1730 int nbufs;
1731 int ninodes;
Dave Chinner5b257b42010-06-03 16:22:29 +10001732 int i, j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 xfs_daddr_t blkno;
1734 xfs_buf_t *bp;
Dave Chinner5b257b42010-06-03 16:22:29 +10001735 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 xfs_inode_log_item_t *iip;
1737 xfs_log_item_t *lip;
Dave Chinner5017e972010-01-11 11:47:40 +00001738 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
Dave Chinner5017e972010-01-11 11:47:40 +00001740 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1742 blks_per_cluster = 1;
1743 ninodes = mp->m_sb.sb_inopblock;
1744 nbufs = XFS_IALLOC_BLOCKS(mp);
1745 } else {
1746 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1747 mp->m_sb.sb_blocksize;
1748 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1749 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1750 }
1751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 for (j = 0; j < nbufs; j++, inum += ninodes) {
1753 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1754 XFS_INO_TO_AGBNO(mp, inum));
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 /*
Dave Chinner5b257b42010-06-03 16:22:29 +10001757 * We obtain and lock the backing buffer first in the process
1758 * here, as we have to ensure that any dirty inode that we
1759 * can't get the flush lock on is attached to the buffer.
1760 * If we scan the in-memory inodes first, then buffer IO can
1761 * complete before we get a lock on it, and hence we may fail
1762 * to mark all the active inodes on the buffer stale.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 */
Dave Chinner5b257b42010-06-03 16:22:29 +10001764 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
Dave Chinnerb6aff292012-11-02 11:38:42 +11001765 mp->m_bsize * blks_per_cluster,
1766 XBF_UNMAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00001768 if (!bp)
1769 return ENOMEM;
Dave Chinner5b257b42010-06-03 16:22:29 +10001770 /*
1771 * Walk the inodes already attached to the buffer and mark them
1772 * stale. These will all have the flush locks held, so an
Dave Chinner5b3eed72010-08-24 11:42:41 +10001773 * in-memory inode walk can't lock them. By marking them all
1774 * stale first, we will not attempt to lock them in the loop
1775 * below as the XFS_ISTALE flag will be set.
Dave Chinner5b257b42010-06-03 16:22:29 +10001776 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001777 lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 while (lip) {
1779 if (lip->li_type == XFS_LI_INODE) {
1780 iip = (xfs_inode_log_item_t *)lip;
1781 ASSERT(iip->ili_logged == 1);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001782 lip->li_cb = xfs_istale_done;
David Chinner7b2e2a32008-10-30 17:39:12 +11001783 xfs_trans_ail_copy_lsn(mp->m_ail,
1784 &iip->ili_flush_lsn,
1785 &iip->ili_item.li_lsn);
David Chinnere5ffd2b2006-11-21 18:55:33 +11001786 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 }
1788 lip = lip->li_bio_list;
1789 }
1790
Dave Chinner5b3eed72010-08-24 11:42:41 +10001791
Dave Chinner5b257b42010-06-03 16:22:29 +10001792 /*
1793 * For each inode in memory attempt to add it to the inode
1794 * buffer and set it up for being staled on buffer IO
1795 * completion. This is safe as we've locked out tail pushing
1796 * and flushing by locking the buffer.
1797 *
1798 * We have already marked every inode that was part of a
1799 * transaction stale above, which means there is no point in
1800 * even trying to lock them.
1801 */
1802 for (i = 0; i < ninodes; i++) {
Dave Chinner5b3eed72010-08-24 11:42:41 +10001803retry:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001804 rcu_read_lock();
Dave Chinner5b257b42010-06-03 16:22:29 +10001805 ip = radix_tree_lookup(&pag->pag_ici_root,
1806 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001808 /* Inode not in memory, nothing to do */
1809 if (!ip) {
1810 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10001811 continue;
1812 }
1813
Dave Chinner5b3eed72010-08-24 11:42:41 +10001814 /*
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001815 * because this is an RCU protected lookup, we could
1816 * find a recently freed or even reallocated inode
1817 * during the lookup. We need to check under the
1818 * i_flags_lock for a valid inode here. Skip it if it
1819 * is not valid, the wrong inode or stale.
1820 */
1821 spin_lock(&ip->i_flags_lock);
1822 if (ip->i_ino != inum + i ||
1823 __xfs_iflags_test(ip, XFS_ISTALE)) {
1824 spin_unlock(&ip->i_flags_lock);
1825 rcu_read_unlock();
1826 continue;
1827 }
1828 spin_unlock(&ip->i_flags_lock);
1829
1830 /*
Dave Chinner5b3eed72010-08-24 11:42:41 +10001831 * Don't try to lock/unlock the current inode, but we
1832 * _cannot_ skip the other inodes that we did not find
1833 * in the list attached to the buffer and are not
1834 * already marked stale. If we can't lock it, back off
1835 * and retry.
1836 */
Dave Chinner5b257b42010-06-03 16:22:29 +10001837 if (ip != free_ip &&
1838 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001839 rcu_read_unlock();
Dave Chinner5b3eed72010-08-24 11:42:41 +10001840 delay(1);
1841 goto retry;
Dave Chinner5b257b42010-06-03 16:22:29 +10001842 }
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001843 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10001844
Dave Chinner5b3eed72010-08-24 11:42:41 +10001845 xfs_iflock(ip);
Dave Chinner5b257b42010-06-03 16:22:29 +10001846 xfs_iflags_set(ip, XFS_ISTALE);
Dave Chinner5b257b42010-06-03 16:22:29 +10001847
Dave Chinner5b3eed72010-08-24 11:42:41 +10001848 /*
1849 * we don't need to attach clean inodes or those only
1850 * with unlogged changes (which we throw away, anyway).
1851 */
Dave Chinner5b257b42010-06-03 16:22:29 +10001852 iip = ip->i_itemp;
Dave Chinner5b3eed72010-08-24 11:42:41 +10001853 if (!iip || xfs_inode_clean(ip)) {
Dave Chinner5b257b42010-06-03 16:22:29 +10001854 ASSERT(ip != free_ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 xfs_ifunlock(ip);
1856 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1857 continue;
1858 }
1859
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00001860 iip->ili_last_fields = iip->ili_fields;
1861 iip->ili_fields = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 iip->ili_logged = 1;
David Chinner7b2e2a32008-10-30 17:39:12 +11001863 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1864 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001866 xfs_buf_attach_iodone(bp, xfs_istale_done,
1867 &iip->ili_item);
Dave Chinner5b257b42010-06-03 16:22:29 +10001868
1869 if (ip != free_ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 }
1872
Dave Chinner5b3eed72010-08-24 11:42:41 +10001873 xfs_trans_stale_inode_buf(tp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 xfs_trans_binval(tp, bp);
1875 }
1876
Dave Chinner5017e972010-01-11 11:47:40 +00001877 xfs_perag_put(pag);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00001878 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879}
1880
1881/*
1882 * This is called to return an inode to the inode free list.
1883 * The inode should already be truncated to 0 length and have
1884 * no pages associated with it. This routine also assumes that
1885 * the inode is already a part of the transaction.
1886 *
1887 * The on-disk copy of the inode will have been added to the list
1888 * of unlinked inodes in the AGI. We need to remove the inode from
1889 * that list atomically with respect to freeing it here.
1890 */
1891int
1892xfs_ifree(
1893 xfs_trans_t *tp,
1894 xfs_inode_t *ip,
1895 xfs_bmap_free_t *flist)
1896{
1897 int error;
1898 int delete;
1899 xfs_ino_t first_ino;
Vlad Apostolovc319b582007-11-23 16:27:51 +11001900 xfs_dinode_t *dip;
1901 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001903 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 ASSERT(ip->i_d.di_nlink == 0);
1905 ASSERT(ip->i_d.di_nextents == 0);
1906 ASSERT(ip->i_d.di_anextents == 0);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00001907 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 ASSERT(ip->i_d.di_nblocks == 0);
1909
1910 /*
1911 * Pull the on-disk inode from the AGI unlinked list.
1912 */
1913 error = xfs_iunlink_remove(tp, ip);
1914 if (error != 0) {
1915 return error;
1916 }
1917
1918 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
1919 if (error != 0) {
1920 return error;
1921 }
1922 ip->i_d.di_mode = 0; /* mark incore inode as free */
1923 ip->i_d.di_flags = 0;
1924 ip->i_d.di_dmevmask = 0;
1925 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1927 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1928 /*
1929 * Bump the generation count so no one will be confused
1930 * by reincarnations of this inode.
1931 */
1932 ip->i_d.di_gen++;
Vlad Apostolovc319b582007-11-23 16:27:51 +11001933
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1935
Christoph Hellwig475ee412012-07-03 12:21:22 -04001936 error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &dip, &ibp,
1937 0, 0);
Vlad Apostolovc319b582007-11-23 16:27:51 +11001938 if (error)
1939 return error;
1940
1941 /*
1942 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1943 * from picking up this inode when it is reclaimed (its incore state
1944 * initialzed but not flushed to disk yet). The in-core di_mode is
1945 * already cleared and a corresponding transaction logged.
1946 * The hack here just synchronizes the in-core to on-disk
1947 * di_mode value in advance before the actual inode sync to disk.
1948 * This is OK because the inode is already unlinked and would never
1949 * change its di_mode again for this inode generation.
1950 * This is a temporary hack that would require a proper fix
1951 * in the future.
1952 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11001953 dip->di_mode = 0;
Vlad Apostolovc319b582007-11-23 16:27:51 +11001954
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 if (delete) {
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00001956 error = xfs_ifree_cluster(ip, tp, first_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 }
1958
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00001959 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960}
1961
1962/*
1963 * Reallocate the space for if_broot based on the number of records
1964 * being added or deleted as indicated in rec_diff. Move the records
1965 * and pointers in if_broot to fit the new size. When shrinking this
1966 * will eliminate holes between the records and pointers created by
1967 * the caller. When growing this will create holes to be filled in
1968 * by the caller.
1969 *
1970 * The caller must not request to add more records than would fit in
1971 * the on-disk inode root. If the if_broot is currently NULL, then
1972 * if we adding records one will be allocated. The caller must also
1973 * not request that the number of records go below zero, although
1974 * it can go to zero.
1975 *
1976 * ip -- the inode whose if_broot area is changing
1977 * ext_diff -- the change in the number of records, positive or negative,
1978 * requested for the if_broot array.
1979 */
1980void
1981xfs_iroot_realloc(
1982 xfs_inode_t *ip,
1983 int rec_diff,
1984 int whichfork)
1985{
Christoph Hellwig60197e82008-10-30 17:11:19 +11001986 struct xfs_mount *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 int cur_max;
1988 xfs_ifork_t *ifp;
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11001989 struct xfs_btree_block *new_broot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 int new_max;
1991 size_t new_size;
1992 char *np;
1993 char *op;
1994
1995 /*
1996 * Handle the degenerate case quietly.
1997 */
1998 if (rec_diff == 0) {
1999 return;
2000 }
2001
2002 ifp = XFS_IFORK_PTR(ip, whichfork);
2003 if (rec_diff > 0) {
2004 /*
2005 * If there wasn't any memory allocated before, just
2006 * allocate it now and get out.
2007 */
2008 if (ifp->if_broot_bytes == 0) {
2009 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002010 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 ifp->if_broot_bytes = (int)new_size;
2012 return;
2013 }
2014
2015 /*
2016 * If there is already an existing if_broot, then we need
2017 * to realloc() it and shift the pointers to their new
2018 * location. The records don't change location because
2019 * they are kept butted up against the btree block header.
2020 */
Christoph Hellwig60197e82008-10-30 17:11:19 +11002021 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 new_max = cur_max + rec_diff;
2023 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002024 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002026 KM_SLEEP | KM_NOFS);
Christoph Hellwig60197e82008-10-30 17:11:19 +11002027 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2028 ifp->if_broot_bytes);
2029 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2030 (int)new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 ifp->if_broot_bytes = (int)new_size;
2032 ASSERT(ifp->if_broot_bytes <=
2033 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2034 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2035 return;
2036 }
2037
2038 /*
2039 * rec_diff is less than 0. In this case, we are shrinking the
2040 * if_broot buffer. It must already exist. If we go to zero
2041 * records, just get rid of the root and clear the status bit.
2042 */
2043 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
Christoph Hellwig60197e82008-10-30 17:11:19 +11002044 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 new_max = cur_max + rec_diff;
2046 ASSERT(new_max >= 0);
2047 if (new_max > 0)
2048 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2049 else
2050 new_size = 0;
2051 if (new_size > 0) {
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002052 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 /*
2054 * First copy over the btree block header.
2055 */
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002056 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 } else {
2058 new_broot = NULL;
2059 ifp->if_flags &= ~XFS_IFBROOT;
2060 }
2061
2062 /*
2063 * Only copy the records and pointers if there are any.
2064 */
2065 if (new_max > 0) {
2066 /*
2067 * First copy the records.
2068 */
Christoph Hellwig136341b2008-10-30 17:11:40 +11002069 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2070 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2072
2073 /*
2074 * Then copy the pointers.
2075 */
Christoph Hellwig60197e82008-10-30 17:11:19 +11002076 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 ifp->if_broot_bytes);
Christoph Hellwig60197e82008-10-30 17:11:19 +11002078 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 (int)new_size);
2080 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2081 }
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002082 kmem_free(ifp->if_broot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 ifp->if_broot = new_broot;
2084 ifp->if_broot_bytes = (int)new_size;
2085 ASSERT(ifp->if_broot_bytes <=
2086 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2087 return;
2088}
2089
2090
2091/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 * This is called when the amount of space needed for if_data
2093 * is increased or decreased. The change in size is indicated by
2094 * the number of bytes that need to be added or deleted in the
2095 * byte_diff parameter.
2096 *
2097 * If the amount of space needed has decreased below the size of the
2098 * inline buffer, then switch to using the inline buffer. Otherwise,
2099 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2100 * to what is needed.
2101 *
2102 * ip -- the inode whose if_data area is changing
2103 * byte_diff -- the change in the number of bytes, positive or negative,
2104 * requested for the if_data array.
2105 */
2106void
2107xfs_idata_realloc(
2108 xfs_inode_t *ip,
2109 int byte_diff,
2110 int whichfork)
2111{
2112 xfs_ifork_t *ifp;
2113 int new_size;
2114 int real_size;
2115
2116 if (byte_diff == 0) {
2117 return;
2118 }
2119
2120 ifp = XFS_IFORK_PTR(ip, whichfork);
2121 new_size = (int)ifp->if_bytes + byte_diff;
2122 ASSERT(new_size >= 0);
2123
2124 if (new_size == 0) {
2125 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002126 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 }
2128 ifp->if_u1.if_data = NULL;
2129 real_size = 0;
2130 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2131 /*
2132 * If the valid extents/data can fit in if_inline_ext/data,
2133 * copy them from the malloc'd vector and free it.
2134 */
2135 if (ifp->if_u1.if_data == NULL) {
2136 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2137 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2138 ASSERT(ifp->if_real_bytes != 0);
2139 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2140 new_size);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002141 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2143 }
2144 real_size = 0;
2145 } else {
2146 /*
2147 * Stuck with malloc/realloc.
2148 * For inline data, the underlying buffer must be
2149 * a multiple of 4 bytes in size so that it can be
2150 * logged and stay on word boundaries. We enforce
2151 * that here.
2152 */
2153 real_size = roundup(new_size, 4);
2154 if (ifp->if_u1.if_data == NULL) {
2155 ASSERT(ifp->if_real_bytes == 0);
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002156 ifp->if_u1.if_data = kmem_alloc(real_size,
2157 KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2159 /*
2160 * Only do the realloc if the underlying size
2161 * is really changing.
2162 */
2163 if (ifp->if_real_bytes != real_size) {
2164 ifp->if_u1.if_data =
2165 kmem_realloc(ifp->if_u1.if_data,
2166 real_size,
2167 ifp->if_real_bytes,
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002168 KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 }
2170 } else {
2171 ASSERT(ifp->if_real_bytes == 0);
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002172 ifp->if_u1.if_data = kmem_alloc(real_size,
2173 KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2175 ifp->if_bytes);
2176 }
2177 }
2178 ifp->if_real_bytes = real_size;
2179 ifp->if_bytes = new_size;
2180 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2181}
2182
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183void
2184xfs_idestroy_fork(
2185 xfs_inode_t *ip,
2186 int whichfork)
2187{
2188 xfs_ifork_t *ifp;
2189
2190 ifp = XFS_IFORK_PTR(ip, whichfork);
2191 if (ifp->if_broot != NULL) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002192 kmem_free(ifp->if_broot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 ifp->if_broot = NULL;
2194 }
2195
2196 /*
2197 * If the format is local, then we can't have an extents
2198 * array so just look for an inline data array. If we're
2199 * not local then we may or may not have an extents list,
2200 * so check and free it up if we do.
2201 */
2202 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2203 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2204 (ifp->if_u1.if_data != NULL)) {
2205 ASSERT(ifp->if_real_bytes != 0);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002206 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 ifp->if_u1.if_data = NULL;
2208 ifp->if_real_bytes = 0;
2209 }
2210 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11002211 ((ifp->if_flags & XFS_IFEXTIREC) ||
2212 ((ifp->if_u1.if_extents != NULL) &&
2213 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 ASSERT(ifp->if_real_bytes != 0);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002215 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 }
2217 ASSERT(ifp->if_u1.if_extents == NULL ||
2218 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2219 ASSERT(ifp->if_real_bytes == 0);
2220 if (whichfork == XFS_ATTR_FORK) {
2221 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2222 ip->i_afp = NULL;
2223 }
2224}
2225
2226/*
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002227 * This is called to unpin an inode. The caller must have the inode locked
2228 * in at least shared mode so that the buffer cannot be subsequently pinned
2229 * once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002231static void
Christoph Hellwigf392e632011-12-18 20:00:10 +00002232xfs_iunpin(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002233 struct xfs_inode *ip)
David Chinnera3f74ff2008-03-06 13:43:42 +11002234{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002235 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002236
Dave Chinner4aaf15d2010-03-08 11:24:07 +11002237 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2238
David Chinnera3f74ff2008-03-06 13:43:42 +11002239 /* Give the log a push to start the unpinning I/O */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002240 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002241
David Chinnera3f74ff2008-03-06 13:43:42 +11002242}
2243
Christoph Hellwigf392e632011-12-18 20:00:10 +00002244static void
2245__xfs_iunpin_wait(
2246 struct xfs_inode *ip)
2247{
2248 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2249 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2250
2251 xfs_iunpin(ip);
2252
2253 do {
2254 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2255 if (xfs_ipincount(ip))
2256 io_schedule();
2257 } while (xfs_ipincount(ip));
2258 finish_wait(wq, &wait.wait);
2259}
2260
Dave Chinner777df5a2010-02-06 12:37:26 +11002261void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262xfs_iunpin_wait(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002263 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264{
Christoph Hellwigf392e632011-12-18 20:00:10 +00002265 if (xfs_ipincount(ip))
2266 __xfs_iunpin_wait(ip);
David Chinnera3f74ff2008-03-06 13:43:42 +11002267}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269/*
2270 * xfs_iextents_copy()
2271 *
2272 * This is called to copy the REAL extents (as opposed to the delayed
2273 * allocation extents) from the inode into the given buffer. It
2274 * returns the number of bytes copied into the buffer.
2275 *
2276 * If there are no delayed allocation extents, then we can just
2277 * memcpy() the extents into the buffer. Otherwise, we need to
2278 * examine each extent in turn and skip those which are delayed.
2279 */
2280int
2281xfs_iextents_copy(
2282 xfs_inode_t *ip,
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002283 xfs_bmbt_rec_t *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 int whichfork)
2285{
2286 int copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 int i;
2288 xfs_ifork_t *ifp;
2289 int nrecs;
2290 xfs_fsblock_t start_block;
2291
2292 ifp = XFS_IFORK_PTR(ip, whichfork);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002293 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 ASSERT(ifp->if_bytes > 0);
2295
2296 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
Eric Sandeen3a59c942007-07-11 11:09:47 +10002297 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 ASSERT(nrecs > 0);
2299
2300 /*
2301 * There are some delayed allocation extents in the
2302 * inode, so copy the extents one at a time and skip
2303 * the delayed ones. There must be at least one
2304 * non-delayed extent.
2305 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 copied = 0;
2307 for (i = 0; i < nrecs; i++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002308 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 start_block = xfs_bmbt_get_startblock(ep);
Eric Sandeen9d87c312009-01-14 23:22:07 -06002310 if (isnullstartblock(start_block)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 /*
2312 * It's a delayed allocation extent, so skip it.
2313 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 continue;
2315 }
2316
2317 /* Translate to on disk format */
Christoph Hellwigcd8b0a92007-08-16 16:24:15 +10002318 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2319 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002320 dp++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 copied++;
2322 }
2323 ASSERT(copied != 0);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002324 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
2326 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2327}
2328
2329/*
2330 * Each of the following cases stores data into the same region
2331 * of the on-disk inode, so only one of them can be valid at
2332 * any given time. While it is possible to have conflicting formats
2333 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2334 * in EXTENTS format, this can only happen when the fork has
2335 * changed formats after being modified but before being flushed.
2336 * In these cases, the format always takes precedence, because the
2337 * format indicates the current state of the fork.
2338 */
2339/*ARGSUSED*/
David Chinnere4ac9672008-04-10 12:23:58 +10002340STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341xfs_iflush_fork(
2342 xfs_inode_t *ip,
2343 xfs_dinode_t *dip,
2344 xfs_inode_log_item_t *iip,
2345 int whichfork,
2346 xfs_buf_t *bp)
2347{
2348 char *cp;
2349 xfs_ifork_t *ifp;
2350 xfs_mount_t *mp;
2351#ifdef XFS_TRANS_DEBUG
2352 int first;
2353#endif
2354 static const short brootflag[2] =
2355 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2356 static const short dataflag[2] =
2357 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2358 static const short extflag[2] =
2359 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2360
David Chinnere4ac9672008-04-10 12:23:58 +10002361 if (!iip)
2362 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 ifp = XFS_IFORK_PTR(ip, whichfork);
2364 /*
2365 * This can happen if we gave up in iformat in an error path,
2366 * for the attribute fork.
2367 */
David Chinnere4ac9672008-04-10 12:23:58 +10002368 if (!ifp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 ASSERT(whichfork == XFS_ATTR_FORK);
David Chinnere4ac9672008-04-10 12:23:58 +10002370 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 }
2372 cp = XFS_DFORK_PTR(dip, whichfork);
2373 mp = ip->i_mount;
2374 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2375 case XFS_DINODE_FMT_LOCAL:
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002376 if ((iip->ili_fields & dataflag[whichfork]) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 (ifp->if_bytes > 0)) {
2378 ASSERT(ifp->if_u1.if_data != NULL);
2379 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2380 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 break;
2383
2384 case XFS_DINODE_FMT_EXTENTS:
2385 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002386 !(iip->ili_fields & extflag[whichfork]));
2387 if ((iip->ili_fields & extflag[whichfork]) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 (ifp->if_bytes > 0)) {
Christoph Hellwigab1908a2011-05-11 15:04:10 +00002389 ASSERT(xfs_iext_get_ext(ifp, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2391 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2392 whichfork);
2393 }
2394 break;
2395
2396 case XFS_DINODE_FMT_BTREE:
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002397 if ((iip->ili_fields & brootflag[whichfork]) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 (ifp->if_broot_bytes > 0)) {
2399 ASSERT(ifp->if_broot != NULL);
2400 ASSERT(ifp->if_broot_bytes <=
2401 (XFS_IFORK_SIZE(ip, whichfork) +
2402 XFS_BROOT_SIZE_ADJ));
Christoph Hellwig60197e82008-10-30 17:11:19 +11002403 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 (xfs_bmdr_block_t *)cp,
2405 XFS_DFORK_SIZE(dip, mp, whichfork));
2406 }
2407 break;
2408
2409 case XFS_DINODE_FMT_DEV:
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002410 if (iip->ili_fields & XFS_ILOG_DEV) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 ASSERT(whichfork == XFS_DATA_FORK);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002412 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 }
2414 break;
2415
2416 case XFS_DINODE_FMT_UUID:
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002417 if (iip->ili_fields & XFS_ILOG_UUID) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 ASSERT(whichfork == XFS_DATA_FORK);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002419 memcpy(XFS_DFORK_DPTR(dip),
2420 &ip->i_df.if_u2.if_uuid,
2421 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 }
2423 break;
2424
2425 default:
2426 ASSERT(0);
2427 break;
2428 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429}
2430
David Chinnerbad55842008-03-06 13:43:49 +11002431STATIC int
2432xfs_iflush_cluster(
2433 xfs_inode_t *ip,
2434 xfs_buf_t *bp)
2435{
2436 xfs_mount_t *mp = ip->i_mount;
Dave Chinner5017e972010-01-11 11:47:40 +00002437 struct xfs_perag *pag;
David Chinnerbad55842008-03-06 13:43:49 +11002438 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10002439 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11002440 int ilist_size;
2441 xfs_inode_t **ilist;
2442 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11002443 int nr_found;
2444 int clcount = 0;
2445 int bufwasdelwri;
2446 int i;
2447
Dave Chinner5017e972010-01-11 11:47:40 +00002448 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
David Chinnerbad55842008-03-06 13:43:49 +11002449
David Chinnerc8f5f122008-05-20 11:30:15 +10002450 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2451 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10002452 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11002453 if (!ilist)
Dave Chinner44b56e02010-01-11 11:47:43 +00002454 goto out_put;
David Chinnerbad55842008-03-06 13:43:49 +11002455
2456 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2457 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002458 rcu_read_lock();
David Chinnerbad55842008-03-06 13:43:49 +11002459 /* really need a gang lookup range call here */
2460 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10002461 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11002462 if (nr_found == 0)
2463 goto out_free;
2464
2465 for (i = 0; i < nr_found; i++) {
2466 iq = ilist[i];
2467 if (iq == ip)
2468 continue;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002469
2470 /*
2471 * because this is an RCU protected lookup, we could find a
2472 * recently freed or even reallocated inode during the lookup.
2473 * We need to check under the i_flags_lock for a valid inode
2474 * here. Skip it if it is not valid or the wrong inode.
2475 */
2476 spin_lock(&ip->i_flags_lock);
2477 if (!ip->i_ino ||
2478 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2479 spin_unlock(&ip->i_flags_lock);
2480 continue;
2481 }
2482 spin_unlock(&ip->i_flags_lock);
2483
David Chinnerbad55842008-03-06 13:43:49 +11002484 /*
2485 * Do an un-protected check to see if the inode is dirty and
2486 * is a candidate for flushing. These checks will be repeated
2487 * later after the appropriate locks are acquired.
2488 */
David Chinner33540402008-03-06 13:43:59 +11002489 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11002490 continue;
David Chinnerbad55842008-03-06 13:43:49 +11002491
2492 /*
2493 * Try to get locks. If any are unavailable or it is pinned,
2494 * then this inode cannot be flushed and is skipped.
2495 */
2496
2497 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2498 continue;
2499 if (!xfs_iflock_nowait(iq)) {
2500 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2501 continue;
2502 }
2503 if (xfs_ipincount(iq)) {
2504 xfs_ifunlock(iq);
2505 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2506 continue;
2507 }
2508
2509 /*
2510 * arriving here means that this inode can be flushed. First
2511 * re-check that it's dirty before flushing.
2512 */
David Chinner33540402008-03-06 13:43:59 +11002513 if (!xfs_inode_clean(iq)) {
2514 int error;
David Chinnerbad55842008-03-06 13:43:49 +11002515 error = xfs_iflush_int(iq, bp);
2516 if (error) {
2517 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2518 goto cluster_corrupt_out;
2519 }
2520 clcount++;
2521 } else {
2522 xfs_ifunlock(iq);
2523 }
2524 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2525 }
2526
2527 if (clcount) {
2528 XFS_STATS_INC(xs_icluster_flushcnt);
2529 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2530 }
2531
2532out_free:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002533 rcu_read_unlock();
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002534 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00002535out_put:
2536 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11002537 return 0;
2538
2539
2540cluster_corrupt_out:
2541 /*
2542 * Corruption detected in the clustering loop. Invalidate the
2543 * inode buffer and shut down the filesystem.
2544 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002545 rcu_read_unlock();
David Chinnerbad55842008-03-06 13:43:49 +11002546 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002547 * Clean up the buffer. If it was delwri, just release it --
David Chinnerbad55842008-03-06 13:43:49 +11002548 * brelse can handle it with no problems. If not, shut down the
2549 * filesystem before releasing the buffer.
2550 */
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002551 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
David Chinnerbad55842008-03-06 13:43:49 +11002552 if (bufwasdelwri)
2553 xfs_buf_relse(bp);
2554
2555 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2556
2557 if (!bufwasdelwri) {
2558 /*
2559 * Just like incore_relse: if we have b_iodone functions,
2560 * mark the buffer as an error and call them. Otherwise
2561 * mark it as stale and brelse.
2562 */
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002563 if (bp->b_iodone) {
David Chinnerbad55842008-03-06 13:43:49 +11002564 XFS_BUF_UNDONE(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002565 xfs_buf_stale(bp);
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002566 xfs_buf_ioerror(bp, EIO);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00002567 xfs_buf_ioend(bp, 0);
David Chinnerbad55842008-03-06 13:43:49 +11002568 } else {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002569 xfs_buf_stale(bp);
David Chinnerbad55842008-03-06 13:43:49 +11002570 xfs_buf_relse(bp);
2571 }
2572 }
2573
2574 /*
2575 * Unlocks the flush lock
2576 */
Dave Chinner04913fd2012-04-23 15:58:41 +10002577 xfs_iflush_abort(iq, false);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002578 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00002579 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11002580 return XFS_ERROR(EFSCORRUPTED);
2581}
2582
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583/*
Christoph Hellwig4c468192012-04-23 15:58:36 +10002584 * Flush dirty inode metadata into the backing buffer.
2585 *
2586 * The caller must have the inode lock and the inode flush lock held. The
2587 * inode lock will still be held upon return to the caller, and the inode
2588 * flush lock will be released after the inode has reached the disk.
2589 *
2590 * The caller must write out the buffer returned in *bpp and release it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 */
2592int
2593xfs_iflush(
Christoph Hellwig4c468192012-04-23 15:58:36 +10002594 struct xfs_inode *ip,
2595 struct xfs_buf **bpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596{
Christoph Hellwig4c468192012-04-23 15:58:36 +10002597 struct xfs_mount *mp = ip->i_mount;
2598 struct xfs_buf *bp;
2599 struct xfs_dinode *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601
2602 XFS_STATS_INC(xs_iflush_count);
2603
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002604 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00002605 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00002607 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608
Christoph Hellwig4c468192012-04-23 15:58:36 +10002609 *bpp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 xfs_iunpin_wait(ip);
2612
2613 /*
Dave Chinner4b6a4682010-01-11 11:45:21 +00002614 * For stale inodes we cannot rely on the backing buffer remaining
2615 * stale in cache for the remaining life of the stale inode and so
Christoph Hellwig475ee412012-07-03 12:21:22 -04002616 * xfs_imap_to_bp() below may give us a buffer that no longer contains
Dave Chinner4b6a4682010-01-11 11:45:21 +00002617 * inodes below. We have to check this after ensuring the inode is
2618 * unpinned so that it is safe to reclaim the stale inode after the
2619 * flush call.
2620 */
2621 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2622 xfs_ifunlock(ip);
2623 return 0;
2624 }
2625
2626 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 * This may have been unpinned because the filesystem is shutting
2628 * down forcibly. If that's the case we must not write this inode
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10002629 * to disk, because the log record didn't make it to disk.
2630 *
2631 * We also have to remove the log item from the AIL in this case,
2632 * as we wait for an empty AIL as part of the unmount process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 */
2634 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10002635 error = XFS_ERROR(EIO);
2636 goto abort_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 }
2638
2639 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002640 * Get the buffer containing the on-disk inode.
2641 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002642 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
2643 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11002644 if (error || !bp) {
2645 xfs_ifunlock(ip);
2646 return error;
2647 }
2648
2649 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 * First flush out the inode that xfs_iflush was called with.
2651 */
2652 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11002653 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
2656 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002657 * If the buffer is pinned then push on the log now so we won't
2658 * get stuck waiting in the write for too long.
2659 */
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00002660 if (xfs_buf_ispinned(bp))
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002661 xfs_log_force(mp, 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11002662
2663 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 * inode clustering:
2665 * see if other inodes can be gathered into this write
2666 */
David Chinnerbad55842008-03-06 13:43:49 +11002667 error = xfs_iflush_cluster(ip, bp);
2668 if (error)
2669 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670
Christoph Hellwig4c468192012-04-23 15:58:36 +10002671 *bpp = bp;
2672 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673
2674corrupt_out:
2675 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10002676 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677cluster_corrupt_out:
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10002678 error = XFS_ERROR(EFSCORRUPTED);
2679abort_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 /*
2681 * Unlocks the flush lock
2682 */
Dave Chinner04913fd2012-04-23 15:58:41 +10002683 xfs_iflush_abort(ip, false);
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10002684 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685}
2686
2687
2688STATIC int
2689xfs_iflush_int(
2690 xfs_inode_t *ip,
2691 xfs_buf_t *bp)
2692{
2693 xfs_inode_log_item_t *iip;
2694 xfs_dinode_t *dip;
2695 xfs_mount_t *mp;
2696#ifdef XFS_TRANS_DEBUG
2697 int first;
2698#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002700 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00002701 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00002703 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704
2705 iip = ip->i_itemp;
2706 mp = ip->i_mount;
2707
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 /* set *dip = inode's place in the buffer */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002709 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002711 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002713 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2714 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2715 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 goto corrupt_out;
2717 }
2718 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2719 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002720 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2721 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2722 __func__, ip->i_ino, ip, ip->i_d.di_magic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 goto corrupt_out;
2724 }
Al Viroabbede12011-07-26 02:31:30 -04002725 if (S_ISREG(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 if (XFS_TEST_ERROR(
2727 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2728 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2729 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002730 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2731 "%s: Bad regular inode %Lu, ptr 0x%p",
2732 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 goto corrupt_out;
2734 }
Al Viroabbede12011-07-26 02:31:30 -04002735 } else if (S_ISDIR(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 if (XFS_TEST_ERROR(
2737 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2738 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2739 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2740 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002741 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2742 "%s: Bad directory inode %Lu, ptr 0x%p",
2743 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 goto corrupt_out;
2745 }
2746 }
2747 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2748 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2749 XFS_RANDOM_IFLUSH_5)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002750 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2751 "%s: detected corrupt incore inode %Lu, "
2752 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2753 __func__, ip->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 ip->i_d.di_nextents + ip->i_d.di_anextents,
Dave Chinner6a19d932011-03-07 10:02:35 +11002755 ip->i_d.di_nblocks, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 goto corrupt_out;
2757 }
2758 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2759 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002760 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2761 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2762 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 goto corrupt_out;
2764 }
2765 /*
2766 * bump the flush iteration count, used to detect flushes which
2767 * postdate a log record during recovery.
2768 */
2769
2770 ip->i_d.di_flushiter++;
2771
2772 /*
2773 * Copy the dirty parts of the inode into the on-disk
2774 * inode. We always copy out the core of the inode,
2775 * because if the inode is dirty at all the core must
2776 * be.
2777 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002778 xfs_dinode_to_disk(dip, &ip->i_d);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779
2780 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2781 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2782 ip->i_d.di_flushiter = 0;
2783
2784 /*
2785 * If this is really an old format inode and the superblock version
2786 * has not been updated to support only new format inodes, then
2787 * convert back to the old inode format. If the superblock version
2788 * has been updated, then make the conversion permanent.
2789 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11002790 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2791 if (ip->i_d.di_version == 1) {
Eric Sandeen62118702008-03-06 13:44:28 +11002792 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 /*
2794 * Convert it back.
2795 */
2796 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002797 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 } else {
2799 /*
2800 * The superblock version has already been bumped,
2801 * so just make the conversion to the new inode
2802 * format permanent.
2803 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11002804 ip->i_d.di_version = 2;
2805 dip->di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 ip->i_d.di_onlink = 0;
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002807 dip->di_onlink = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002809 memset(&(dip->di_pad[0]), 0,
2810 sizeof(dip->di_pad));
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00002811 ASSERT(xfs_get_projid(ip) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 }
2813 }
2814
David Chinnere4ac9672008-04-10 12:23:58 +10002815 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
2816 if (XFS_IFORK_Q(ip))
2817 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 xfs_inobp_check(mp, bp);
2819
2820 /*
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002821 * We've recorded everything logged in the inode, so we'd like to clear
2822 * the ili_fields bits so we don't log and flush things unnecessarily.
2823 * However, we can't stop logging all this information until the data
2824 * we've copied into the disk buffer is written to disk. If we did we
2825 * might overwrite the copy of the inode in the log with all the data
2826 * after re-logging only part of it, and in the face of a crash we
2827 * wouldn't have all the data we need to recover.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002829 * What we do is move the bits to the ili_last_fields field. When
2830 * logging the inode, these bits are moved back to the ili_fields field.
2831 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
2832 * know that the information those bits represent is permanently on
2833 * disk. As long as the flush completes before the inode is logged
2834 * again, then both ili_fields and ili_last_fields will be cleared.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002836 * We can play with the ili_fields bits here, because the inode lock
2837 * must be held exclusively in order to set bits there and the flush
2838 * lock protects the ili_last_fields bits. Set ili_logged so the flush
2839 * done routine can tell whether or not to look in the AIL. Also, store
2840 * the current LSN of the inode so that we can tell whether the item has
2841 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
2842 * need the AIL lock, because it is a 64 bit value that cannot be read
2843 * atomically.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 */
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002845 if (iip != NULL && iip->ili_fields != 0) {
2846 iip->ili_last_fields = iip->ili_fields;
2847 iip->ili_fields = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 iip->ili_logged = 1;
2849
David Chinner7b2e2a32008-10-30 17:39:12 +11002850 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2851 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852
2853 /*
2854 * Attach the function xfs_iflush_done to the inode's
2855 * buffer. This will remove the inode from the AIL
2856 * and unlock the inode's flush lock when the inode is
2857 * completely written to disk.
2858 */
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002859 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860
Christoph Hellwigadadbee2011-07-13 13:43:49 +02002861 ASSERT(bp->b_fspriv != NULL);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002862 ASSERT(bp->b_iodone != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 } else {
2864 /*
2865 * We're flushing an inode which is not in the AIL and has
Christoph Hellwig8a9c9982012-02-29 09:53:52 +00002866 * not been logged. For this case we can immediately drop
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 * the inode flush lock because we can avoid the whole
2868 * AIL state thing. It's OK to drop the flush lock now,
2869 * because we've already locked the buffer and to do anything
2870 * you really need both.
2871 */
2872 if (iip != NULL) {
2873 ASSERT(iip->ili_logged == 0);
2874 ASSERT(iip->ili_last_fields == 0);
2875 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
2876 }
2877 xfs_ifunlock(ip);
2878 }
2879
2880 return 0;
2881
2882corrupt_out:
2883 return XFS_ERROR(EFSCORRUPTED);
2884}
2885
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002886/*
2887 * Return a pointer to the extent record at file index idx.
2888 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002889xfs_bmbt_rec_host_t *
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002890xfs_iext_get_ext(
2891 xfs_ifork_t *ifp, /* inode fork pointer */
2892 xfs_extnum_t idx) /* index of target extent */
2893{
2894 ASSERT(idx >= 0);
Christoph Hellwig87bef182011-05-11 15:04:11 +00002895 ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
2896
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11002897 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
2898 return ifp->if_u1.if_ext_irec->er_extbuf;
2899 } else if (ifp->if_flags & XFS_IFEXTIREC) {
2900 xfs_ext_irec_t *erp; /* irec pointer */
2901 int erp_idx = 0; /* irec index */
2902 xfs_extnum_t page_idx = idx; /* ext index in target list */
2903
2904 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
2905 return &erp->er_extbuf[page_idx];
2906 } else if (ifp->if_bytes) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002907 return &ifp->if_u1.if_extents[idx];
2908 } else {
2909 return NULL;
2910 }
2911}
2912
2913/*
2914 * Insert new item(s) into the extent records for incore inode
2915 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2916 */
2917void
2918xfs_iext_insert(
Christoph Hellwig6ef35542009-11-25 00:00:21 +00002919 xfs_inode_t *ip, /* incore inode pointer */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002920 xfs_extnum_t idx, /* starting index of new items */
2921 xfs_extnum_t count, /* number of inserted items */
Christoph Hellwig6ef35542009-11-25 00:00:21 +00002922 xfs_bmbt_irec_t *new, /* items to insert */
2923 int state) /* type of extent conversion */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002924{
Christoph Hellwig6ef35542009-11-25 00:00:21 +00002925 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002926 xfs_extnum_t i; /* extent record index */
2927
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002928 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
2929
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002930 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2931 xfs_iext_add(ifp, idx, count);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002932 for (i = idx; i < idx + count; i++, new++)
2933 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002934}
2935
2936/*
2937 * This is called when the amount of space required for incore file
2938 * extents needs to be increased. The ext_diff parameter stores the
2939 * number of new extents being added and the idx parameter contains
2940 * the extent index where the new extents will be added. If the new
2941 * extents are being appended, then we just need to (re)allocate and
2942 * initialize the space. Otherwise, if the new extents are being
2943 * inserted into the middle of the existing entries, a bit more work
2944 * is required to make room for the new extents to be inserted. The
2945 * caller is responsible for filling in the new extent entries upon
2946 * return.
2947 */
2948void
2949xfs_iext_add(
2950 xfs_ifork_t *ifp, /* inode fork pointer */
2951 xfs_extnum_t idx, /* index to begin adding exts */
Nathan Scottc41564b2006-03-29 08:55:14 +10002952 int ext_diff) /* number of extents to add */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002953{
2954 int byte_diff; /* new bytes being added */
2955 int new_size; /* size of extents after adding */
2956 xfs_extnum_t nextents; /* number of extents in file */
2957
2958 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2959 ASSERT((idx >= 0) && (idx <= nextents));
2960 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
2961 new_size = ifp->if_bytes + byte_diff;
2962 /*
2963 * If the new number of extents (nextents + ext_diff)
2964 * fits inside the inode, then continue to use the inline
2965 * extent buffer.
2966 */
2967 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
2968 if (idx < nextents) {
2969 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
2970 &ifp->if_u2.if_inline_ext[idx],
2971 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2972 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
2973 }
2974 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2975 ifp->if_real_bytes = 0;
2976 }
2977 /*
2978 * Otherwise use a linear (direct) extent list.
2979 * If the extents are currently inside the inode,
2980 * xfs_iext_realloc_direct will switch us from
2981 * inline to direct extent allocation mode.
2982 */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11002983 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002984 xfs_iext_realloc_direct(ifp, new_size);
2985 if (idx < nextents) {
2986 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
2987 &ifp->if_u1.if_extents[idx],
2988 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2989 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
2990 }
2991 }
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11002992 /* Indirection array */
2993 else {
2994 xfs_ext_irec_t *erp;
2995 int erp_idx = 0;
2996 int page_idx = idx;
2997
2998 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
2999 if (ifp->if_flags & XFS_IFEXTIREC) {
3000 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3001 } else {
3002 xfs_iext_irec_init(ifp);
3003 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3004 erp = ifp->if_u1.if_ext_irec;
3005 }
3006 /* Extents fit in target extent page */
3007 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3008 if (page_idx < erp->er_extcount) {
3009 memmove(&erp->er_extbuf[page_idx + ext_diff],
3010 &erp->er_extbuf[page_idx],
3011 (erp->er_extcount - page_idx) *
3012 sizeof(xfs_bmbt_rec_t));
3013 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3014 }
3015 erp->er_extcount += ext_diff;
3016 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3017 }
3018 /* Insert a new extent page */
3019 else if (erp) {
3020 xfs_iext_add_indirect_multi(ifp,
3021 erp_idx, page_idx, ext_diff);
3022 }
3023 /*
3024 * If extent(s) are being appended to the last page in
3025 * the indirection array and the new extent(s) don't fit
3026 * in the page, then erp is NULL and erp_idx is set to
3027 * the next index needed in the indirection array.
3028 */
3029 else {
3030 int count = ext_diff;
3031
3032 while (count) {
3033 erp = xfs_iext_irec_new(ifp, erp_idx);
3034 erp->er_extcount = count;
3035 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3036 if (count) {
3037 erp_idx++;
3038 }
3039 }
3040 }
3041 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003042 ifp->if_bytes = new_size;
3043}
3044
3045/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003046 * This is called when incore extents are being added to the indirection
3047 * array and the new extents do not fit in the target extent list. The
3048 * erp_idx parameter contains the irec index for the target extent list
3049 * in the indirection array, and the idx parameter contains the extent
3050 * index within the list. The number of extents being added is stored
3051 * in the count parameter.
3052 *
3053 * |-------| |-------|
3054 * | | | | idx - number of extents before idx
3055 * | idx | | count |
3056 * | | | | count - number of extents being inserted at idx
3057 * |-------| |-------|
3058 * | count | | nex2 | nex2 - number of extents after idx + count
3059 * |-------| |-------|
3060 */
3061void
3062xfs_iext_add_indirect_multi(
3063 xfs_ifork_t *ifp, /* inode fork pointer */
3064 int erp_idx, /* target extent irec index */
3065 xfs_extnum_t idx, /* index within target list */
3066 int count) /* new extents being added */
3067{
3068 int byte_diff; /* new bytes being added */
3069 xfs_ext_irec_t *erp; /* pointer to irec entry */
3070 xfs_extnum_t ext_diff; /* number of extents to add */
3071 xfs_extnum_t ext_cnt; /* new extents still needed */
3072 xfs_extnum_t nex2; /* extents after idx + count */
3073 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3074 int nlists; /* number of irec's (lists) */
3075
3076 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3077 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3078 nex2 = erp->er_extcount - idx;
3079 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3080
3081 /*
3082 * Save second part of target extent list
3083 * (all extents past */
3084 if (nex2) {
3085 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
David Chinner67850732008-08-13 16:02:51 +10003086 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003087 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3088 erp->er_extcount -= nex2;
3089 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3090 memset(&erp->er_extbuf[idx], 0, byte_diff);
3091 }
3092
3093 /*
3094 * Add the new extents to the end of the target
3095 * list, then allocate new irec record(s) and
3096 * extent buffer(s) as needed to store the rest
3097 * of the new extents.
3098 */
3099 ext_cnt = count;
3100 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3101 if (ext_diff) {
3102 erp->er_extcount += ext_diff;
3103 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3104 ext_cnt -= ext_diff;
3105 }
3106 while (ext_cnt) {
3107 erp_idx++;
3108 erp = xfs_iext_irec_new(ifp, erp_idx);
3109 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3110 erp->er_extcount = ext_diff;
3111 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3112 ext_cnt -= ext_diff;
3113 }
3114
3115 /* Add nex2 extents back to indirection array */
3116 if (nex2) {
3117 xfs_extnum_t ext_avail;
3118 int i;
3119
3120 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3121 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3122 i = 0;
3123 /*
3124 * If nex2 extents fit in the current page, append
3125 * nex2_ep after the new extents.
3126 */
3127 if (nex2 <= ext_avail) {
3128 i = erp->er_extcount;
3129 }
3130 /*
3131 * Otherwise, check if space is available in the
3132 * next page.
3133 */
3134 else if ((erp_idx < nlists - 1) &&
3135 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3136 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3137 erp_idx++;
3138 erp++;
3139 /* Create a hole for nex2 extents */
3140 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3141 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3142 }
3143 /*
3144 * Final choice, create a new extent page for
3145 * nex2 extents.
3146 */
3147 else {
3148 erp_idx++;
3149 erp = xfs_iext_irec_new(ifp, erp_idx);
3150 }
3151 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003152 kmem_free(nex2_ep);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003153 erp->er_extcount += nex2;
3154 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3155 }
3156}
3157
3158/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003159 * This is called when the amount of space required for incore file
3160 * extents needs to be decreased. The ext_diff parameter stores the
3161 * number of extents to be removed and the idx parameter contains
3162 * the extent index where the extents will be removed from.
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003163 *
3164 * If the amount of space needed has decreased below the linear
3165 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3166 * extent array. Otherwise, use kmem_realloc() to adjust the
3167 * size to what is needed.
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003168 */
3169void
3170xfs_iext_remove(
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003171 xfs_inode_t *ip, /* incore inode pointer */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003172 xfs_extnum_t idx, /* index to begin removing exts */
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003173 int ext_diff, /* number of extents to remove */
3174 int state) /* type of extent conversion */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003175{
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003176 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003177 xfs_extnum_t nextents; /* number of extents in file */
3178 int new_size; /* size of extents after removal */
3179
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003180 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3181
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003182 ASSERT(ext_diff > 0);
3183 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3184 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3185
3186 if (new_size == 0) {
3187 xfs_iext_destroy(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003188 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3189 xfs_iext_remove_indirect(ifp, idx, ext_diff);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003190 } else if (ifp->if_real_bytes) {
3191 xfs_iext_remove_direct(ifp, idx, ext_diff);
3192 } else {
3193 xfs_iext_remove_inline(ifp, idx, ext_diff);
3194 }
3195 ifp->if_bytes = new_size;
3196}
3197
3198/*
3199 * This removes ext_diff extents from the inline buffer, beginning
3200 * at extent index idx.
3201 */
3202void
3203xfs_iext_remove_inline(
3204 xfs_ifork_t *ifp, /* inode fork pointer */
3205 xfs_extnum_t idx, /* index to begin removing exts */
3206 int ext_diff) /* number of extents to remove */
3207{
3208 int nextents; /* number of extents in file */
3209
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003210 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003211 ASSERT(idx < XFS_INLINE_EXTS);
3212 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3213 ASSERT(((nextents - ext_diff) > 0) &&
3214 (nextents - ext_diff) < XFS_INLINE_EXTS);
3215
3216 if (idx + ext_diff < nextents) {
3217 memmove(&ifp->if_u2.if_inline_ext[idx],
3218 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3219 (nextents - (idx + ext_diff)) *
3220 sizeof(xfs_bmbt_rec_t));
3221 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3222 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3223 } else {
3224 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3225 ext_diff * sizeof(xfs_bmbt_rec_t));
3226 }
3227}
3228
3229/*
3230 * This removes ext_diff extents from a linear (direct) extent list,
3231 * beginning at extent index idx. If the extents are being removed
3232 * from the end of the list (ie. truncate) then we just need to re-
3233 * allocate the list to remove the extra space. Otherwise, if the
3234 * extents are being removed from the middle of the existing extent
3235 * entries, then we first need to move the extent records beginning
3236 * at idx + ext_diff up in the list to overwrite the records being
3237 * removed, then remove the extra space via kmem_realloc.
3238 */
3239void
3240xfs_iext_remove_direct(
3241 xfs_ifork_t *ifp, /* inode fork pointer */
3242 xfs_extnum_t idx, /* index to begin removing exts */
3243 int ext_diff) /* number of extents to remove */
3244{
3245 xfs_extnum_t nextents; /* number of extents in file */
3246 int new_size; /* size of extents after removal */
3247
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003248 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003249 new_size = ifp->if_bytes -
3250 (ext_diff * sizeof(xfs_bmbt_rec_t));
3251 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3252
3253 if (new_size == 0) {
3254 xfs_iext_destroy(ifp);
3255 return;
3256 }
3257 /* Move extents up in the list (if needed) */
3258 if (idx + ext_diff < nextents) {
3259 memmove(&ifp->if_u1.if_extents[idx],
3260 &ifp->if_u1.if_extents[idx + ext_diff],
3261 (nextents - (idx + ext_diff)) *
3262 sizeof(xfs_bmbt_rec_t));
3263 }
3264 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3265 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3266 /*
3267 * Reallocate the direct extent list. If the extents
3268 * will fit inside the inode then xfs_iext_realloc_direct
3269 * will switch from direct to inline extent allocation
3270 * mode for us.
3271 */
3272 xfs_iext_realloc_direct(ifp, new_size);
3273 ifp->if_bytes = new_size;
3274}
3275
3276/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003277 * This is called when incore extents are being removed from the
3278 * indirection array and the extents being removed span multiple extent
3279 * buffers. The idx parameter contains the file extent index where we
3280 * want to begin removing extents, and the count parameter contains
3281 * how many extents need to be removed.
3282 *
3283 * |-------| |-------|
3284 * | nex1 | | | nex1 - number of extents before idx
3285 * |-------| | count |
3286 * | | | | count - number of extents being removed at idx
3287 * | count | |-------|
3288 * | | | nex2 | nex2 - number of extents after idx + count
3289 * |-------| |-------|
3290 */
3291void
3292xfs_iext_remove_indirect(
3293 xfs_ifork_t *ifp, /* inode fork pointer */
3294 xfs_extnum_t idx, /* index to begin removing extents */
3295 int count) /* number of extents to remove */
3296{
3297 xfs_ext_irec_t *erp; /* indirection array pointer */
3298 int erp_idx = 0; /* indirection array index */
3299 xfs_extnum_t ext_cnt; /* extents left to remove */
3300 xfs_extnum_t ext_diff; /* extents to remove in current list */
3301 xfs_extnum_t nex1; /* number of extents before idx */
3302 xfs_extnum_t nex2; /* extents after idx + count */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003303 int page_idx = idx; /* index in target extent list */
3304
3305 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3306 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3307 ASSERT(erp != NULL);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003308 nex1 = page_idx;
3309 ext_cnt = count;
3310 while (ext_cnt) {
3311 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3312 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3313 /*
3314 * Check for deletion of entire list;
3315 * xfs_iext_irec_remove() updates extent offsets.
3316 */
3317 if (ext_diff == erp->er_extcount) {
3318 xfs_iext_irec_remove(ifp, erp_idx);
3319 ext_cnt -= ext_diff;
3320 nex1 = 0;
3321 if (ext_cnt) {
3322 ASSERT(erp_idx < ifp->if_real_bytes /
3323 XFS_IEXT_BUFSZ);
3324 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3325 nex1 = 0;
3326 continue;
3327 } else {
3328 break;
3329 }
3330 }
3331 /* Move extents up (if needed) */
3332 if (nex2) {
3333 memmove(&erp->er_extbuf[nex1],
3334 &erp->er_extbuf[nex1 + ext_diff],
3335 nex2 * sizeof(xfs_bmbt_rec_t));
3336 }
3337 /* Zero out rest of page */
3338 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3339 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3340 /* Update remaining counters */
3341 erp->er_extcount -= ext_diff;
3342 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3343 ext_cnt -= ext_diff;
3344 nex1 = 0;
3345 erp_idx++;
3346 erp++;
3347 }
3348 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3349 xfs_iext_irec_compact(ifp);
3350}
3351
3352/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003353 * Create, destroy, or resize a linear (direct) block of extents.
3354 */
3355void
3356xfs_iext_realloc_direct(
3357 xfs_ifork_t *ifp, /* inode fork pointer */
3358 int new_size) /* new size of extents */
3359{
3360 int rnew_size; /* real new size of extents */
3361
3362 rnew_size = new_size;
3363
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003364 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3365 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3366 (new_size != ifp->if_real_bytes)));
3367
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003368 /* Free extent records */
3369 if (new_size == 0) {
3370 xfs_iext_destroy(ifp);
3371 }
3372 /* Resize direct extent list and zero any new bytes */
3373 else if (ifp->if_real_bytes) {
3374 /* Check if extents will fit inside the inode */
3375 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3376 xfs_iext_direct_to_inline(ifp, new_size /
3377 (uint)sizeof(xfs_bmbt_rec_t));
3378 ifp->if_bytes = new_size;
3379 return;
3380 }
Vignesh Babu16a087d2007-06-28 16:46:37 +10003381 if (!is_power_of_2(new_size)){
Robert P. J. Day40ebd812007-11-23 16:30:51 +11003382 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003383 }
3384 if (rnew_size != ifp->if_real_bytes) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003385 ifp->if_u1.if_extents =
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003386 kmem_realloc(ifp->if_u1.if_extents,
3387 rnew_size,
David Chinner67850732008-08-13 16:02:51 +10003388 ifp->if_real_bytes, KM_NOFS);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003389 }
3390 if (rnew_size > ifp->if_real_bytes) {
3391 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3392 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3393 rnew_size - ifp->if_real_bytes);
3394 }
3395 }
3396 /*
3397 * Switch from the inline extent buffer to a direct
3398 * extent list. Be sure to include the inline extent
3399 * bytes in new_size.
3400 */
3401 else {
3402 new_size += ifp->if_bytes;
Vignesh Babu16a087d2007-06-28 16:46:37 +10003403 if (!is_power_of_2(new_size)) {
Robert P. J. Day40ebd812007-11-23 16:30:51 +11003404 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003405 }
3406 xfs_iext_inline_to_direct(ifp, rnew_size);
3407 }
3408 ifp->if_real_bytes = rnew_size;
3409 ifp->if_bytes = new_size;
3410}
3411
3412/*
3413 * Switch from linear (direct) extent records to inline buffer.
3414 */
3415void
3416xfs_iext_direct_to_inline(
3417 xfs_ifork_t *ifp, /* inode fork pointer */
3418 xfs_extnum_t nextents) /* number of extents in file */
3419{
3420 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3421 ASSERT(nextents <= XFS_INLINE_EXTS);
3422 /*
3423 * The inline buffer was zeroed when we switched
3424 * from inline to direct extent allocation mode,
3425 * so we don't need to clear it here.
3426 */
3427 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3428 nextents * sizeof(xfs_bmbt_rec_t));
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003429 kmem_free(ifp->if_u1.if_extents);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003430 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3431 ifp->if_real_bytes = 0;
3432}
3433
3434/*
3435 * Switch from inline buffer to linear (direct) extent records.
3436 * new_size should already be rounded up to the next power of 2
3437 * by the caller (when appropriate), so use new_size as it is.
3438 * However, since new_size may be rounded up, we can't update
3439 * if_bytes here. It is the caller's responsibility to update
3440 * if_bytes upon return.
3441 */
3442void
3443xfs_iext_inline_to_direct(
3444 xfs_ifork_t *ifp, /* inode fork pointer */
3445 int new_size) /* number of extents in file */
3446{
David Chinner67850732008-08-13 16:02:51 +10003447 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003448 memset(ifp->if_u1.if_extents, 0, new_size);
3449 if (ifp->if_bytes) {
3450 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3451 ifp->if_bytes);
3452 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3453 sizeof(xfs_bmbt_rec_t));
3454 }
3455 ifp->if_real_bytes = new_size;
3456}
3457
3458/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003459 * Resize an extent indirection array to new_size bytes.
3460 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003461STATIC void
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003462xfs_iext_realloc_indirect(
3463 xfs_ifork_t *ifp, /* inode fork pointer */
3464 int new_size) /* new indirection array size */
3465{
3466 int nlists; /* number of irec's (ex lists) */
3467 int size; /* current indirection array size */
3468
3469 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3470 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3471 size = nlists * sizeof(xfs_ext_irec_t);
3472 ASSERT(ifp->if_real_bytes);
3473 ASSERT((new_size >= 0) && (new_size != size));
3474 if (new_size == 0) {
3475 xfs_iext_destroy(ifp);
3476 } else {
3477 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3478 kmem_realloc(ifp->if_u1.if_ext_irec,
David Chinner67850732008-08-13 16:02:51 +10003479 new_size, size, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003480 }
3481}
3482
3483/*
3484 * Switch from indirection array to linear (direct) extent allocations.
3485 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003486STATIC void
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003487xfs_iext_indirect_to_direct(
3488 xfs_ifork_t *ifp) /* inode fork pointer */
3489{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003490 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003491 xfs_extnum_t nextents; /* number of extents in file */
3492 int size; /* size of file extents */
3493
3494 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3495 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3496 ASSERT(nextents <= XFS_LINEAR_EXTS);
3497 size = nextents * sizeof(xfs_bmbt_rec_t);
3498
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10003499 xfs_iext_irec_compact_pages(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003500 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3501
3502 ep = ifp->if_u1.if_ext_irec->er_extbuf;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003503 kmem_free(ifp->if_u1.if_ext_irec);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003504 ifp->if_flags &= ~XFS_IFEXTIREC;
3505 ifp->if_u1.if_extents = ep;
3506 ifp->if_bytes = size;
3507 if (nextents < XFS_LINEAR_EXTS) {
3508 xfs_iext_realloc_direct(ifp, size);
3509 }
3510}
3511
3512/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003513 * Free incore file extents.
3514 */
3515void
3516xfs_iext_destroy(
3517 xfs_ifork_t *ifp) /* inode fork pointer */
3518{
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003519 if (ifp->if_flags & XFS_IFEXTIREC) {
3520 int erp_idx;
3521 int nlists;
3522
3523 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3524 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3525 xfs_iext_irec_remove(ifp, erp_idx);
3526 }
3527 ifp->if_flags &= ~XFS_IFEXTIREC;
3528 } else if (ifp->if_real_bytes) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003529 kmem_free(ifp->if_u1.if_extents);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003530 } else if (ifp->if_bytes) {
3531 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3532 sizeof(xfs_bmbt_rec_t));
3533 }
3534 ifp->if_u1.if_extents = NULL;
3535 ifp->if_real_bytes = 0;
3536 ifp->if_bytes = 0;
3537}
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003538
3539/*
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003540 * Return a pointer to the extent record for file system block bno.
3541 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003542xfs_bmbt_rec_host_t * /* pointer to found extent record */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003543xfs_iext_bno_to_ext(
3544 xfs_ifork_t *ifp, /* inode fork pointer */
3545 xfs_fileoff_t bno, /* block number to search for */
3546 xfs_extnum_t *idxp) /* index of target extent */
3547{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003548 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003549 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003550 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003551 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
Nathan Scottc41564b2006-03-29 08:55:14 +10003552 int high; /* upper boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003553 xfs_extnum_t idx = 0; /* index of target extent */
Nathan Scottc41564b2006-03-29 08:55:14 +10003554 int low; /* lower boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003555 xfs_extnum_t nextents; /* number of file extents */
3556 xfs_fileoff_t startoff = 0; /* start offset of extent */
3557
3558 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3559 if (nextents == 0) {
3560 *idxp = 0;
3561 return NULL;
3562 }
3563 low = 0;
3564 if (ifp->if_flags & XFS_IFEXTIREC) {
3565 /* Find target extent list */
3566 int erp_idx = 0;
3567 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3568 base = erp->er_extbuf;
3569 high = erp->er_extcount - 1;
3570 } else {
3571 base = ifp->if_u1.if_extents;
3572 high = nextents - 1;
3573 }
3574 /* Binary search extent records */
3575 while (low <= high) {
3576 idx = (low + high) >> 1;
3577 ep = base + idx;
3578 startoff = xfs_bmbt_get_startoff(ep);
3579 blockcount = xfs_bmbt_get_blockcount(ep);
3580 if (bno < startoff) {
3581 high = idx - 1;
3582 } else if (bno >= startoff + blockcount) {
3583 low = idx + 1;
3584 } else {
3585 /* Convert back to file-based extent index */
3586 if (ifp->if_flags & XFS_IFEXTIREC) {
3587 idx += erp->er_extoff;
3588 }
3589 *idxp = idx;
3590 return ep;
3591 }
3592 }
3593 /* Convert back to file-based extent index */
3594 if (ifp->if_flags & XFS_IFEXTIREC) {
3595 idx += erp->er_extoff;
3596 }
3597 if (bno >= startoff + blockcount) {
3598 if (++idx == nextents) {
3599 ep = NULL;
3600 } else {
3601 ep = xfs_iext_get_ext(ifp, idx);
3602 }
3603 }
3604 *idxp = idx;
3605 return ep;
3606}
3607
3608/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003609 * Return a pointer to the indirection array entry containing the
3610 * extent record for filesystem block bno. Store the index of the
3611 * target irec in *erp_idxp.
3612 */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003613xfs_ext_irec_t * /* pointer to found extent record */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003614xfs_iext_bno_to_irec(
3615 xfs_ifork_t *ifp, /* inode fork pointer */
3616 xfs_fileoff_t bno, /* block number to search for */
3617 int *erp_idxp) /* irec index of target ext list */
3618{
3619 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3620 xfs_ext_irec_t *erp_next; /* next indirection array entry */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003621 int erp_idx; /* indirection array index */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003622 int nlists; /* number of extent irec's (lists) */
3623 int high; /* binary search upper limit */
3624 int low; /* binary search lower limit */
3625
3626 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3627 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3628 erp_idx = 0;
3629 low = 0;
3630 high = nlists - 1;
3631 while (low <= high) {
3632 erp_idx = (low + high) >> 1;
3633 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3634 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3635 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3636 high = erp_idx - 1;
3637 } else if (erp_next && bno >=
3638 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3639 low = erp_idx + 1;
3640 } else {
3641 break;
3642 }
3643 }
3644 *erp_idxp = erp_idx;
3645 return erp;
3646}
3647
3648/*
3649 * Return a pointer to the indirection array entry containing the
3650 * extent record at file extent index *idxp. Store the index of the
3651 * target irec in *erp_idxp and store the page index of the target
3652 * extent record in *idxp.
3653 */
3654xfs_ext_irec_t *
3655xfs_iext_idx_to_irec(
3656 xfs_ifork_t *ifp, /* inode fork pointer */
3657 xfs_extnum_t *idxp, /* extent index (file -> page) */
3658 int *erp_idxp, /* pointer to target irec */
3659 int realloc) /* new bytes were just added */
3660{
3661 xfs_ext_irec_t *prev; /* pointer to previous irec */
3662 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3663 int erp_idx; /* indirection array index */
3664 int nlists; /* number of irec's (ex lists) */
3665 int high; /* binary search upper limit */
3666 int low; /* binary search lower limit */
3667 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3668
3669 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
Christoph Hellwig87bef182011-05-11 15:04:11 +00003670 ASSERT(page_idx >= 0);
3671 ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
3672 ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
3673
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003674 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3675 erp_idx = 0;
3676 low = 0;
3677 high = nlists - 1;
3678
3679 /* Binary search extent irec's */
3680 while (low <= high) {
3681 erp_idx = (low + high) >> 1;
3682 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3683 prev = erp_idx > 0 ? erp - 1 : NULL;
3684 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3685 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3686 high = erp_idx - 1;
3687 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3688 (page_idx == erp->er_extoff + erp->er_extcount &&
3689 !realloc)) {
3690 low = erp_idx + 1;
3691 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3692 erp->er_extcount == XFS_LINEAR_EXTS) {
3693 ASSERT(realloc);
3694 page_idx = 0;
3695 erp_idx++;
3696 erp = erp_idx < nlists ? erp + 1 : NULL;
3697 break;
3698 } else {
3699 page_idx -= erp->er_extoff;
3700 break;
3701 }
3702 }
3703 *idxp = page_idx;
3704 *erp_idxp = erp_idx;
3705 return(erp);
3706}
3707
3708/*
3709 * Allocate and initialize an indirection array once the space needed
3710 * for incore extents increases above XFS_IEXT_BUFSZ.
3711 */
3712void
3713xfs_iext_irec_init(
3714 xfs_ifork_t *ifp) /* inode fork pointer */
3715{
3716 xfs_ext_irec_t *erp; /* indirection array pointer */
3717 xfs_extnum_t nextents; /* number of extents in file */
3718
3719 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3720 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3721 ASSERT(nextents <= XFS_LINEAR_EXTS);
3722
David Chinner67850732008-08-13 16:02:51 +10003723 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003724
3725 if (nextents == 0) {
David Chinner67850732008-08-13 16:02:51 +10003726 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003727 } else if (!ifp->if_real_bytes) {
3728 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3729 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3730 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3731 }
3732 erp->er_extbuf = ifp->if_u1.if_extents;
3733 erp->er_extcount = nextents;
3734 erp->er_extoff = 0;
3735
3736 ifp->if_flags |= XFS_IFEXTIREC;
3737 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3738 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3739 ifp->if_u1.if_ext_irec = erp;
3740
3741 return;
3742}
3743
3744/*
3745 * Allocate and initialize a new entry in the indirection array.
3746 */
3747xfs_ext_irec_t *
3748xfs_iext_irec_new(
3749 xfs_ifork_t *ifp, /* inode fork pointer */
3750 int erp_idx) /* index for new irec */
3751{
3752 xfs_ext_irec_t *erp; /* indirection array pointer */
3753 int i; /* loop counter */
3754 int nlists; /* number of irec's (ex lists) */
3755
3756 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3757 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3758
3759 /* Resize indirection array */
3760 xfs_iext_realloc_indirect(ifp, ++nlists *
3761 sizeof(xfs_ext_irec_t));
3762 /*
3763 * Move records down in the array so the
3764 * new page can use erp_idx.
3765 */
3766 erp = ifp->if_u1.if_ext_irec;
3767 for (i = nlists - 1; i > erp_idx; i--) {
3768 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3769 }
3770 ASSERT(i == erp_idx);
3771
3772 /* Initialize new extent record */
3773 erp = ifp->if_u1.if_ext_irec;
David Chinner67850732008-08-13 16:02:51 +10003774 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003775 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3776 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3777 erp[erp_idx].er_extcount = 0;
3778 erp[erp_idx].er_extoff = erp_idx > 0 ?
3779 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3780 return (&erp[erp_idx]);
3781}
3782
3783/*
3784 * Remove a record from the indirection array.
3785 */
3786void
3787xfs_iext_irec_remove(
3788 xfs_ifork_t *ifp, /* inode fork pointer */
3789 int erp_idx) /* irec index to remove */
3790{
3791 xfs_ext_irec_t *erp; /* indirection array pointer */
3792 int i; /* loop counter */
3793 int nlists; /* number of irec's (ex lists) */
3794
3795 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3796 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3797 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3798 if (erp->er_extbuf) {
3799 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
3800 -erp->er_extcount);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003801 kmem_free(erp->er_extbuf);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003802 }
3803 /* Compact extent records */
3804 erp = ifp->if_u1.if_ext_irec;
3805 for (i = erp_idx; i < nlists - 1; i++) {
3806 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
3807 }
3808 /*
3809 * Manually free the last extent record from the indirection
3810 * array. A call to xfs_iext_realloc_indirect() with a size
3811 * of zero would result in a call to xfs_iext_destroy() which
3812 * would in turn call this function again, creating a nasty
3813 * infinite loop.
3814 */
3815 if (--nlists) {
3816 xfs_iext_realloc_indirect(ifp,
3817 nlists * sizeof(xfs_ext_irec_t));
3818 } else {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003819 kmem_free(ifp->if_u1.if_ext_irec);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003820 }
3821 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3822}
3823
3824/*
3825 * This is called to clean up large amounts of unused memory allocated
3826 * by the indirection array. Before compacting anything though, verify
3827 * that the indirection array is still needed and switch back to the
3828 * linear extent list (or even the inline buffer) if possible. The
3829 * compaction policy is as follows:
3830 *
3831 * Full Compaction: Extents fit into a single page (or inline buffer)
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10003832 * Partial Compaction: Extents occupy less than 50% of allocated space
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003833 * No Compaction: Extents occupy at least 50% of allocated space
3834 */
3835void
3836xfs_iext_irec_compact(
3837 xfs_ifork_t *ifp) /* inode fork pointer */
3838{
3839 xfs_extnum_t nextents; /* number of extents in file */
3840 int nlists; /* number of irec's (ex lists) */
3841
3842 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3843 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3844 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3845
3846 if (nextents == 0) {
3847 xfs_iext_destroy(ifp);
3848 } else if (nextents <= XFS_INLINE_EXTS) {
3849 xfs_iext_indirect_to_direct(ifp);
3850 xfs_iext_direct_to_inline(ifp, nextents);
3851 } else if (nextents <= XFS_LINEAR_EXTS) {
3852 xfs_iext_indirect_to_direct(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003853 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
3854 xfs_iext_irec_compact_pages(ifp);
3855 }
3856}
3857
3858/*
3859 * Combine extents from neighboring extent pages.
3860 */
3861void
3862xfs_iext_irec_compact_pages(
3863 xfs_ifork_t *ifp) /* inode fork pointer */
3864{
3865 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
3866 int erp_idx = 0; /* indirection array index */
3867 int nlists; /* number of irec's (ex lists) */
3868
3869 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3870 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3871 while (erp_idx < nlists - 1) {
3872 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3873 erp_next = erp + 1;
3874 if (erp_next->er_extcount <=
3875 (XFS_LINEAR_EXTS - erp->er_extcount)) {
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10003876 memcpy(&erp->er_extbuf[erp->er_extcount],
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003877 erp_next->er_extbuf, erp_next->er_extcount *
3878 sizeof(xfs_bmbt_rec_t));
3879 erp->er_extcount += erp_next->er_extcount;
3880 /*
3881 * Free page before removing extent record
3882 * so er_extoffs don't get modified in
3883 * xfs_iext_irec_remove.
3884 */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003885 kmem_free(erp_next->er_extbuf);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003886 erp_next->er_extbuf = NULL;
3887 xfs_iext_irec_remove(ifp, erp_idx + 1);
3888 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3889 } else {
3890 erp_idx++;
3891 }
3892 }
3893}
3894
3895/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003896 * This is called to update the er_extoff field in the indirection
3897 * array when extents have been added or removed from one of the
3898 * extent lists. erp_idx contains the irec index to begin updating
3899 * at and ext_diff contains the number of extents that were added
3900 * or removed.
3901 */
3902void
3903xfs_iext_irec_update_extoffs(
3904 xfs_ifork_t *ifp, /* inode fork pointer */
3905 int erp_idx, /* irec index to update */
3906 int ext_diff) /* number of new extents */
3907{
3908 int i; /* loop counter */
3909 int nlists; /* number of irec's (ex lists */
3910
3911 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3912 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3913 for (i = erp_idx; i < nlists; i++) {
3914 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
3915 }
3916}
Brian Foster72b53ef2012-11-06 09:50:40 -05003917
3918/*
3919 * Test whether it is appropriate to check an inode for and free post EOF
3920 * blocks. The 'force' parameter determines whether we should also consider
3921 * regular files that are marked preallocated or append-only.
3922 */
3923bool
3924xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
3925{
3926 /* prealloc/delalloc exists only on regular files */
3927 if (!S_ISREG(ip->i_d.di_mode))
3928 return false;
3929
3930 /*
3931 * Zero sized files with no cached pages and delalloc blocks will not
3932 * have speculative prealloc/delalloc blocks to remove.
3933 */
3934 if (VFS_I(ip)->i_size == 0 &&
3935 VN_CACHED(VFS_I(ip)) == 0 &&
3936 ip->i_delayed_blks == 0)
3937 return false;
3938
3939 /* If we haven't read in the extent list, then don't do it now. */
3940 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
3941 return false;
3942
3943 /*
3944 * Do not free real preallocated or append-only files unless the file
3945 * has delalloc blocks and we are forced to remove them.
3946 */
3947 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
3948 if (!force || ip->i_delayed_blks == 0)
3949 return false;
3950
3951 return true;
3952}
3953