blob: c1c89dac19cceceed3aff020651b9eac349cf98e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_dinode.h"
36#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_btree.h"
38#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_quota.h"
40#include "xfs_utils.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42/*
43 * Initialize the inode hash table for the newly mounted file system.
44 * Choose an initial table size based on user specified value, else
45 * use a simple algorithm using the maximum number of inodes as an
46 * indicator for table size, and clamp it between one and some large
47 * number of pages.
48 */
49void
50xfs_ihash_init(xfs_mount_t *mp)
51{
52 __uint64_t icount;
Nathan Scott77e46352006-09-28 11:03:27 +100053 uint i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55 if (!mp->m_ihsize) {
56 icount = mp->m_maxicount ? mp->m_maxicount :
57 (mp->m_sb.sb_dblocks << mp->m_sb.sb_inopblog);
58 mp->m_ihsize = 1 << max_t(uint, 8,
59 (xfs_highbit64(icount) + 1) / 2);
60 mp->m_ihsize = min_t(uint, mp->m_ihsize,
61 (64 * NBPP) / sizeof(xfs_ihash_t));
62 }
63
Nathan Scott77e46352006-09-28 11:03:27 +100064 mp->m_ihash = kmem_zalloc_greedy(&mp->m_ihsize,
65 NBPC * sizeof(xfs_ihash_t),
66 mp->m_ihsize * sizeof(xfs_ihash_t),
67 KM_SLEEP | KM_MAYFAIL | KM_LARGE);
68 mp->m_ihsize /= sizeof(xfs_ihash_t);
69 for (i = 0; i < mp->m_ihsize; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 rwlock_init(&(mp->m_ihash[i].ih_lock));
Linus Torvalds1da177e2005-04-16 15:20:36 -070071}
72
73/*
74 * Free up structures allocated by xfs_ihash_init, at unmount time.
75 */
76void
77xfs_ihash_free(xfs_mount_t *mp)
78{
Nathan Scott77e46352006-09-28 11:03:27 +100079 kmem_free(mp->m_ihash, mp->m_ihsize * sizeof(xfs_ihash_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 mp->m_ihash = NULL;
81}
82
83/*
84 * Initialize the inode cluster hash table for the newly mounted file system.
85 * Its size is derived from the ihash table size.
86 */
87void
88xfs_chash_init(xfs_mount_t *mp)
89{
90 uint i;
91
92 mp->m_chsize = max_t(uint, 1, mp->m_ihsize /
93 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog));
94 mp->m_chsize = min_t(uint, mp->m_chsize, mp->m_ihsize);
95 mp->m_chash = (xfs_chash_t *)kmem_zalloc(mp->m_chsize
96 * sizeof(xfs_chash_t),
Nathan Scottefb8ad72006-09-28 11:03:05 +100097 KM_SLEEP | KM_LARGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 for (i = 0; i < mp->m_chsize; i++) {
99 spinlock_init(&mp->m_chash[i].ch_lock,"xfshash");
100 }
101}
102
103/*
104 * Free up structures allocated by xfs_chash_init, at unmount time.
105 */
106void
107xfs_chash_free(xfs_mount_t *mp)
108{
109 int i;
110
111 for (i = 0; i < mp->m_chsize; i++) {
112 spinlock_destroy(&mp->m_chash[i].ch_lock);
113 }
114
115 kmem_free(mp->m_chash, mp->m_chsize*sizeof(xfs_chash_t));
116 mp->m_chash = NULL;
117}
118
119/*
Nathan Scott71bce252005-05-05 13:23:27 -0700120 * Try to move an inode to the front of its hash list if possible
121 * (and if its not there already). Called right after obtaining
122 * the list version number and then dropping the read_lock on the
123 * hash list in question (which is done right after looking up the
124 * inode in question...).
125 */
126STATIC void
127xfs_ihash_promote(
128 xfs_ihash_t *ih,
129 xfs_inode_t *ip,
130 ulong version)
131{
132 xfs_inode_t *iq;
133
134 if ((ip->i_prevp != &ih->ih_next) && write_trylock(&ih->ih_lock)) {
135 if (likely(version == ih->ih_version)) {
136 /* remove from list */
137 if ((iq = ip->i_next)) {
138 iq->i_prevp = ip->i_prevp;
139 }
140 *ip->i_prevp = iq;
141
142 /* insert at list head */
143 iq = ih->ih_next;
144 iq->i_prevp = &ip->i_next;
145 ip->i_next = iq;
146 ip->i_prevp = &ih->ih_next;
147 ih->ih_next = ip;
148 }
149 write_unlock(&ih->ih_lock);
150 }
151}
152
153/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 * Look up an inode by number in the given file system.
155 * The inode is looked up in the hash table for the file system
156 * represented by the mount point parameter mp. Each bucket of
157 * the hash table is guarded by an individual semaphore.
158 *
159 * If the inode is found in the hash table, its corresponding vnode
160 * is obtained with a call to vn_get(). This call takes care of
161 * coordination with the reclamation of the inode and vnode. Note
162 * that the vmap structure is filled in while holding the hash lock.
163 * This gives us the state of the inode/vnode when we found it and
164 * is used for coordination in vn_get().
165 *
166 * If it is not in core, read it in from the file system's device and
167 * add the inode into the hash table.
168 *
169 * The inode is locked according to the value of the lock_flags parameter.
170 * This flag parameter indicates how and if the inode's IO lock and inode lock
171 * should be taken.
172 *
173 * mp -- the mount point structure for the current file system. It points
174 * to the inode hash table.
175 * tp -- a pointer to the current transaction if there is one. This is
176 * simply passed through to the xfs_iread() call.
177 * ino -- the number of the inode desired. This is the unique identifier
178 * within the file system for the inode being requested.
179 * lock_flags -- flags indicating how to lock the inode. See the comment
180 * for xfs_ilock() for a list of valid values.
181 * bno -- the block number starting the buffer containing the inode,
182 * if known (as by bulkstat), else 0.
183 */
184STATIC int
185xfs_iget_core(
Nathan Scott67fcaa72006-06-09 17:00:52 +1000186 bhv_vnode_t *vp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 xfs_mount_t *mp,
188 xfs_trans_t *tp,
189 xfs_ino_t ino,
190 uint flags,
191 uint lock_flags,
192 xfs_inode_t **ipp,
193 xfs_daddr_t bno)
194{
195 xfs_ihash_t *ih;
196 xfs_inode_t *ip;
197 xfs_inode_t *iq;
Nathan Scott67fcaa72006-06-09 17:00:52 +1000198 bhv_vnode_t *inode_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 ulong version;
200 int error;
201 /* REFERENCED */
202 xfs_chash_t *ch;
203 xfs_chashlist_t *chl, *chlnew;
204 SPLDECL(s);
205
206
207 ih = XFS_IHASH(mp, ino);
208
209again:
210 read_lock(&ih->ih_lock);
211
212 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
213 if (ip->i_ino == ino) {
214 /*
215 * If INEW is set this inode is being set up
216 * we need to pause and try again.
217 */
David Chinner7a18c382006-11-11 18:04:54 +1100218 if (xfs_iflags_test(ip, XFS_INEW)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 read_unlock(&ih->ih_lock);
220 delay(1);
221 XFS_STATS_INC(xs_ig_frecycle);
222
223 goto again;
224 }
225
226 inode_vp = XFS_ITOV_NULL(ip);
227 if (inode_vp == NULL) {
228 /*
229 * If IRECLAIM is set this inode is
230 * on its way out of the system,
231 * we need to pause and try again.
232 */
David Chinner7a18c382006-11-11 18:04:54 +1100233 if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 read_unlock(&ih->ih_lock);
235 delay(1);
236 XFS_STATS_INC(xs_ig_frecycle);
237
238 goto again;
239 }
David Chinner4c606582006-11-11 18:05:00 +1100240 ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE));
241
242 /*
243 * If lookup is racing with unlink, then we
244 * should return an error immediately so we
245 * don't remove it from the reclaim list and
246 * potentially leak the inode.
247 */
248 if ((ip->i_d.di_mode == 0) &&
249 !(flags & XFS_IGET_CREATE)) {
250 read_unlock(&ih->ih_lock);
251 return ENOENT;
252 }
253
254 /*
255 * There may be transactions sitting in the
256 * incore log buffers or being flushed to disk
257 * at this time. We can't clear the
258 * XFS_IRECLAIMABLE flag until these
259 * transactions have hit the disk, otherwise we
260 * will void the guarantee the flag provides
261 * xfs_iunpin()
262 */
263 if (xfs_ipincount(ip)) {
264 read_unlock(&ih->ih_lock);
265 xfs_log_force(mp, 0,
266 XFS_LOG_FORCE|XFS_LOG_SYNC);
267 XFS_STATS_INC(xs_ig_frecycle);
268 goto again;
269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
271 vn_trace_exit(vp, "xfs_iget.alloc",
272 (inst_t *)__return_address);
273
274 XFS_STATS_INC(xs_ig_found);
275
David Chinner7a18c382006-11-11 18:04:54 +1100276 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
Nathan Scott71bce252005-05-05 13:23:27 -0700277 version = ih->ih_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 read_unlock(&ih->ih_lock);
Nathan Scott71bce252005-05-05 13:23:27 -0700279 xfs_ihash_promote(ih, ip, version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281 XFS_MOUNT_ILOCK(mp);
282 list_del_init(&ip->i_reclaim);
283 XFS_MOUNT_IUNLOCK(mp);
284
285 goto finish_inode;
286
287 } else if (vp != inode_vp) {
Nathan Scottec86dc02006-03-17 17:25:36 +1100288 struct inode *inode = vn_to_inode(inode_vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
290 /* The inode is being torn down, pause and
291 * try again.
292 */
293 if (inode->i_state & (I_FREEING | I_CLEAR)) {
294 read_unlock(&ih->ih_lock);
295 delay(1);
296 XFS_STATS_INC(xs_ig_frecycle);
297
298 goto again;
299 }
300/* Chances are the other vnode (the one in the inode) is being torn
301 * down right now, and we landed on top of it. Question is, what do
302 * we do? Unhook the old inode and hook up the new one?
303 */
304 cmn_err(CE_PANIC,
305 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
306 inode_vp, vp);
307 }
308
Nathan Scott71bce252005-05-05 13:23:27 -0700309 /*
310 * Inode cache hit: if ip is not at the front of
311 * its hash chain, move it there now.
312 * Do this with the lock held for update, but
313 * do statistics after releasing the lock.
314 */
315 version = ih->ih_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 read_unlock(&ih->ih_lock);
Nathan Scott71bce252005-05-05 13:23:27 -0700317 xfs_ihash_promote(ih, ip, version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 XFS_STATS_INC(xs_ig_found);
319
320finish_inode:
321 if (ip->i_d.di_mode == 0) {
Nathan Scott745b1f472006-09-28 11:02:23 +1000322 if (!(flags & XFS_IGET_CREATE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 return ENOENT;
324 xfs_iocore_inode_reinit(ip);
325 }
Nathan Scott745b1f472006-09-28 11:02:23 +1000326
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 if (lock_flags != 0)
328 xfs_ilock(ip, lock_flags);
329
David Chinner7a18c382006-11-11 18:04:54 +1100330 xfs_iflags_clear(ip, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 vn_trace_exit(vp, "xfs_iget.found",
332 (inst_t *)__return_address);
333 goto return_ip;
334 }
335 }
336
337 /*
338 * Inode cache miss: save the hash chain version stamp and unlock
339 * the chain, so we don't deadlock in vn_alloc.
340 */
341 XFS_STATS_INC(xs_ig_missed);
342
343 version = ih->ih_version;
344
345 read_unlock(&ih->ih_lock);
346
347 /*
348 * Read the disk inode attributes into a new inode structure and get
349 * a new vnode for it. This should also initialize i_ino and i_mount.
350 */
Nathan Scott745b1f472006-09-28 11:02:23 +1000351 error = xfs_iread(mp, tp, ino, &ip, bno,
352 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
353 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
356 vn_trace_exit(vp, "xfs_iget.alloc", (inst_t *)__return_address);
357
358 xfs_inode_lock_init(ip, vp);
359 xfs_iocore_inode_init(ip);
360
Nathan Scott745b1f472006-09-28 11:02:23 +1000361 if (lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 xfs_ilock(ip, lock_flags);
Nathan Scott745b1f472006-09-28 11:02:23 +1000363
364 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 xfs_idestroy(ip);
366 return ENOENT;
367 }
368
369 /*
370 * Put ip on its hash chain, unless someone else hashed a duplicate
371 * after we released the hash lock.
372 */
373 write_lock(&ih->ih_lock);
374
375 if (ih->ih_version != version) {
376 for (iq = ih->ih_next; iq != NULL; iq = iq->i_next) {
377 if (iq->i_ino == ino) {
378 write_unlock(&ih->ih_lock);
379 xfs_idestroy(ip);
380
381 XFS_STATS_INC(xs_ig_dup);
382 goto again;
383 }
384 }
385 }
386
387 /*
388 * These values _must_ be set before releasing ihlock!
389 */
390 ip->i_hash = ih;
391 if ((iq = ih->ih_next)) {
392 iq->i_prevp = &ip->i_next;
393 }
394 ip->i_next = iq;
395 ip->i_prevp = &ih->ih_next;
396 ih->ih_next = ip;
397 ip->i_udquot = ip->i_gdquot = NULL;
398 ih->ih_version++;
David Chinner7a18c382006-11-11 18:04:54 +1100399 xfs_iflags_set(ip, XFS_INEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 write_unlock(&ih->ih_lock);
401
402 /*
403 * put ip on its cluster's hash chain
404 */
405 ASSERT(ip->i_chash == NULL && ip->i_cprev == NULL &&
406 ip->i_cnext == NULL);
407
408 chlnew = NULL;
409 ch = XFS_CHASH(mp, ip->i_blkno);
410 chlredo:
411 s = mutex_spinlock(&ch->ch_lock);
412 for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) {
413 if (chl->chl_blkno == ip->i_blkno) {
414
415 /* insert this inode into the doubly-linked list
416 * where chl points */
417 if ((iq = chl->chl_ip)) {
418 ip->i_cprev = iq->i_cprev;
419 iq->i_cprev->i_cnext = ip;
420 iq->i_cprev = ip;
421 ip->i_cnext = iq;
422 } else {
423 ip->i_cnext = ip;
424 ip->i_cprev = ip;
425 }
426 chl->chl_ip = ip;
427 ip->i_chash = chl;
428 break;
429 }
430 }
431
432 /* no hash list found for this block; add a new hash list */
433 if (chl == NULL) {
434 if (chlnew == NULL) {
435 mutex_spinunlock(&ch->ch_lock, s);
436 ASSERT(xfs_chashlist_zone != NULL);
437 chlnew = (xfs_chashlist_t *)
438 kmem_zone_alloc(xfs_chashlist_zone,
439 KM_SLEEP);
440 ASSERT(chlnew != NULL);
441 goto chlredo;
442 } else {
443 ip->i_cnext = ip;
444 ip->i_cprev = ip;
445 ip->i_chash = chlnew;
446 chlnew->chl_ip = ip;
447 chlnew->chl_blkno = ip->i_blkno;
David Chinner1fc5d952006-04-11 15:11:12 +1000448 if (ch->ch_list)
449 ch->ch_list->chl_prev = chlnew;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 chlnew->chl_next = ch->ch_list;
David Chinner1fc5d952006-04-11 15:11:12 +1000451 chlnew->chl_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 ch->ch_list = chlnew;
453 chlnew = NULL;
454 }
455 } else {
456 if (chlnew != NULL) {
457 kmem_zone_free(xfs_chashlist_zone, chlnew);
458 }
459 }
460
461 mutex_spinunlock(&ch->ch_lock, s);
462
463
464 /*
465 * Link ip to its mount and thread it on the mount's inode list.
466 */
467 XFS_MOUNT_ILOCK(mp);
468 if ((iq = mp->m_inodes)) {
469 ASSERT(iq->i_mprev->i_mnext == iq);
470 ip->i_mprev = iq->i_mprev;
471 iq->i_mprev->i_mnext = ip;
472 iq->i_mprev = ip;
473 ip->i_mnext = iq;
474 } else {
475 ip->i_mnext = ip;
476 ip->i_mprev = ip;
477 }
478 mp->m_inodes = ip;
479
480 XFS_MOUNT_IUNLOCK(mp);
481
482 return_ip:
483 ASSERT(ip->i_df.if_ext_max ==
484 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
485
486 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
487 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
488
489 *ipp = ip;
490
491 /*
492 * If we have a real type for an on-disk inode, we can set ops(&unlock)
493 * now. If it's a new inode being created, xfs_ialloc will handle it.
494 */
Nathan Scottb83bd132006-06-09 16:48:30 +1000495 bhv_vfs_init_vnode(XFS_MTOVFS(mp), vp, XFS_ITOBHV(ip), 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
497 return 0;
498}
499
500
501/*
502 * The 'normal' internal xfs_iget, if needed it will
503 * 'allocate', or 'get', the vnode.
504 */
505int
506xfs_iget(
507 xfs_mount_t *mp,
508 xfs_trans_t *tp,
509 xfs_ino_t ino,
510 uint flags,
511 uint lock_flags,
512 xfs_inode_t **ipp,
513 xfs_daddr_t bno)
514{
515 struct inode *inode;
Nathan Scott67fcaa72006-06-09 17:00:52 +1000516 bhv_vnode_t *vp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 int error;
518
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 XFS_STATS_INC(xs_ig_attempts);
520
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000521retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if ((inode = iget_locked(XFS_MTOVFS(mp)->vfs_super, ino))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Nathan Scottec86dc02006-03-17 17:25:36 +1100525 vp = vn_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 if (inode->i_state & I_NEW) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 vn_initialize(inode);
528 error = xfs_iget_core(vp, mp, tp, ino, flags,
529 lock_flags, ipp, bno);
530 if (error) {
531 vn_mark_bad(vp);
532 if (inode->i_state & I_NEW)
533 unlock_new_inode(inode);
534 iput(inode);
535 }
536 } else {
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000537 /*
538 * If the inode is not fully constructed due to
Nathan Scottc41564b2006-03-29 08:55:14 +1000539 * filehandle mismatches wait for the inode to go
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000540 * away and try again.
541 *
542 * iget_locked will call __wait_on_freeing_inode
543 * to wait for the inode to go away.
544 */
545 if (is_bad_inode(inode) ||
Christoph Hellwig75e17b32006-01-11 20:58:44 +1100546 ((ip = xfs_vtoi(vp)) == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 iput(inode);
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000548 delay(1);
549 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 }
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 if (lock_flags != 0)
553 xfs_ilock(ip, lock_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 XFS_STATS_INC(xs_ig_found);
555 *ipp = ip;
556 error = 0;
557 }
558 } else
559 error = ENOMEM; /* If we got no inode we are out of memory */
560
561 return error;
562}
563
564/*
565 * Do the setup for the various locks within the incore inode.
566 */
567void
568xfs_inode_lock_init(
569 xfs_inode_t *ip,
Nathan Scott67fcaa72006-06-09 17:00:52 +1000570 bhv_vnode_t *vp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
572 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
573 "xfsino", (long)vp->v_number);
574 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", vp->v_number);
575 init_waitqueue_head(&ip->i_ipin_wait);
576 atomic_set(&ip->i_pincount, 0);
Eric Sandeen7ae67d72006-09-28 11:05:46 +1000577 initnsema(&ip->i_flock, 1, "xfsfino");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
580/*
581 * Look for the inode corresponding to the given ino in the hash table.
582 * If it is there and its i_transp pointer matches tp, return it.
583 * Otherwise, return NULL.
584 */
585xfs_inode_t *
586xfs_inode_incore(xfs_mount_t *mp,
587 xfs_ino_t ino,
588 xfs_trans_t *tp)
589{
590 xfs_ihash_t *ih;
591 xfs_inode_t *ip;
Nathan Scott71bce252005-05-05 13:23:27 -0700592 ulong version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594 ih = XFS_IHASH(mp, ino);
595 read_lock(&ih->ih_lock);
596 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
597 if (ip->i_ino == ino) {
598 /*
599 * If we find it and tp matches, return it.
Nathan Scott71bce252005-05-05 13:23:27 -0700600 * Also move it to the front of the hash list
601 * if we find it and it is not already there.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 * Otherwise break from the loop and return
603 * NULL.
604 */
605 if (ip->i_transp == tp) {
Nathan Scott71bce252005-05-05 13:23:27 -0700606 version = ih->ih_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 read_unlock(&ih->ih_lock);
Nathan Scott71bce252005-05-05 13:23:27 -0700608 xfs_ihash_promote(ih, ip, version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 return (ip);
610 }
611 break;
612 }
613 }
614 read_unlock(&ih->ih_lock);
615 return (NULL);
616}
617
618/*
619 * Decrement reference count of an inode structure and unlock it.
620 *
621 * ip -- the inode being released
622 * lock_flags -- this parameter indicates the inode's locks to be
623 * to be released. See the comment on xfs_iunlock() for a list
624 * of valid values.
625 */
626void
627xfs_iput(xfs_inode_t *ip,
628 uint lock_flags)
629{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000630 bhv_vnode_t *vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
632 vn_trace_entry(vp, "xfs_iput", (inst_t *)__return_address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 xfs_iunlock(ip, lock_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 VN_RELE(vp);
635}
636
637/*
638 * Special iput for brand-new inodes that are still locked
639 */
640void
641xfs_iput_new(xfs_inode_t *ip,
642 uint lock_flags)
643{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000644 bhv_vnode_t *vp = XFS_ITOV(ip);
Nathan Scottec86dc02006-03-17 17:25:36 +1100645 struct inode *inode = vn_to_inode(vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647 vn_trace_entry(vp, "xfs_iput_new", (inst_t *)__return_address);
648
649 if ((ip->i_d.di_mode == 0)) {
David Chinner7a18c382006-11-11 18:04:54 +1100650 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 vn_mark_bad(vp);
652 }
653 if (inode->i_state & I_NEW)
654 unlock_new_inode(inode);
655 if (lock_flags)
656 xfs_iunlock(ip, lock_flags);
657 VN_RELE(vp);
658}
659
660
661/*
662 * This routine embodies the part of the reclaim code that pulls
663 * the inode from the inode hash table and the mount structure's
664 * inode list.
665 * This should only be called from xfs_reclaim().
666 */
667void
668xfs_ireclaim(xfs_inode_t *ip)
669{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000670 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
672 /*
673 * Remove from old hash list and mount list.
674 */
675 XFS_STATS_INC(xs_ig_reclaims);
676
677 xfs_iextract(ip);
678
679 /*
680 * Here we do a spurious inode lock in order to coordinate with
681 * xfs_sync(). This is because xfs_sync() references the inodes
682 * in the mount list without taking references on the corresponding
683 * vnodes. We make that OK here by ensuring that we wait until
684 * the inode is unlocked in xfs_sync() before we go ahead and
685 * free it. We get both the regular lock and the io lock because
686 * the xfs_sync() code may need to drop the regular one but will
687 * still hold the io lock.
688 */
689 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
690
691 /*
692 * Release dquots (and their references) if any. An inode may escape
693 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
694 */
695 XFS_QM_DQDETACH(ip->i_mount, ip);
696
697 /*
698 * Pull our behavior descriptor from the vnode chain.
699 */
700 vp = XFS_ITOV_NULL(ip);
701 if (vp) {
702 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
703 }
704
705 /*
706 * Free all memory associated with the inode.
707 */
Tim Shimmin439b8432006-11-11 18:04:34 +1100708 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 xfs_idestroy(ip);
710}
711
712/*
713 * This routine removes an about-to-be-destroyed inode from
714 * all of the lists in which it is located with the exception
715 * of the behavior chain.
716 */
717void
718xfs_iextract(
719 xfs_inode_t *ip)
720{
721 xfs_ihash_t *ih;
722 xfs_inode_t *iq;
723 xfs_mount_t *mp;
724 xfs_chash_t *ch;
725 xfs_chashlist_t *chl, *chm;
726 SPLDECL(s);
727
728 ih = ip->i_hash;
729 write_lock(&ih->ih_lock);
730 if ((iq = ip->i_next)) {
731 iq->i_prevp = ip->i_prevp;
732 }
733 *ip->i_prevp = iq;
Nathan Scott71bce252005-05-05 13:23:27 -0700734 ih->ih_version++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 write_unlock(&ih->ih_lock);
736
737 /*
738 * Remove from cluster hash list
739 * 1) delete the chashlist if this is the last inode on the chashlist
740 * 2) unchain from list of inodes
741 * 3) point chashlist->chl_ip to 'chl_next' if to this inode.
742 */
743 mp = ip->i_mount;
744 ch = XFS_CHASH(mp, ip->i_blkno);
745 s = mutex_spinlock(&ch->ch_lock);
746
747 if (ip->i_cnext == ip) {
748 /* Last inode on chashlist */
749 ASSERT(ip->i_cnext == ip && ip->i_cprev == ip);
750 ASSERT(ip->i_chash != NULL);
751 chm=NULL;
David Chinner1fc5d952006-04-11 15:11:12 +1000752 chl = ip->i_chash;
753 if (chl->chl_prev)
754 chl->chl_prev->chl_next = chl->chl_next;
755 else
756 ch->ch_list = chl->chl_next;
757 if (chl->chl_next)
758 chl->chl_next->chl_prev = chl->chl_prev;
759 kmem_zone_free(xfs_chashlist_zone, chl);
760 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 /* delete one inode from a non-empty list */
762 iq = ip->i_cnext;
763 iq->i_cprev = ip->i_cprev;
764 ip->i_cprev->i_cnext = iq;
765 if (ip->i_chash->chl_ip == ip) {
766 ip->i_chash->chl_ip = iq;
767 }
768 ip->i_chash = __return_address;
769 ip->i_cprev = __return_address;
770 ip->i_cnext = __return_address;
771 }
772 mutex_spinunlock(&ch->ch_lock, s);
773
774 /*
775 * Remove from mount's inode list.
776 */
777 XFS_MOUNT_ILOCK(mp);
778 ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
779 iq = ip->i_mnext;
780 iq->i_mprev = ip->i_mprev;
781 ip->i_mprev->i_mnext = iq;
782
783 /*
784 * Fix up the head pointer if it points to the inode being deleted.
785 */
786 if (mp->m_inodes == ip) {
787 if (ip == iq) {
788 mp->m_inodes = NULL;
789 } else {
790 mp->m_inodes = iq;
791 }
792 }
793
794 /* Deal with the deleted inodes list */
795 list_del_init(&ip->i_reclaim);
796
797 mp->m_ireclaims++;
798 XFS_MOUNT_IUNLOCK(mp);
799}
800
801/*
802 * This is a wrapper routine around the xfs_ilock() routine
803 * used to centralize some grungy code. It is used in places
804 * that wish to lock the inode solely for reading the extents.
805 * The reason these places can't just call xfs_ilock(SHARED)
806 * is that the inode lock also guards to bringing in of the
807 * extents from disk for a file in b-tree format. If the inode
808 * is in b-tree format, then we need to lock the inode exclusively
809 * until the extents are read in. Locking it exclusively all
810 * the time would limit our parallelism unnecessarily, though.
811 * What we do instead is check to see if the extents have been
812 * read in yet, and only lock the inode exclusively if they
813 * have not.
814 *
815 * The function returns a value which should be given to the
816 * corresponding xfs_iunlock_map_shared(). This value is
817 * the mode in which the lock was actually taken.
818 */
819uint
820xfs_ilock_map_shared(
821 xfs_inode_t *ip)
822{
823 uint lock_mode;
824
825 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
826 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
827 lock_mode = XFS_ILOCK_EXCL;
828 } else {
829 lock_mode = XFS_ILOCK_SHARED;
830 }
831
832 xfs_ilock(ip, lock_mode);
833
834 return lock_mode;
835}
836
837/*
838 * This is simply the unlock routine to go with xfs_ilock_map_shared().
839 * All it does is call xfs_iunlock() with the given lock_mode.
840 */
841void
842xfs_iunlock_map_shared(
843 xfs_inode_t *ip,
844 unsigned int lock_mode)
845{
846 xfs_iunlock(ip, lock_mode);
847}
848
849/*
850 * The xfs inode contains 2 locks: a multi-reader lock called the
851 * i_iolock and a multi-reader lock called the i_lock. This routine
852 * allows either or both of the locks to be obtained.
853 *
854 * The 2 locks should always be ordered so that the IO lock is
855 * obtained first in order to prevent deadlock.
856 *
857 * ip -- the inode being locked
858 * lock_flags -- this parameter indicates the inode's locks
859 * to be locked. It can be:
860 * XFS_IOLOCK_SHARED,
861 * XFS_IOLOCK_EXCL,
862 * XFS_ILOCK_SHARED,
863 * XFS_ILOCK_EXCL,
864 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
865 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
866 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
867 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
868 */
869void
870xfs_ilock(xfs_inode_t *ip,
871 uint lock_flags)
872{
873 /*
874 * You can't set both SHARED and EXCL for the same lock,
875 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
876 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
877 */
878 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
879 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
880 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
881 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
882 ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0);
883
884 if (lock_flags & XFS_IOLOCK_EXCL) {
885 mrupdate(&ip->i_iolock);
886 } else if (lock_flags & XFS_IOLOCK_SHARED) {
887 mraccess(&ip->i_iolock);
888 }
889 if (lock_flags & XFS_ILOCK_EXCL) {
890 mrupdate(&ip->i_lock);
891 } else if (lock_flags & XFS_ILOCK_SHARED) {
892 mraccess(&ip->i_lock);
893 }
894 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
895}
896
897/*
898 * This is just like xfs_ilock(), except that the caller
899 * is guaranteed not to sleep. It returns 1 if it gets
900 * the requested locks and 0 otherwise. If the IO lock is
901 * obtained but the inode lock cannot be, then the IO lock
902 * is dropped before returning.
903 *
904 * ip -- the inode being locked
905 * lock_flags -- this parameter indicates the inode's locks to be
906 * to be locked. See the comment for xfs_ilock() for a list
907 * of valid values.
908 *
909 */
910int
911xfs_ilock_nowait(xfs_inode_t *ip,
912 uint lock_flags)
913{
914 int iolocked;
915 int ilocked;
916
917 /*
918 * You can't set both SHARED and EXCL for the same lock,
919 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
920 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
921 */
922 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
923 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
924 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
925 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
926 ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0);
927
928 iolocked = 0;
929 if (lock_flags & XFS_IOLOCK_EXCL) {
930 iolocked = mrtryupdate(&ip->i_iolock);
931 if (!iolocked) {
932 return 0;
933 }
934 } else if (lock_flags & XFS_IOLOCK_SHARED) {
935 iolocked = mrtryaccess(&ip->i_iolock);
936 if (!iolocked) {
937 return 0;
938 }
939 }
940 if (lock_flags & XFS_ILOCK_EXCL) {
941 ilocked = mrtryupdate(&ip->i_lock);
942 if (!ilocked) {
943 if (iolocked) {
944 mrunlock(&ip->i_iolock);
945 }
946 return 0;
947 }
948 } else if (lock_flags & XFS_ILOCK_SHARED) {
949 ilocked = mrtryaccess(&ip->i_lock);
950 if (!ilocked) {
951 if (iolocked) {
952 mrunlock(&ip->i_iolock);
953 }
954 return 0;
955 }
956 }
957 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
958 return 1;
959}
960
961/*
962 * xfs_iunlock() is used to drop the inode locks acquired with
963 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
964 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
965 * that we know which locks to drop.
966 *
967 * ip -- the inode being unlocked
968 * lock_flags -- this parameter indicates the inode's locks to be
969 * to be unlocked. See the comment for xfs_ilock() for a list
970 * of valid values for this parameter.
971 *
972 */
973void
974xfs_iunlock(xfs_inode_t *ip,
975 uint lock_flags)
976{
977 /*
978 * You can't set both SHARED and EXCL for the same lock,
979 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
980 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
981 */
982 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
983 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
984 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
985 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
986 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY)) == 0);
987 ASSERT(lock_flags != 0);
988
989 if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) {
990 ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) ||
991 (ismrlocked(&ip->i_iolock, MR_ACCESS)));
992 ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) ||
993 (ismrlocked(&ip->i_iolock, MR_UPDATE)));
994 mrunlock(&ip->i_iolock);
995 }
996
997 if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) {
998 ASSERT(!(lock_flags & XFS_ILOCK_SHARED) ||
999 (ismrlocked(&ip->i_lock, MR_ACCESS)));
1000 ASSERT(!(lock_flags & XFS_ILOCK_EXCL) ||
1001 (ismrlocked(&ip->i_lock, MR_UPDATE)));
1002 mrunlock(&ip->i_lock);
1003
1004 /*
1005 * Let the AIL know that this item has been unlocked in case
1006 * it is in the AIL and anyone is waiting on it. Don't do
1007 * this if the caller has asked us not to.
1008 */
1009 if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) &&
1010 ip->i_itemp != NULL) {
1011 xfs_trans_unlocked_item(ip->i_mount,
1012 (xfs_log_item_t*)(ip->i_itemp));
1013 }
1014 }
1015 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
1016}
1017
1018/*
1019 * give up write locks. the i/o lock cannot be held nested
1020 * if it is being demoted.
1021 */
1022void
1023xfs_ilock_demote(xfs_inode_t *ip,
1024 uint lock_flags)
1025{
1026 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
1027 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
1028
1029 if (lock_flags & XFS_ILOCK_EXCL) {
1030 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
1031 mrdemote(&ip->i_lock);
1032 }
1033 if (lock_flags & XFS_IOLOCK_EXCL) {
1034 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1035 mrdemote(&ip->i_iolock);
1036 }
1037}
1038
1039/*
1040 * The following three routines simply manage the i_flock
1041 * semaphore embedded in the inode. This semaphore synchronizes
1042 * processes attempting to flush the in-core inode back to disk.
1043 */
1044void
1045xfs_iflock(xfs_inode_t *ip)
1046{
1047 psema(&(ip->i_flock), PINOD|PLTWAIT);
1048}
1049
1050int
1051xfs_iflock_nowait(xfs_inode_t *ip)
1052{
1053 return (cpsema(&(ip->i_flock)));
1054}
1055
1056void
1057xfs_ifunlock(xfs_inode_t *ip)
1058{
Al Viro0d8fee32006-06-19 08:41:30 +10001059 ASSERT(issemalocked(&(ip->i_flock)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 vsema(&(ip->i_flock));
1061}