blob: 0724df7fabb755fea973ca3ac87176b4e4c24e93 [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;
53 uint i, flags = KM_SLEEP | KM_MAYFAIL;
54
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
64 while (!(mp->m_ihash = (xfs_ihash_t *)kmem_zalloc(mp->m_ihsize *
65 sizeof(xfs_ihash_t), flags))) {
66 if ((mp->m_ihsize >>= 1) <= NBPP)
67 flags = KM_SLEEP;
68 }
69 for (i = 0; i < mp->m_ihsize; i++) {
70 rwlock_init(&(mp->m_ihash[i].ih_lock));
71 }
72}
73
74/*
75 * Free up structures allocated by xfs_ihash_init, at unmount time.
76 */
77void
78xfs_ihash_free(xfs_mount_t *mp)
79{
80 kmem_free(mp->m_ihash, mp->m_ihsize*sizeof(xfs_ihash_t));
81 mp->m_ihash = NULL;
82}
83
84/*
85 * Initialize the inode cluster hash table for the newly mounted file system.
86 * Its size is derived from the ihash table size.
87 */
88void
89xfs_chash_init(xfs_mount_t *mp)
90{
91 uint i;
92
93 mp->m_chsize = max_t(uint, 1, mp->m_ihsize /
94 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog));
95 mp->m_chsize = min_t(uint, mp->m_chsize, mp->m_ihsize);
96 mp->m_chash = (xfs_chash_t *)kmem_zalloc(mp->m_chsize
97 * sizeof(xfs_chash_t),
98 KM_SLEEP);
99 for (i = 0; i < mp->m_chsize; i++) {
100 spinlock_init(&mp->m_chash[i].ch_lock,"xfshash");
101 }
102}
103
104/*
105 * Free up structures allocated by xfs_chash_init, at unmount time.
106 */
107void
108xfs_chash_free(xfs_mount_t *mp)
109{
110 int i;
111
112 for (i = 0; i < mp->m_chsize; i++) {
113 spinlock_destroy(&mp->m_chash[i].ch_lock);
114 }
115
116 kmem_free(mp->m_chash, mp->m_chsize*sizeof(xfs_chash_t));
117 mp->m_chash = NULL;
118}
119
120/*
Nathan Scott71bce252005-05-05 13:23:27 -0700121 * Try to move an inode to the front of its hash list if possible
122 * (and if its not there already). Called right after obtaining
123 * the list version number and then dropping the read_lock on the
124 * hash list in question (which is done right after looking up the
125 * inode in question...).
126 */
127STATIC void
128xfs_ihash_promote(
129 xfs_ihash_t *ih,
130 xfs_inode_t *ip,
131 ulong version)
132{
133 xfs_inode_t *iq;
134
135 if ((ip->i_prevp != &ih->ih_next) && write_trylock(&ih->ih_lock)) {
136 if (likely(version == ih->ih_version)) {
137 /* remove from list */
138 if ((iq = ip->i_next)) {
139 iq->i_prevp = ip->i_prevp;
140 }
141 *ip->i_prevp = iq;
142
143 /* insert at list head */
144 iq = ih->ih_next;
145 iq->i_prevp = &ip->i_next;
146 ip->i_next = iq;
147 ip->i_prevp = &ih->ih_next;
148 ih->ih_next = ip;
149 }
150 write_unlock(&ih->ih_lock);
151 }
152}
153
154/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 * Look up an inode by number in the given file system.
156 * The inode is looked up in the hash table for the file system
157 * represented by the mount point parameter mp. Each bucket of
158 * the hash table is guarded by an individual semaphore.
159 *
160 * If the inode is found in the hash table, its corresponding vnode
161 * is obtained with a call to vn_get(). This call takes care of
162 * coordination with the reclamation of the inode and vnode. Note
163 * that the vmap structure is filled in while holding the hash lock.
164 * This gives us the state of the inode/vnode when we found it and
165 * is used for coordination in vn_get().
166 *
167 * If it is not in core, read it in from the file system's device and
168 * add the inode into the hash table.
169 *
170 * The inode is locked according to the value of the lock_flags parameter.
171 * This flag parameter indicates how and if the inode's IO lock and inode lock
172 * should be taken.
173 *
174 * mp -- the mount point structure for the current file system. It points
175 * to the inode hash table.
176 * tp -- a pointer to the current transaction if there is one. This is
177 * simply passed through to the xfs_iread() call.
178 * ino -- the number of the inode desired. This is the unique identifier
179 * within the file system for the inode being requested.
180 * lock_flags -- flags indicating how to lock the inode. See the comment
181 * for xfs_ilock() for a list of valid values.
182 * bno -- the block number starting the buffer containing the inode,
183 * if known (as by bulkstat), else 0.
184 */
185STATIC int
186xfs_iget_core(
Nathan Scott67fcaa72006-06-09 17:00:52 +1000187 bhv_vnode_t *vp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 xfs_mount_t *mp,
189 xfs_trans_t *tp,
190 xfs_ino_t ino,
191 uint flags,
192 uint lock_flags,
193 xfs_inode_t **ipp,
194 xfs_daddr_t bno)
195{
196 xfs_ihash_t *ih;
197 xfs_inode_t *ip;
198 xfs_inode_t *iq;
Nathan Scott67fcaa72006-06-09 17:00:52 +1000199 bhv_vnode_t *inode_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 ulong version;
201 int error;
202 /* REFERENCED */
203 xfs_chash_t *ch;
204 xfs_chashlist_t *chl, *chlnew;
205 SPLDECL(s);
206
207
208 ih = XFS_IHASH(mp, ino);
209
210again:
211 read_lock(&ih->ih_lock);
212
213 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
214 if (ip->i_ino == ino) {
215 /*
216 * If INEW is set this inode is being set up
217 * we need to pause and try again.
218 */
219 if (ip->i_flags & XFS_INEW) {
220 read_unlock(&ih->ih_lock);
221 delay(1);
222 XFS_STATS_INC(xs_ig_frecycle);
223
224 goto again;
225 }
226
227 inode_vp = XFS_ITOV_NULL(ip);
228 if (inode_vp == NULL) {
229 /*
230 * If IRECLAIM is set this inode is
231 * on its way out of the system,
232 * we need to pause and try again.
233 */
234 if (ip->i_flags & XFS_IRECLAIM) {
235 read_unlock(&ih->ih_lock);
236 delay(1);
237 XFS_STATS_INC(xs_ig_frecycle);
238
239 goto again;
240 }
241
242 vn_trace_exit(vp, "xfs_iget.alloc",
243 (inst_t *)__return_address);
244
245 XFS_STATS_INC(xs_ig_found);
246
247 ip->i_flags &= ~XFS_IRECLAIMABLE;
Nathan Scott71bce252005-05-05 13:23:27 -0700248 version = ih->ih_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 read_unlock(&ih->ih_lock);
Nathan Scott71bce252005-05-05 13:23:27 -0700250 xfs_ihash_promote(ih, ip, version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 XFS_MOUNT_ILOCK(mp);
253 list_del_init(&ip->i_reclaim);
254 XFS_MOUNT_IUNLOCK(mp);
255
256 goto finish_inode;
257
258 } else if (vp != inode_vp) {
Nathan Scottec86dc02006-03-17 17:25:36 +1100259 struct inode *inode = vn_to_inode(inode_vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 /* The inode is being torn down, pause and
262 * try again.
263 */
264 if (inode->i_state & (I_FREEING | I_CLEAR)) {
265 read_unlock(&ih->ih_lock);
266 delay(1);
267 XFS_STATS_INC(xs_ig_frecycle);
268
269 goto again;
270 }
271/* Chances are the other vnode (the one in the inode) is being torn
272 * down right now, and we landed on top of it. Question is, what do
273 * we do? Unhook the old inode and hook up the new one?
274 */
275 cmn_err(CE_PANIC,
276 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
277 inode_vp, vp);
278 }
279
Nathan Scott71bce252005-05-05 13:23:27 -0700280 /*
281 * Inode cache hit: if ip is not at the front of
282 * its hash chain, move it there now.
283 * Do this with the lock held for update, but
284 * do statistics after releasing the lock.
285 */
286 version = ih->ih_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 read_unlock(&ih->ih_lock);
Nathan Scott71bce252005-05-05 13:23:27 -0700288 xfs_ihash_promote(ih, ip, version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 XFS_STATS_INC(xs_ig_found);
290
291finish_inode:
292 if (ip->i_d.di_mode == 0) {
293 if (!(flags & IGET_CREATE))
294 return ENOENT;
295 xfs_iocore_inode_reinit(ip);
296 }
297
298 if (lock_flags != 0)
299 xfs_ilock(ip, lock_flags);
300
301 ip->i_flags &= ~XFS_ISTALE;
302
303 vn_trace_exit(vp, "xfs_iget.found",
304 (inst_t *)__return_address);
305 goto return_ip;
306 }
307 }
308
309 /*
310 * Inode cache miss: save the hash chain version stamp and unlock
311 * the chain, so we don't deadlock in vn_alloc.
312 */
313 XFS_STATS_INC(xs_ig_missed);
314
315 version = ih->ih_version;
316
317 read_unlock(&ih->ih_lock);
318
319 /*
320 * Read the disk inode attributes into a new inode structure and get
321 * a new vnode for it. This should also initialize i_ino and i_mount.
322 */
323 error = xfs_iread(mp, tp, ino, &ip, bno);
324 if (error) {
325 return error;
326 }
327
328 vn_trace_exit(vp, "xfs_iget.alloc", (inst_t *)__return_address);
329
330 xfs_inode_lock_init(ip, vp);
331 xfs_iocore_inode_init(ip);
332
333 if (lock_flags != 0) {
334 xfs_ilock(ip, lock_flags);
335 }
336
337 if ((ip->i_d.di_mode == 0) && !(flags & IGET_CREATE)) {
338 xfs_idestroy(ip);
339 return ENOENT;
340 }
341
342 /*
343 * Put ip on its hash chain, unless someone else hashed a duplicate
344 * after we released the hash lock.
345 */
346 write_lock(&ih->ih_lock);
347
348 if (ih->ih_version != version) {
349 for (iq = ih->ih_next; iq != NULL; iq = iq->i_next) {
350 if (iq->i_ino == ino) {
351 write_unlock(&ih->ih_lock);
352 xfs_idestroy(ip);
353
354 XFS_STATS_INC(xs_ig_dup);
355 goto again;
356 }
357 }
358 }
359
360 /*
361 * These values _must_ be set before releasing ihlock!
362 */
363 ip->i_hash = ih;
364 if ((iq = ih->ih_next)) {
365 iq->i_prevp = &ip->i_next;
366 }
367 ip->i_next = iq;
368 ip->i_prevp = &ih->ih_next;
369 ih->ih_next = ip;
370 ip->i_udquot = ip->i_gdquot = NULL;
371 ih->ih_version++;
372 ip->i_flags |= XFS_INEW;
373
374 write_unlock(&ih->ih_lock);
375
376 /*
377 * put ip on its cluster's hash chain
378 */
379 ASSERT(ip->i_chash == NULL && ip->i_cprev == NULL &&
380 ip->i_cnext == NULL);
381
382 chlnew = NULL;
383 ch = XFS_CHASH(mp, ip->i_blkno);
384 chlredo:
385 s = mutex_spinlock(&ch->ch_lock);
386 for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) {
387 if (chl->chl_blkno == ip->i_blkno) {
388
389 /* insert this inode into the doubly-linked list
390 * where chl points */
391 if ((iq = chl->chl_ip)) {
392 ip->i_cprev = iq->i_cprev;
393 iq->i_cprev->i_cnext = ip;
394 iq->i_cprev = ip;
395 ip->i_cnext = iq;
396 } else {
397 ip->i_cnext = ip;
398 ip->i_cprev = ip;
399 }
400 chl->chl_ip = ip;
401 ip->i_chash = chl;
402 break;
403 }
404 }
405
406 /* no hash list found for this block; add a new hash list */
407 if (chl == NULL) {
408 if (chlnew == NULL) {
409 mutex_spinunlock(&ch->ch_lock, s);
410 ASSERT(xfs_chashlist_zone != NULL);
411 chlnew = (xfs_chashlist_t *)
412 kmem_zone_alloc(xfs_chashlist_zone,
413 KM_SLEEP);
414 ASSERT(chlnew != NULL);
415 goto chlredo;
416 } else {
417 ip->i_cnext = ip;
418 ip->i_cprev = ip;
419 ip->i_chash = chlnew;
420 chlnew->chl_ip = ip;
421 chlnew->chl_blkno = ip->i_blkno;
David Chinner1fc5d952006-04-11 15:11:12 +1000422 if (ch->ch_list)
423 ch->ch_list->chl_prev = chlnew;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 chlnew->chl_next = ch->ch_list;
David Chinner1fc5d952006-04-11 15:11:12 +1000425 chlnew->chl_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 ch->ch_list = chlnew;
427 chlnew = NULL;
428 }
429 } else {
430 if (chlnew != NULL) {
431 kmem_zone_free(xfs_chashlist_zone, chlnew);
432 }
433 }
434
435 mutex_spinunlock(&ch->ch_lock, s);
436
437
438 /*
439 * Link ip to its mount and thread it on the mount's inode list.
440 */
441 XFS_MOUNT_ILOCK(mp);
442 if ((iq = mp->m_inodes)) {
443 ASSERT(iq->i_mprev->i_mnext == iq);
444 ip->i_mprev = iq->i_mprev;
445 iq->i_mprev->i_mnext = ip;
446 iq->i_mprev = ip;
447 ip->i_mnext = iq;
448 } else {
449 ip->i_mnext = ip;
450 ip->i_mprev = ip;
451 }
452 mp->m_inodes = ip;
453
454 XFS_MOUNT_IUNLOCK(mp);
455
456 return_ip:
457 ASSERT(ip->i_df.if_ext_max ==
458 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
459
460 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
461 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
462
463 *ipp = ip;
464
465 /*
466 * If we have a real type for an on-disk inode, we can set ops(&unlock)
467 * now. If it's a new inode being created, xfs_ialloc will handle it.
468 */
Nathan Scottb83bd132006-06-09 16:48:30 +1000469 bhv_vfs_init_vnode(XFS_MTOVFS(mp), vp, XFS_ITOBHV(ip), 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
471 return 0;
472}
473
474
475/*
476 * The 'normal' internal xfs_iget, if needed it will
477 * 'allocate', or 'get', the vnode.
478 */
479int
480xfs_iget(
481 xfs_mount_t *mp,
482 xfs_trans_t *tp,
483 xfs_ino_t ino,
484 uint flags,
485 uint lock_flags,
486 xfs_inode_t **ipp,
487 xfs_daddr_t bno)
488{
489 struct inode *inode;
Nathan Scott67fcaa72006-06-09 17:00:52 +1000490 bhv_vnode_t *vp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 int error;
492
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 XFS_STATS_INC(xs_ig_attempts);
494
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000495retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 if ((inode = iget_locked(XFS_MTOVFS(mp)->vfs_super, ino))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Nathan Scottec86dc02006-03-17 17:25:36 +1100499 vp = vn_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (inode->i_state & I_NEW) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 vn_initialize(inode);
502 error = xfs_iget_core(vp, mp, tp, ino, flags,
503 lock_flags, ipp, bno);
504 if (error) {
505 vn_mark_bad(vp);
506 if (inode->i_state & I_NEW)
507 unlock_new_inode(inode);
508 iput(inode);
509 }
510 } else {
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000511 /*
512 * If the inode is not fully constructed due to
Nathan Scottc41564b2006-03-29 08:55:14 +1000513 * filehandle mismatches wait for the inode to go
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000514 * away and try again.
515 *
516 * iget_locked will call __wait_on_freeing_inode
517 * to wait for the inode to go away.
518 */
519 if (is_bad_inode(inode) ||
Christoph Hellwig75e17b32006-01-11 20:58:44 +1100520 ((ip = xfs_vtoi(vp)) == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 iput(inode);
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000522 delay(1);
523 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 }
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 if (lock_flags != 0)
527 xfs_ilock(ip, lock_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 XFS_STATS_INC(xs_ig_found);
529 *ipp = ip;
530 error = 0;
531 }
532 } else
533 error = ENOMEM; /* If we got no inode we are out of memory */
534
535 return error;
536}
537
538/*
539 * Do the setup for the various locks within the incore inode.
540 */
541void
542xfs_inode_lock_init(
543 xfs_inode_t *ip,
Nathan Scott67fcaa72006-06-09 17:00:52 +1000544 bhv_vnode_t *vp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
546 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
547 "xfsino", (long)vp->v_number);
548 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", vp->v_number);
549 init_waitqueue_head(&ip->i_ipin_wait);
550 atomic_set(&ip->i_pincount, 0);
551 init_sema(&ip->i_flock, 1, "xfsfino", vp->v_number);
552}
553
554/*
555 * Look for the inode corresponding to the given ino in the hash table.
556 * If it is there and its i_transp pointer matches tp, return it.
557 * Otherwise, return NULL.
558 */
559xfs_inode_t *
560xfs_inode_incore(xfs_mount_t *mp,
561 xfs_ino_t ino,
562 xfs_trans_t *tp)
563{
564 xfs_ihash_t *ih;
565 xfs_inode_t *ip;
Nathan Scott71bce252005-05-05 13:23:27 -0700566 ulong version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
568 ih = XFS_IHASH(mp, ino);
569 read_lock(&ih->ih_lock);
570 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
571 if (ip->i_ino == ino) {
572 /*
573 * If we find it and tp matches, return it.
Nathan Scott71bce252005-05-05 13:23:27 -0700574 * Also move it to the front of the hash list
575 * if we find it and it is not already there.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 * Otherwise break from the loop and return
577 * NULL.
578 */
579 if (ip->i_transp == tp) {
Nathan Scott71bce252005-05-05 13:23:27 -0700580 version = ih->ih_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 read_unlock(&ih->ih_lock);
Nathan Scott71bce252005-05-05 13:23:27 -0700582 xfs_ihash_promote(ih, ip, version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 return (ip);
584 }
585 break;
586 }
587 }
588 read_unlock(&ih->ih_lock);
589 return (NULL);
590}
591
592/*
593 * Decrement reference count of an inode structure and unlock it.
594 *
595 * ip -- the inode being released
596 * lock_flags -- this parameter indicates the inode's locks to be
597 * to be released. See the comment on xfs_iunlock() for a list
598 * of valid values.
599 */
600void
601xfs_iput(xfs_inode_t *ip,
602 uint lock_flags)
603{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000604 bhv_vnode_t *vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606 vn_trace_entry(vp, "xfs_iput", (inst_t *)__return_address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 xfs_iunlock(ip, lock_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 VN_RELE(vp);
609}
610
611/*
612 * Special iput for brand-new inodes that are still locked
613 */
614void
615xfs_iput_new(xfs_inode_t *ip,
616 uint lock_flags)
617{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000618 bhv_vnode_t *vp = XFS_ITOV(ip);
Nathan Scottec86dc02006-03-17 17:25:36 +1100619 struct inode *inode = vn_to_inode(vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 vn_trace_entry(vp, "xfs_iput_new", (inst_t *)__return_address);
622
623 if ((ip->i_d.di_mode == 0)) {
624 ASSERT(!(ip->i_flags & XFS_IRECLAIMABLE));
625 vn_mark_bad(vp);
626 }
627 if (inode->i_state & I_NEW)
628 unlock_new_inode(inode);
629 if (lock_flags)
630 xfs_iunlock(ip, lock_flags);
631 VN_RELE(vp);
632}
633
634
635/*
636 * This routine embodies the part of the reclaim code that pulls
637 * the inode from the inode hash table and the mount structure's
638 * inode list.
639 * This should only be called from xfs_reclaim().
640 */
641void
642xfs_ireclaim(xfs_inode_t *ip)
643{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000644 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
646 /*
647 * Remove from old hash list and mount list.
648 */
649 XFS_STATS_INC(xs_ig_reclaims);
650
651 xfs_iextract(ip);
652
653 /*
654 * Here we do a spurious inode lock in order to coordinate with
655 * xfs_sync(). This is because xfs_sync() references the inodes
656 * in the mount list without taking references on the corresponding
657 * vnodes. We make that OK here by ensuring that we wait until
658 * the inode is unlocked in xfs_sync() before we go ahead and
659 * free it. We get both the regular lock and the io lock because
660 * the xfs_sync() code may need to drop the regular one but will
661 * still hold the io lock.
662 */
663 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
664
665 /*
666 * Release dquots (and their references) if any. An inode may escape
667 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
668 */
669 XFS_QM_DQDETACH(ip->i_mount, ip);
670
671 /*
672 * Pull our behavior descriptor from the vnode chain.
673 */
674 vp = XFS_ITOV_NULL(ip);
675 if (vp) {
676 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
677 }
678
679 /*
680 * Free all memory associated with the inode.
681 */
682 xfs_idestroy(ip);
683}
684
685/*
686 * This routine removes an about-to-be-destroyed inode from
687 * all of the lists in which it is located with the exception
688 * of the behavior chain.
689 */
690void
691xfs_iextract(
692 xfs_inode_t *ip)
693{
694 xfs_ihash_t *ih;
695 xfs_inode_t *iq;
696 xfs_mount_t *mp;
697 xfs_chash_t *ch;
698 xfs_chashlist_t *chl, *chm;
699 SPLDECL(s);
700
701 ih = ip->i_hash;
702 write_lock(&ih->ih_lock);
703 if ((iq = ip->i_next)) {
704 iq->i_prevp = ip->i_prevp;
705 }
706 *ip->i_prevp = iq;
Nathan Scott71bce252005-05-05 13:23:27 -0700707 ih->ih_version++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 write_unlock(&ih->ih_lock);
709
710 /*
711 * Remove from cluster hash list
712 * 1) delete the chashlist if this is the last inode on the chashlist
713 * 2) unchain from list of inodes
714 * 3) point chashlist->chl_ip to 'chl_next' if to this inode.
715 */
716 mp = ip->i_mount;
717 ch = XFS_CHASH(mp, ip->i_blkno);
718 s = mutex_spinlock(&ch->ch_lock);
719
720 if (ip->i_cnext == ip) {
721 /* Last inode on chashlist */
722 ASSERT(ip->i_cnext == ip && ip->i_cprev == ip);
723 ASSERT(ip->i_chash != NULL);
724 chm=NULL;
David Chinner1fc5d952006-04-11 15:11:12 +1000725 chl = ip->i_chash;
726 if (chl->chl_prev)
727 chl->chl_prev->chl_next = chl->chl_next;
728 else
729 ch->ch_list = chl->chl_next;
730 if (chl->chl_next)
731 chl->chl_next->chl_prev = chl->chl_prev;
732 kmem_zone_free(xfs_chashlist_zone, chl);
733 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 /* delete one inode from a non-empty list */
735 iq = ip->i_cnext;
736 iq->i_cprev = ip->i_cprev;
737 ip->i_cprev->i_cnext = iq;
738 if (ip->i_chash->chl_ip == ip) {
739 ip->i_chash->chl_ip = iq;
740 }
741 ip->i_chash = __return_address;
742 ip->i_cprev = __return_address;
743 ip->i_cnext = __return_address;
744 }
745 mutex_spinunlock(&ch->ch_lock, s);
746
747 /*
748 * Remove from mount's inode list.
749 */
750 XFS_MOUNT_ILOCK(mp);
751 ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
752 iq = ip->i_mnext;
753 iq->i_mprev = ip->i_mprev;
754 ip->i_mprev->i_mnext = iq;
755
756 /*
757 * Fix up the head pointer if it points to the inode being deleted.
758 */
759 if (mp->m_inodes == ip) {
760 if (ip == iq) {
761 mp->m_inodes = NULL;
762 } else {
763 mp->m_inodes = iq;
764 }
765 }
766
767 /* Deal with the deleted inodes list */
768 list_del_init(&ip->i_reclaim);
769
770 mp->m_ireclaims++;
771 XFS_MOUNT_IUNLOCK(mp);
772}
773
774/*
775 * This is a wrapper routine around the xfs_ilock() routine
776 * used to centralize some grungy code. It is used in places
777 * that wish to lock the inode solely for reading the extents.
778 * The reason these places can't just call xfs_ilock(SHARED)
779 * is that the inode lock also guards to bringing in of the
780 * extents from disk for a file in b-tree format. If the inode
781 * is in b-tree format, then we need to lock the inode exclusively
782 * until the extents are read in. Locking it exclusively all
783 * the time would limit our parallelism unnecessarily, though.
784 * What we do instead is check to see if the extents have been
785 * read in yet, and only lock the inode exclusively if they
786 * have not.
787 *
788 * The function returns a value which should be given to the
789 * corresponding xfs_iunlock_map_shared(). This value is
790 * the mode in which the lock was actually taken.
791 */
792uint
793xfs_ilock_map_shared(
794 xfs_inode_t *ip)
795{
796 uint lock_mode;
797
798 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
799 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
800 lock_mode = XFS_ILOCK_EXCL;
801 } else {
802 lock_mode = XFS_ILOCK_SHARED;
803 }
804
805 xfs_ilock(ip, lock_mode);
806
807 return lock_mode;
808}
809
810/*
811 * This is simply the unlock routine to go with xfs_ilock_map_shared().
812 * All it does is call xfs_iunlock() with the given lock_mode.
813 */
814void
815xfs_iunlock_map_shared(
816 xfs_inode_t *ip,
817 unsigned int lock_mode)
818{
819 xfs_iunlock(ip, lock_mode);
820}
821
822/*
823 * The xfs inode contains 2 locks: a multi-reader lock called the
824 * i_iolock and a multi-reader lock called the i_lock. This routine
825 * allows either or both of the locks to be obtained.
826 *
827 * The 2 locks should always be ordered so that the IO lock is
828 * obtained first in order to prevent deadlock.
829 *
830 * ip -- the inode being locked
831 * lock_flags -- this parameter indicates the inode's locks
832 * to be locked. It can be:
833 * XFS_IOLOCK_SHARED,
834 * XFS_IOLOCK_EXCL,
835 * XFS_ILOCK_SHARED,
836 * XFS_ILOCK_EXCL,
837 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
838 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
839 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
840 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
841 */
842void
843xfs_ilock(xfs_inode_t *ip,
844 uint lock_flags)
845{
846 /*
847 * You can't set both SHARED and EXCL for the same lock,
848 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
849 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
850 */
851 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
852 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
853 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
854 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
855 ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0);
856
857 if (lock_flags & XFS_IOLOCK_EXCL) {
858 mrupdate(&ip->i_iolock);
859 } else if (lock_flags & XFS_IOLOCK_SHARED) {
860 mraccess(&ip->i_iolock);
861 }
862 if (lock_flags & XFS_ILOCK_EXCL) {
863 mrupdate(&ip->i_lock);
864 } else if (lock_flags & XFS_ILOCK_SHARED) {
865 mraccess(&ip->i_lock);
866 }
867 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
868}
869
870/*
871 * This is just like xfs_ilock(), except that the caller
872 * is guaranteed not to sleep. It returns 1 if it gets
873 * the requested locks and 0 otherwise. If the IO lock is
874 * obtained but the inode lock cannot be, then the IO lock
875 * is dropped before returning.
876 *
877 * ip -- the inode being locked
878 * lock_flags -- this parameter indicates the inode's locks to be
879 * to be locked. See the comment for xfs_ilock() for a list
880 * of valid values.
881 *
882 */
883int
884xfs_ilock_nowait(xfs_inode_t *ip,
885 uint lock_flags)
886{
887 int iolocked;
888 int ilocked;
889
890 /*
891 * You can't set both SHARED and EXCL for the same lock,
892 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
893 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
894 */
895 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
896 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
897 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
898 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
899 ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0);
900
901 iolocked = 0;
902 if (lock_flags & XFS_IOLOCK_EXCL) {
903 iolocked = mrtryupdate(&ip->i_iolock);
904 if (!iolocked) {
905 return 0;
906 }
907 } else if (lock_flags & XFS_IOLOCK_SHARED) {
908 iolocked = mrtryaccess(&ip->i_iolock);
909 if (!iolocked) {
910 return 0;
911 }
912 }
913 if (lock_flags & XFS_ILOCK_EXCL) {
914 ilocked = mrtryupdate(&ip->i_lock);
915 if (!ilocked) {
916 if (iolocked) {
917 mrunlock(&ip->i_iolock);
918 }
919 return 0;
920 }
921 } else if (lock_flags & XFS_ILOCK_SHARED) {
922 ilocked = mrtryaccess(&ip->i_lock);
923 if (!ilocked) {
924 if (iolocked) {
925 mrunlock(&ip->i_iolock);
926 }
927 return 0;
928 }
929 }
930 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
931 return 1;
932}
933
934/*
935 * xfs_iunlock() is used to drop the inode locks acquired with
936 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
937 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
938 * that we know which locks to drop.
939 *
940 * ip -- the inode being unlocked
941 * lock_flags -- this parameter indicates the inode's locks to be
942 * to be unlocked. See the comment for xfs_ilock() for a list
943 * of valid values for this parameter.
944 *
945 */
946void
947xfs_iunlock(xfs_inode_t *ip,
948 uint lock_flags)
949{
950 /*
951 * You can't set both SHARED and EXCL for the same lock,
952 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
953 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
954 */
955 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
956 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
957 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
958 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
959 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY)) == 0);
960 ASSERT(lock_flags != 0);
961
962 if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) {
963 ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) ||
964 (ismrlocked(&ip->i_iolock, MR_ACCESS)));
965 ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) ||
966 (ismrlocked(&ip->i_iolock, MR_UPDATE)));
967 mrunlock(&ip->i_iolock);
968 }
969
970 if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) {
971 ASSERT(!(lock_flags & XFS_ILOCK_SHARED) ||
972 (ismrlocked(&ip->i_lock, MR_ACCESS)));
973 ASSERT(!(lock_flags & XFS_ILOCK_EXCL) ||
974 (ismrlocked(&ip->i_lock, MR_UPDATE)));
975 mrunlock(&ip->i_lock);
976
977 /*
978 * Let the AIL know that this item has been unlocked in case
979 * it is in the AIL and anyone is waiting on it. Don't do
980 * this if the caller has asked us not to.
981 */
982 if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) &&
983 ip->i_itemp != NULL) {
984 xfs_trans_unlocked_item(ip->i_mount,
985 (xfs_log_item_t*)(ip->i_itemp));
986 }
987 }
988 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
989}
990
991/*
992 * give up write locks. the i/o lock cannot be held nested
993 * if it is being demoted.
994 */
995void
996xfs_ilock_demote(xfs_inode_t *ip,
997 uint lock_flags)
998{
999 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
1000 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
1001
1002 if (lock_flags & XFS_ILOCK_EXCL) {
1003 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
1004 mrdemote(&ip->i_lock);
1005 }
1006 if (lock_flags & XFS_IOLOCK_EXCL) {
1007 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1008 mrdemote(&ip->i_iolock);
1009 }
1010}
1011
1012/*
1013 * The following three routines simply manage the i_flock
1014 * semaphore embedded in the inode. This semaphore synchronizes
1015 * processes attempting to flush the in-core inode back to disk.
1016 */
1017void
1018xfs_iflock(xfs_inode_t *ip)
1019{
1020 psema(&(ip->i_flock), PINOD|PLTWAIT);
1021}
1022
1023int
1024xfs_iflock_nowait(xfs_inode_t *ip)
1025{
1026 return (cpsema(&(ip->i_flock)));
1027}
1028
1029void
1030xfs_ifunlock(xfs_inode_t *ip)
1031{
Al Viro0d8fee32006-06-19 08:41:30 +10001032 ASSERT(issemalocked(&(ip->i_flock)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 vsema(&(ip->i_flock));
1034}