1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2001-2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include <linux/sched.h>
24 #include <linux/smp_lock.h>
25 #include <linux/quotaops.h>
27 #define DEBUG_SUBSYSTEM S_LLITE
29 #include <linux/obd_support.h>
30 #include <linux/lustre_lite.h>
31 #include <linux/lustre_idl.h>
32 #include <linux/lustre_dlm.h>
33 #include <linux/lustre_version.h>
35 #include "llite_internal.h"
37 /* should NOT be called with the dcache lock, see fs/dcache.c */
38 static void ll_release(struct dentry *de)
40 struct ll_dentry_data *lld;
44 CDEBUG(D_DENTRY, "releasing dentry %p\n", de);
47 if (lld) { /* Root dentry does not have ll_dentry_data */
48 LASSERT(lld->lld_cwd_count == 0);
49 LASSERT(lld->lld_mnt_count == 0);
50 OBD_FREE(de->d_fsdata, sizeof(struct ll_dentry_data));
56 /* Compare if two dentries are the same. Don't match if the existing dentry
57 * is marked DCACHE_LUSTRE_INVALID. Returns 1 if different, 0 if the same.
59 * This avoids a race where ll_lookup_it() instantiates a dentry, but we get
60 * an AST before calling d_revalidate_it(). The dentry still exists (marked
61 * INVALID) so d_lookup() matches it, but we have no lock on it (so
62 * lock_match() fails) and we spin around real_lookup(). */
63 static int ll_dcompare(struct dentry *parent, struct qstr *d_name,
65 struct dentry *dchild;
68 if (d_name->len != name->len)
71 if (memcmp(d_name->name, name->name, name->len))
74 dchild = container_of(d_name, struct dentry, d_name); /* ugh */
75 if (dchild->d_flags & DCACHE_LUSTRE_INVALID) {
76 CDEBUG(D_DENTRY,"INVALID dentry %p not matched, was bug 3784\n",
84 /* should NOT be called with the dcache lock, see fs/dcache.c */
85 static int ll_ddelete(struct dentry *de)
89 CDEBUG(D_DENTRY, "%s dentry %*s (%p, parent %p, inode %p) %s%s\n",
90 (de->d_flags & DCACHE_LUSTRE_INVALID ? "deleting" : "keeping"),
91 de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode,
92 d_unhashed(de) ? "" : "hashed,",
93 list_empty(&de->d_subdirs) ? "" : "subdirs");
97 void ll_set_dd(struct dentry *de)
102 CDEBUG(D_DENTRY, "ldd on dentry %.*s (%p) parent %p inode %p refc %d\n",
103 de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode,
104 atomic_read(&de->d_count));
106 if (de->d_fsdata == NULL) {
107 OBD_ALLOC(de->d_fsdata, sizeof(struct ll_dentry_data));
114 void ll_intent_drop_lock(struct lookup_intent *it)
116 struct lustre_handle *handle;
117 struct lustre_intent_data *itdata = LUSTRE_IT(it);
119 if (it->it_op && itdata && itdata->it_lock_mode) {
120 handle = (struct lustre_handle *)&itdata->it_lock_handle;
121 CDEBUG(D_DLMTRACE, "releasing lock with cookie "LPX64
122 " from it %p\n", handle->cookie, it);
123 ldlm_lock_decref(handle, itdata->it_lock_mode);
125 /* bug 494: intent_release may be called multiple times, from
126 * this thread and we don't want to double-decref this lock */
127 itdata->it_lock_mode = 0;
131 void ll_intent_release(struct lookup_intent *it)
135 ll_intent_drop_lock(it);
137 it->it_op_release = 0;
142 void ll_unhash_aliases(struct inode *inode)
144 struct list_head *tmp, *head;
145 struct ll_sb_info *sbi;
149 CERROR("unexpected NULL inode, tell phil\n");
154 CDEBUG(D_INODE, "marking dentries for ino %lu/%u(%p) invalid\n",
155 inode->i_ino, inode->i_generation, inode);
157 sbi = ll_i2sbi(inode);
158 head = &inode->i_dentry;
160 spin_lock(&dcache_lock);
162 while ((tmp = tmp->next) != head) {
163 struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
164 if (atomic_read(&dentry->d_count) == 0) {
165 CDEBUG(D_DENTRY, "deleting dentry %.*s (%p) parent %p "
166 "inode %p\n", dentry->d_name.len,
167 dentry->d_name.name, dentry, dentry->d_parent,
171 spin_unlock(&dcache_lock);
174 } else if (!(dentry->d_flags & DCACHE_LUSTRE_INVALID)) {
175 CDEBUG(D_DENTRY, "unhashing dentry %.*s (%p) parent %p "
176 "inode %p refc %d\n", dentry->d_name.len,
177 dentry->d_name.name, dentry, dentry->d_parent,
178 dentry->d_inode, atomic_read(&dentry->d_count));
179 hlist_del_init(&dentry->d_hash);
180 dentry->d_flags |= DCACHE_LUSTRE_INVALID;
181 hlist_add_head(&dentry->d_hash,
182 &sbi->ll_orphan_dentry_list);
185 spin_unlock(&dcache_lock);
189 extern struct dentry *ll_find_alias(struct inode *, struct dentry *);
191 int revalidate_it_finish(struct ptlrpc_request *request, int offset,
192 struct lookup_intent *it, struct dentry *de)
194 struct ll_sb_info *sbi;
201 if (it_disposition(it, DISP_LOOKUP_NEG))
204 sbi = ll_i2sbi(de->d_inode);
205 rc = ll_prep_inode(sbi->ll_dt_exp, sbi->ll_md_exp,
206 &de->d_inode, request, offset, NULL);
211 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry)
214 LASSERT(dentry != NULL);
216 if (LUSTRE_IT(it)->it_lock_mode && dentry->d_inode != NULL) {
217 struct inode *inode = dentry->d_inode;
218 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
219 inode, inode->i_ino, inode->i_generation);
220 mdc_set_lock_data(NULL, &LUSTRE_IT(it)->it_lock_handle, inode);
223 /* drop lookup or getattr locks immediately */
224 if (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR ||
225 it->it_op == IT_CHDIR) {
226 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
228 * on 2.6 there are situations when several lookups and
229 * revalidations may be requested during single operation.
230 * Therefore, we don't release intent here -bzzz
232 ll_intent_drop_lock(it);
234 ll_intent_release(it);
239 void ll_frob_intent(struct lookup_intent **itp, struct lookup_intent *deft)
241 struct lookup_intent *it = *itp;
243 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
245 LASSERTF(it->it_magic == INTENT_MAGIC, "bad intent magic: %x\n",
250 if (!it || it->it_op == IT_GETXATTR)
256 if (ll_intent_alloc(it)) {
257 CERROR("Failed to allocate memory for lustre specific intent "
259 /* XXX: we cannot return status just yet */
264 int ll_intent_alloc(struct lookup_intent *it)
267 CERROR("Intent alloc on already allocated intent\n");
270 OBD_SLAB_ALLOC(it->d.fs_data, ll_intent_slab, SLAB_KERNEL,
271 sizeof(struct lustre_intent_data));
272 if (!it->d.fs_data) {
273 CERROR("Failed to allocate memory for lustre specific intent "
278 it->it_op_release = ll_intent_release;
282 void ll_intent_free(struct lookup_intent *it)
285 OBD_SLAB_FREE(it->d.fs_data, ll_intent_slab,
286 sizeof(struct lustre_intent_data));
287 it->d.fs_data = NULL;
292 ll_special_name(struct dentry *de)
294 if (de->d_name.name[0] == '.') switch (de->d_name.len) {
296 if (de->d_name.name[1] == '.')
306 int ll_revalidate_it(struct dentry *de, int flags, struct nameidata *nd,
307 struct lookup_intent *it)
309 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
310 struct ptlrpc_request *req = NULL;
311 struct obd_export *exp;
312 struct it_cb_data icbd;
313 struct lustre_id pid;
314 struct lustre_id cid;
318 CDEBUG(D_VFSTRACE, "VFS Op:name=%s (%p), intent=%s\n", de->d_name.name,
321 /* Cached negative dentries are unsafe for now - look them up again */
322 if (de->d_inode == NULL)
325 /* Root of the tree is always valid, attributes would be fixed in
326 ll_inode_revalidate_it */
327 if (de->d_sb->s_root == de)
330 CDEBUG(D_INODE, "revalidate 0x%p: %*s -> %lu/%lu\n",
331 de, de->d_name.len, de->d_name.name,
332 (unsigned long) de->d_inode->i_ino,
333 (unsigned long) de->d_inode->i_generation);
335 exp = ll_i2mdexp(de->d_inode);
336 ll_inode2id(&pid, de->d_parent->d_inode);
337 ll_inode2id(&cid, de->d_inode);
338 LASSERT(id_fid(&cid) != 0);
340 icbd.icbd_parent = de->d_parent->d_inode;
341 icbd.icbd_childp = &de;
344 * never execute intents for mount points. Attributes will be fixed up
345 * in ll_inode_revalidate_it().
347 if (d_mountpoint(de))
351 nd->mnt->mnt_last_used = jiffies;
353 OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_REVALIDATE_PAUSE, 5);
355 gns_it = it ? it->it_op : IT_OPEN;
356 ll_frob_intent(&it, &lookup_it);
360 if (it->it_op == IT_GETATTR) { /* We need to check for LOOKUP lock as
362 rc = ll_intent_alloc(&lookup_it);
364 LBUG(); /* Can't think of better idea just yet */
366 rc = md_intent_lock(exp, &pid, de->d_name.name,
367 de->d_name.len, NULL, 0, &cid, &lookup_it,
368 flags, &req, ll_mdc_blocking_ast);
369 /* If there was no lookup lock, no point in even checking for
379 if (it_disposition(&lookup_it, DISP_LOOKUP_NEG)) {
386 ptlrpc_req_finished(req);
388 ll_lookup_finish_locks(&lookup_it, de);
389 /* XXX: on 2.6 ll_lookup_finish_locks does not call ll_intent_release */
390 ll_intent_release(&lookup_it);
393 /* open lock stuff */
394 if ((it->it_op == IT_OPEN) && de->d_inode) {
395 struct inode *inode = de->d_inode;
396 struct ll_inode_info *lli = ll_i2info(inode);
397 struct obd_client_handle **och_p;
399 struct obd_device *obddev;
400 struct lustre_handle lockh;
401 int flags = LDLM_FL_BLOCK_GRANTED;
402 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_OPEN}};
403 struct ldlm_res_id file_res_id = {.name = {id_fid(&lli->lli_id),
404 id_group(&lli->lli_id)}};
407 if (it->it_flags & FMODE_WRITE) {
408 och_p = &lli->lli_mds_write_och;
409 och_usecount = &lli->lli_open_fd_write_count;
411 } else if (it->it_flags & FMODE_EXEC) {
412 och_p = &lli->lli_mds_exec_och;
413 och_usecount = &lli->lli_open_fd_exec_count;
416 och_p = &lli->lli_mds_read_och;
417 och_usecount = &lli->lli_open_fd_read_count;
421 /* Check for the proper lock */
422 obddev = md_get_real_obd(exp, &lli->lli_id);
423 if (!ldlm_lock_match(obddev->obd_namespace, flags, &file_res_id,
424 LDLM_IBITS, &policy, lockmode, &lockh))
426 down(&lli->lli_och_sem);
427 if (*och_p) { /* Everything is open already, do nothing */
428 /*(*och_usecount)++; Do not let them steal our open
429 handle from under us */
430 /* XXX The code above was my original idea, but in case
431 we have the handle, but we cannot use it due to later
432 checks (e.g. O_CREAT|O_EXCL flags set), nobody
433 would decrement counter increased here. So we just
434 hope the lock won't be invalidated in between. But
435 if it would be, we'll reopen the open request to
436 MDS later during file open path */
437 up(&lli->lli_och_sem);
438 memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh,
440 LUSTRE_IT(it)->it_lock_mode = lockmode;
443 /* Hm, interesting. Lock is present, but no open
445 up(&lli->lli_och_sem);
446 ldlm_lock_decref(&lockh, lockmode);
451 rc = md_intent_lock(exp, &pid, de->d_name.name, de->d_name.len,
452 NULL, 0, &cid, it, flags, &req, ll_mdc_blocking_ast);
453 /* If req is NULL, then md_intent_lock() only tried to do a lock match;
454 * if all was well, it will return 1 if it found locks, 0 otherwise. */
455 if (req == NULL && rc >= 0) {
463 CDEBUG(D_INFO, "ll_intent_lock(): rc %d : it->it_status "
464 "%d\n", rc, LUSTRE_IT(it)->it_status);
469 rc = revalidate_it_finish(req, 1, it, de);
471 ll_intent_release(it);
476 /* unfortunately ll_intent_lock may cause a callback and revoke our
478 spin_lock(&dcache_lock);
479 hlist_del_init(&de->d_hash);
481 spin_unlock(&dcache_lock);
485 /* If we had succesful it lookup on mds, but it happened to be negative,
486 we do not free request as it will be reused during lookup (see
487 comment in mdc/mdc_locks.c::mdc_intent_lock(). But if
488 request was not completed, we need to free it. (bug 5154) */
489 if (req != NULL && (rc == 1 || !it_disposition(it, DISP_ENQ_COMPLETE))) {
490 ptlrpc_req_finished(req);
495 if (it == &lookup_it)
496 ll_intent_release(it);
498 ll_unhash_aliases(de->d_inode);
503 * if we found that this is possible GNS mount and dentry is still valid
504 * and may be used by system, we drop the lock and return 0, that means
505 * that re-lookup is needed. Such a way we cause real mounting only in
506 * lookup control path, which is always made with parent's i_sem taken.
510 !(!((de->d_inode->i_mode & S_ISUID) && S_ISDIR(de->d_inode->i_mode)) ||
511 !(flags & LOOKUP_CONTINUE || (gns_it & (IT_CHDIR | IT_OPEN))))) {
513 * special "." and ".." has to be always revalidated because
514 * they never should be passed to lookup()
516 if (!ll_special_name(de)) {
517 ll_intent_drop_lock(it);
519 ll_unhash_aliases(de->d_inode);
524 CDEBUG(D_DENTRY, "revalidated dentry %*s (%p) parent %p "
525 "inode %p refc %d\n", de->d_name.len,
526 de->d_name.name, de, de->d_parent, de->d_inode,
527 atomic_read(&de->d_count));
529 ll_lookup_finish_locks(it, de);
530 de->d_flags &= ~DCACHE_LUSTRE_INVALID;
531 if (it == &lookup_it)
532 ll_intent_release(it);
536 if (ll_intent_alloc(it))
539 // We did that already, right? ll_inode2id(&pid, de->d_parent->d_inode);
540 rc = md_intent_lock(exp, &pid, de->d_name.name,
541 de->d_name.len, NULL, 0, NULL,
542 it, 0, &req, ll_mdc_blocking_ast);
544 struct mds_body *mds_body = lustre_msg_buf(req->rq_repmsg, 1,
547 /* See if we got same inode, if not - return error */
548 if (id_equal_stc(&cid, &mds_body->id1))
549 goto revalidate_finish;
555 /*static*/ void ll_pin(struct dentry *de, struct vfsmount *mnt, int flag)
557 struct inode *inode= de->d_inode;
558 struct ll_sb_info *sbi = ll_i2sbi(inode);
559 struct ll_dentry_data *ldd = ll_d2d(de);
560 struct obd_client_handle *handle;
566 /* Strictly speaking this introduces an additional race: the
567 * increments should wait until the rpc has returned.
568 * However, given that at present the function is void, this
570 if (flag == 1 && (++ldd->lld_mnt_count) > 1) {
576 if (flag == 0 && (++ldd->lld_cwd_count) > 1) {
583 handle = (flag) ? &ldd->lld_mnt_och : &ldd->lld_cwd_och;
584 rc = obd_pin(sbi->ll_md_exp, inode->i_ino, inode->i_generation,
585 inode->i_mode & S_IFMT, handle, flag);
589 memset(handle, 0, sizeof(*handle));
591 ldd->lld_cwd_count--;
593 ldd->lld_mnt_count--;
601 /*static*/ void ll_unpin(struct dentry *de, struct vfsmount *mnt, int flag)
603 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
604 struct ll_dentry_data *ldd = ll_d2d(de);
605 struct obd_client_handle handle;
611 /* Strictly speaking this introduces an additional race: the
612 * increments should wait until the rpc has returned.
613 * However, given that at present the function is void, this
615 handle = (flag) ? ldd->lld_mnt_och : ldd->lld_cwd_och;
616 if (handle.och_magic != OBD_CLIENT_HANDLE_MAGIC) {
617 /* the "pin" failed */
624 count = --ldd->lld_mnt_count;
626 count = --ldd->lld_cwd_count;
634 rc = obd_unpin(sbi->ll_md_exp, &handle, flag);
638 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
639 static int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd)
644 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
645 rc = ll_revalidate_it(dentry, nd->flags, nd, &nd->intent.open);
647 rc = ll_revalidate_it(dentry, 0, nd, NULL);
653 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
654 static void ll_dentry_iput(struct dentry *dentry, struct inode *inode)
656 struct ll_sb_info *sbi = ll_i2sbi(inode);
657 struct lustre_id parent, child;
659 LASSERT(dentry->d_parent && dentry->d_parent->d_inode);
660 ll_inode2id(&parent, dentry->d_parent->d_inode);
661 ll_inode2id(&child, inode);
662 md_change_cbdata_name(sbi->ll_md_exp, &parent,
663 (char *)dentry->d_name.name,
664 dentry->d_name.len, &child,
665 null_if_equal, inode);
669 static void ll_dentry_iput(struct dentry *dentry, struct inode *inode)
671 struct ll_sb_info *sbi = ll_i2sbi(inode);
672 struct lustre_id parent, child;
674 if (dentry->d_parent != dentry) {
675 /* Do not do this for root of the tree */
676 LASSERT(dentry->d_parent && dentry->d_parent->d_inode);
677 ll_inode2id(&parent, dentry->d_parent->d_inode);
678 ll_inode2id(&child, inode);
679 md_change_cbdata_name(sbi->ll_md_exp, &parent,
680 (char *)dentry->d_name.name,
681 dentry->d_name.len, &child,
682 null_if_equal, inode);
689 struct dentry_operations ll_d_ops = {
690 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
691 .d_revalidate = ll_revalidate_nd,
693 .d_revalidate_it = ll_revalidate_it,
695 .d_release = ll_release,
696 .d_iput = ll_dentry_iput,
697 .d_delete = ll_ddelete,
698 .d_compare = ll_dcompare,