1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2004, 2005 Cluster File Systems, Inc.
6 * Author: Phil Schwan <phil@clusterfs.com>
7 * Author: Oleg Drokin <green@clusterfs.com>
8 * Author: Yury Umanets <yury@clusterfs.com>
9 * Review: Nikita Danilov <nikita@clusterfs.com>
11 * This file is part of Lustre, http://www.lustre.org.
13 * Lustre is free software; you can redistribute it and/or
14 * modify it under the terms of version 2 of the GNU General Public
15 * License as published by the Free Software Foundation.
17 * Lustre is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
22 * You should have received a copy of the GNU General Public License
23 * along with Lustre; if not, write to the Free Software
24 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 #define DEBUG_SUBSYSTEM S_LLITE
30 #include <linux/version.h>
31 #include <asm/uaccess.h>
32 #include <linux/file.h>
33 #include <linux/kmod.h>
35 #include <linux/lustre_lite.h>
36 #include "llite_internal.h"
38 static struct list_head gns_sbi_list = LIST_HEAD_INIT(gns_sbi_list);
39 static spinlock_t gns_lock = SPIN_LOCK_UNLOCKED;
40 static struct ptlrpc_thread gns_thread;
41 static struct ll_gns_ctl gns_ctl;
44 * waits until passed dentry gets mountpoint or timeout and attempts are
45 * exhausted. Returns 1 if dentry became mountpoint and 0 otherwise.
48 ll_gns_wait_for_mount(struct dentry *dentry,
49 int timeout, int tries)
51 struct l_wait_info lwi;
52 struct ll_sb_info *sbi;
55 LASSERT(dentry != NULL);
56 LASSERT(!IS_ERR(dentry));
57 sbi = ll_s2sbi(dentry->d_sb);
59 lwi = LWI_TIMEOUT(timeout * HZ, NULL, NULL);
60 for (; !d_mountpoint(dentry) && tries > 0; tries--)
61 l_wait_event(sbi->ll_gns_waitq, d_mountpoint(dentry), &lwi);
63 if (d_mountpoint(dentry)) {
64 spin_lock(&sbi->ll_gns_lock);
65 sbi->ll_gns_state = LL_GNS_FINISHED;
66 spin_unlock(&sbi->ll_gns_lock);
73 * tries to mount the mount object under passed @dentry. In the case of success
74 * @dentry will become mount point and 0 will be returned. Error code will be
78 ll_gns_mount_object(struct dentry *dentry, struct vfsmount *mnt)
80 struct ll_dentry_data *lld = dentry->d_fsdata;
81 char *path, *pathpage, *datapage, *argv[4];
82 struct file *mntinfo_fd = NULL;
83 int cleanup_phase = 0, rc = 0;
84 struct ll_sb_info *sbi;
85 struct dentry *dchild;
89 CERROR("suid directory found, but no "
90 "vfsmount available.\n");
94 CDEBUG(D_INODE, "mounting dentry %p\n", dentry);
96 LASSERT(dentry->d_inode != NULL);
97 LASSERT(S_ISDIR(dentry->d_inode->i_mode));
100 sbi = ll_i2sbi(dentry->d_inode);
101 LASSERT(sbi != NULL);
104 * another thead is in progress or just finished mounting the
105 * dentry. Handling that.
107 spin_lock(&sbi->ll_gns_lock);
108 if (sbi->ll_gns_state == LL_GNS_MOUNTING ||
109 sbi->ll_gns_state == LL_GNS_FINISHED) {
110 spin_unlock(&sbi->ll_gns_lock);
111 CDEBUG(D_INODE, "GNS is in progress now, throwing "
112 "-ERESTARTSYS to restart syscall and let "
114 RETURN(-ERESTARTSYS);
116 LASSERT(sbi->ll_gns_state == LL_GNS_IDLE);
118 /* mounting started */
119 sbi->ll_gns_state = LL_GNS_MOUNTING;
120 spin_unlock(&sbi->ll_gns_lock);
122 /* we need to build an absolute pathname to pass to mount */
123 pathpage = (char *)__get_free_page(GFP_KERNEL);
125 GOTO(cleanup, rc = -ENOMEM);
128 /* getting @dentry path stored in @pathpage. */
129 path = d_path(dentry, mnt, pathpage, PAGE_SIZE);
131 CERROR("can't build mount object path, err %d\n",
132 (int)PTR_ERR(dchild));
133 GOTO(cleanup, rc = PTR_ERR(dchild));
136 /* synchronizing with possible /proc/fs/...write */
137 down(&sbi->ll_gns_sem);
140 * mount object name is taken from sbi, where it is set in mount time or
141 * via /proc/fs... tunable. It may be ".mntinfo" or so.
143 dchild = lookup_one_len(sbi->ll_gns_oname, dentry,
144 strlen(sbi->ll_gns_oname));
145 up(&sbi->ll_gns_sem);
149 if (IS_ERR(dchild)) {
150 rc = PTR_ERR(dchild);
152 if (rc == -ERESTARTSYS) {
153 CDEBUG(D_INODE, "possible endless loop is detected "
154 "due to mount object is directory marked by "
156 GOTO(cleanup, rc = -ELOOP);
159 CERROR("can't find mount object %*s/%*s err = %d.\n",
160 (int)dentry->d_name.len, dentry->d_name.name,
161 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
166 /* mount object is not found */
167 if (!dchild->d_inode)
168 GOTO(cleanup, rc = -ENOENT);
170 /* check if found child is regular file */
171 if (!S_ISREG(dchild->d_inode->i_mode))
172 GOTO(cleanup, rc = -EOPNOTSUPP);
176 /* ok, mount object if found, opening it. */
177 mntinfo_fd = dentry_open(dchild, mnt, 0);
178 if (IS_ERR(mntinfo_fd)) {
179 CERROR("can't open mount object %*s/%*s err = %d.\n",
180 (int)dentry->d_name.len, dentry->d_name.name,
181 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
182 (int)PTR_ERR(mntinfo_fd));
184 GOTO(cleanup, rc = PTR_ERR(mntinfo_fd));
188 if (mntinfo_fd->f_dentry->d_inode->i_size > PAGE_SIZE - 1) {
189 CERROR("mount object %*s/%*s is too big (%Ld)\n",
190 (int)dentry->d_name.len, dentry->d_name.name,
191 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
192 mntinfo_fd->f_dentry->d_inode->i_size);
193 GOTO(cleanup, rc = -EFBIG);
196 datapage = (char *)__get_free_page(GFP_KERNEL);
198 GOTO(cleanup, rc = -ENOMEM);
202 /* read data from mount object. */
203 rc = kernel_read(mntinfo_fd, 0, datapage, PAGE_SIZE);
205 CERROR("can't read mount object %*s/%*s data, err %d\n",
206 (int)dentry->d_name.len, dentry->d_name.name,
207 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
217 /* synchronizing with possible /proc/fs/...write */
218 down(&sbi->ll_gns_sem);
221 * upcall is initialized in mount time or via /proc/fs/... tuneable and
222 * may be /usr/lib/lustre/gns-upcall.sh
224 argv[0] = sbi->ll_gns_upcall;
229 up(&sbi->ll_gns_sem);
231 rc = USERMODEHELPER(argv[0], argv, NULL);
233 CERROR("failed to call GNS upcall %s, err = %d\n",
234 sbi->ll_gns_upcall, rc);
239 * wait for mount completion. This is actually not need, because
240 * USERMODEHELPER() returns only when usermode process finishes. But we
241 * doing this just for case USERMODEHELPER() semantics will be changed
242 * or usermode upcall program will start mounting in backgound and
243 * return instantly. --umka
245 rc = ll_gns_wait_for_mount(dentry, 1, GNS_WAIT_ATTEMPTS);
246 complete_all(&sbi->ll_gns_mount_finished);
248 struct dentry *rdentry;
249 struct vfsmount *rmnt;
251 /* mount is successful */
252 LASSERT(sbi->ll_gns_state == LL_GNS_FINISHED);
255 rdentry = dget(dentry);
257 if (follow_down(&rmnt, &rdentry)) {
259 * registering new mount in GNS mounts list and thus
260 * make it accessible from GNS control thread.
262 spin_lock(&dcache_lock);
263 LASSERT(list_empty(&rmnt->mnt_lustre_list));
264 list_add_tail(&rmnt->mnt_lustre_list,
266 spin_unlock(&dcache_lock);
267 rmnt->mnt_last_used = jiffies;
275 CERROR("usermode upcall %s failed to mount %s, err %d\n",
276 sbi->ll_gns_upcall, path, rc);
281 switch (cleanup_phase) {
283 free_page((unsigned long)datapage);
285 if (mntinfo_fd != NULL)
291 free_page((unsigned long)pathpage);
294 * waking up all waiters after gns state is set to
297 complete_all(&sbi->ll_gns_mount_finished);
299 spin_lock(&sbi->ll_gns_lock);
300 sbi->ll_gns_state = LL_GNS_IDLE;
301 spin_unlock(&sbi->ll_gns_lock);
306 /* tries to umount passed @mnt. */
307 int ll_gns_umount_object(struct vfsmount *mnt)
312 CDEBUG(D_INODE, "unmounting mnt %p\n", mnt);
313 rc = do_umount(mnt, 0);
315 CDEBUG(D_INODE, "can't umount 0x%p, err = %d\n",
322 int ll_gns_check_mounts(struct ll_sb_info *sbi, int flags)
324 struct list_head check_list = LIST_HEAD_INIT(check_list);
325 struct vfsmount *mnt;
329 spin_lock(&dcache_lock);
330 list_splice_init(&sbi->ll_mnt_list, &check_list);
333 * walk the list in reverse order, and put them on the front of the sbi
334 * list each iteration; this avoids list-ordering problems if we race
335 * with another gns-mounting thread.
337 while (!list_empty(&check_list)) {
338 mnt = list_entry(check_list.prev,
344 list_del_init(&mnt->mnt_lustre_list);
346 list_add(&mnt->mnt_lustre_list,
349 /* check for timeout if needed */
350 pass = jiffies - mnt->mnt_last_used;
352 if (flags == LL_GNS_CHECK &&
353 pass < sbi->ll_gns_timeout * HZ)
358 spin_unlock(&dcache_lock);
361 ll_gns_umount_object(mnt);
364 spin_lock(&dcache_lock);
366 spin_unlock(&dcache_lock);
371 * GNS timer callback function. It restarts gns timer and wakes up GNS control
372 * thread to process mounts list.
374 void ll_gns_timer_callback(unsigned long data)
376 struct ll_sb_info *sbi = (void *)data;
379 spin_lock(&gns_lock);
380 if (list_empty(&sbi->ll_gns_sbi_head))
381 list_add(&sbi->ll_gns_sbi_head, &gns_sbi_list);
382 spin_unlock(&gns_lock);
384 wake_up(&gns_thread.t_ctl_waitq);
385 mod_timer(&sbi->ll_gns_timer,
386 jiffies + sbi->ll_gns_tick * HZ);
389 /* this function checks if something new happened to exist in gns list. */
390 static int inline ll_gns_check_event(void)
394 spin_lock(&gns_lock);
395 rc = !list_empty(&gns_sbi_list);
396 spin_unlock(&gns_lock);
401 /* should we stop GNS control thread? */
402 static int inline ll_gns_check_stop(void)
405 return (gns_thread.t_flags & SVC_STOPPING) ? 1 : 0;
408 /* GNS control thread function. */
409 static int ll_gns_thread_main(void *arg)
411 struct ll_gns_ctl *ctl = arg;
416 char name[sizeof(current->comm)];
417 snprintf(name, sizeof(name) - 1, "ll_gns");
418 kportal_daemonize(name);
421 SIGNAL_MASK_LOCK(current, flags);
422 sigfillset(¤t->blocked);
424 SIGNAL_MASK_UNLOCK(current, flags);
427 * letting starting function know, that we are ready and control may be
430 gns_thread.t_flags = SVC_RUNNING;
431 complete(&ctl->gc_starting);
433 while (!ll_gns_check_stop()) {
434 struct l_wait_info lwi = { 0 };
436 l_wait_event(gns_thread.t_ctl_waitq,
437 (ll_gns_check_event() ||
438 ll_gns_check_stop()), &lwi);
440 spin_lock(&gns_lock);
441 while (!list_empty(&gns_sbi_list)) {
442 struct ll_sb_info *sbi;
444 sbi = list_entry(gns_sbi_list.prev,
448 list_del_init(&sbi->ll_gns_sbi_head);
449 spin_unlock(&gns_lock);
450 ll_gns_check_mounts(sbi, LL_GNS_CHECK);
451 spin_lock(&gns_lock);
453 spin_unlock(&gns_lock);
457 gns_thread.t_flags = SVC_STOPPED;
459 /* this is SMP-safe way to finish thread. */
460 complete_and_exit(&ctl->gc_finishing, 0);
463 void ll_gns_add_timer(struct ll_sb_info *sbi)
465 mod_timer(&sbi->ll_gns_timer,
466 jiffies + sbi->ll_gns_tick * HZ);
469 void ll_gns_del_timer(struct ll_sb_info *sbi)
471 del_timer(&sbi->ll_gns_timer);
475 * starts GNS control thread and waits for a signal it is up and work may be
478 int ll_gns_start_thread(void)
483 LASSERT(gns_thread.t_flags == 0);
484 init_completion(&gns_ctl.gc_starting);
485 init_completion(&gns_ctl.gc_finishing);
486 init_waitqueue_head(&gns_thread.t_ctl_waitq);
488 rc = kernel_thread(ll_gns_thread_main, &gns_ctl,
489 (CLONE_VM | CLONE_FILES));
491 CERROR("cannot start GNS control thread, "
495 wait_for_completion(&gns_ctl.gc_starting);
496 LASSERT(gns_thread.t_flags == SVC_RUNNING);
500 /* stops GNS control thread and waits its actual stop. */
501 void ll_gns_stop_thread(void)
504 gns_thread.t_flags = SVC_STOPPING;
505 wake_up(&gns_thread.t_ctl_waitq);
506 wait_for_completion(&gns_ctl.gc_finishing);
507 LASSERT(gns_thread.t_flags == SVC_STOPPED);
508 gns_thread.t_flags = 0;