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 * sending a signal known to be ignored to cause restarting syscall if GNS mount
74 * function returns -ERESTARTSYS.
77 ll_gns_send_signal(void)
79 struct task_struct *task = current;
82 read_lock(&tasklist_lock);
83 spin_lock_irq(&task->sighand->siglock);
84 sigaddset(&task->pending.signal, signal);
85 spin_unlock_irq(&task->sighand->siglock);
86 read_unlock(&tasklist_lock);
87 set_tsk_thread_flag(task, TIF_SIGPENDING);
91 * tries to mount the mount object under passed @dentry. In the case of success
92 * @dentry will become mount point and 0 will be returned. Error code will be
96 ll_gns_mount_object(struct dentry *dentry, struct vfsmount *mnt)
98 char *path, *pathpage, *datapage, *argv[4];
99 struct file *mntinfo_fd = NULL;
100 int cleanup_phase = 0, rc = 0;
101 struct ll_sb_info *sbi;
102 struct dentry *dchild;
105 LASSERT(dentry->d_inode != NULL);
107 if (!S_ISDIR(dentry->d_inode->i_mode))
110 sbi = ll_i2sbi(dentry->d_inode);
111 LASSERT(sbi != NULL);
113 spin_lock(&sbi->ll_gns_lock);
115 if (sbi->ll_gns_state == LL_GNS_DISABLED) {
116 spin_unlock(&sbi->ll_gns_lock);
121 CERROR("suid directory found, but no "
122 "vfsmount available.\n");
127 * another thead is in progress or just finished mounting the
128 * dentry. Handling that.
130 if (sbi->ll_gns_state == LL_GNS_MOUNTING ||
131 sbi->ll_gns_state == LL_GNS_FINISHED) {
133 * another thread is trying to mount GNS dentry. We'd like to
136 spin_unlock(&sbi->ll_gns_lock);
139 * check if dentry is mount point already, if so, do not restart
142 if (d_mountpoint(dentry))
146 * causing syscall to restart and find this dentry already
149 ll_gns_send_signal();
150 RETURN(-ERESTARTSYS);
153 wait_for_completion(&sbi->ll_gns_mount_finished);
154 if (d_mountpoint(dentry))
158 LASSERT(sbi->ll_gns_state == LL_GNS_IDLE);
160 CDEBUG(D_INODE, "mounting dentry %p\n", dentry);
162 /* mounting started */
163 sbi->ll_gns_state = LL_GNS_MOUNTING;
164 spin_unlock(&sbi->ll_gns_lock);
166 /* we need to build an absolute pathname to pass to mount */
167 pathpage = (char *)__get_free_page(GFP_KERNEL);
169 GOTO(cleanup, rc = -ENOMEM);
172 /* getting @dentry path stored in @pathpage. */
173 path = d_path(dentry, mnt, pathpage, PAGE_SIZE);
175 CERROR("can't build mount object path, err %d\n",
176 (int)PTR_ERR(dchild));
177 GOTO(cleanup, rc = PTR_ERR(dchild));
180 /* synchronizing with possible /proc/fs/...write */
181 down(&sbi->ll_gns_sem);
184 * mount object name is taken from sbi, where it is set in mount time or
185 * via /proc/fs... tunable. It may be ".mntinfo" or so.
189 * recursive lookup with trying to mount SUID bit marked directories on
190 * the way is not possible here, as lookup_one_len() does not pass @nd
191 * to ->lookup() and this is checked in ll_lookup_it(). So, do not
192 * handle possible -EAGAIN here.
194 dchild = ll_lookup_one_len(sbi->ll_gns_oname, dentry,
195 strlen(sbi->ll_gns_oname));
196 up(&sbi->ll_gns_sem);
200 if (IS_ERR(dchild)) {
201 rc = PTR_ERR(dchild);
202 CERROR("can't find mount object %*s/%*s err = %d.\n",
203 (int)dentry->d_name.len, dentry->d_name.name,
204 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
209 /* mount object is not found */
210 if (!dchild->d_inode)
211 GOTO(cleanup, rc = -ENOENT);
213 /* check if found child is regular file */
214 if (!S_ISREG(dchild->d_inode->i_mode))
215 GOTO(cleanup, rc = -EOPNOTSUPP);
219 /* ok, mount object if found, opening it. */
220 mntinfo_fd = dentry_open(dchild, mnt, 0);
221 if (IS_ERR(mntinfo_fd)) {
222 CERROR("can't open mount object %*s/%*s err = %d.\n",
223 (int)dentry->d_name.len, dentry->d_name.name,
224 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
225 (int)PTR_ERR(mntinfo_fd));
227 GOTO(cleanup, rc = PTR_ERR(mntinfo_fd));
231 if (mntinfo_fd->f_dentry->d_inode->i_size > PAGE_SIZE - 1) {
232 CERROR("mount object %*s/%*s is too big (%Ld)\n",
233 (int)dentry->d_name.len, dentry->d_name.name,
234 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
235 mntinfo_fd->f_dentry->d_inode->i_size);
236 GOTO(cleanup, rc = -EFBIG);
239 datapage = (char *)__get_free_page(GFP_KERNEL);
241 GOTO(cleanup, rc = -ENOMEM);
245 /* read data from mount object. */
246 rc = kernel_read(mntinfo_fd, 0, datapage, PAGE_SIZE - 1);
248 CERROR("can't read mount object %*s/%*s data, err %d\n",
249 (int)dentry->d_name.len, dentry->d_name.name,
250 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
255 /* no data in mount object? */
257 CERROR("mount object %*s/%*s is empty?\n",
258 (int)dentry->d_name.len, dentry->d_name.name,
259 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname);
268 /* synchronizing with possible /proc/fs/...write */
269 down(&sbi->ll_gns_sem);
272 * upcall is initialized in mount time or via /proc/fs/... tuneable and
273 * may be /usr/lib/lustre/gns-upcall.sh
275 argv[0] = sbi->ll_gns_upcall;
280 up(&sbi->ll_gns_sem);
282 /* do not wait for helper complete here. */
283 rc = call_usermodehelper(argv[0], argv, NULL, 0);
285 CERROR("failed to call GNS upcall %s, err = %d\n",
286 sbi->ll_gns_upcall, rc);
291 * waiting for dentry become mount point GNS_WAIT_ATTEMPTS times by 1
294 rc = ll_gns_wait_for_mount(dentry, 1, GNS_WAIT_ATTEMPTS);
295 complete_all(&sbi->ll_gns_mount_finished);
297 struct dentry *rdentry;
298 struct vfsmount *rmnt;
300 /* mount is successful */
301 LASSERT(sbi->ll_gns_state == LL_GNS_FINISHED);
304 rdentry = dget(dentry);
306 if (follow_down(&rmnt, &rdentry)) {
308 * registering new mount in GNS mounts list and thus
309 * make it accessible from GNS control thread.
311 spin_lock(&dcache_lock);
312 LASSERT(list_empty(&rmnt->mnt_lustre_list));
313 list_add_tail(&rmnt->mnt_lustre_list,
315 spin_unlock(&dcache_lock);
316 rmnt->mnt_last_used = jiffies;
324 CERROR("usermode upcall %s failed to mount %s, err %d\n",
325 sbi->ll_gns_upcall, path, rc);
330 switch (cleanup_phase) {
332 free_page((unsigned long)datapage);
334 if (mntinfo_fd != NULL)
340 free_page((unsigned long)pathpage);
341 complete_all(&sbi->ll_gns_mount_finished);
343 spin_lock(&sbi->ll_gns_lock);
344 sbi->ll_gns_state = LL_GNS_IDLE;
345 spin_unlock(&sbi->ll_gns_lock);
350 /* tries to umount passed @mnt. */
351 int ll_gns_umount_object(struct vfsmount *mnt)
356 CDEBUG(D_INODE, "unmounting mnt %p\n", mnt);
357 rc = do_umount(mnt, 0);
359 CDEBUG(D_INODE, "can't umount 0x%p, err = %d\n",
366 int ll_gns_check_mounts(struct ll_sb_info *sbi, int flags)
368 struct list_head check_list = LIST_HEAD_INIT(check_list);
369 struct vfsmount *mnt;
373 spin_lock(&dcache_lock);
374 list_splice_init(&sbi->ll_mnt_list, &check_list);
377 * walk the list in reverse order, and put them on the front of the sbi
378 * list each iteration; this avoids list-ordering problems if we race
379 * with another gns-mounting thread.
381 while (!list_empty(&check_list)) {
382 mnt = list_entry(check_list.prev,
388 list_del_init(&mnt->mnt_lustre_list);
390 list_add(&mnt->mnt_lustre_list,
393 /* check for timeout if needed */
394 pass = jiffies - mnt->mnt_last_used;
396 if (flags == LL_GNS_CHECK &&
397 pass < sbi->ll_gns_timeout * HZ)
402 spin_unlock(&dcache_lock);
405 ll_gns_umount_object(mnt);
408 spin_lock(&dcache_lock);
410 spin_unlock(&dcache_lock);
415 * GNS timer callback function. It restarts gns timer and wakes up GNS control
416 * thread to process mounts list.
418 void ll_gns_timer_callback(unsigned long data)
420 struct ll_sb_info *sbi = (void *)data;
423 spin_lock(&gns_lock);
424 if (list_empty(&sbi->ll_gns_sbi_head))
425 list_add(&sbi->ll_gns_sbi_head, &gns_sbi_list);
426 spin_unlock(&gns_lock);
428 wake_up(&gns_thread.t_ctl_waitq);
429 mod_timer(&sbi->ll_gns_timer,
430 jiffies + sbi->ll_gns_tick * HZ);
433 /* this function checks if something new happened to exist in gns list. */
434 static int inline ll_gns_check_event(void)
438 spin_lock(&gns_lock);
439 rc = !list_empty(&gns_sbi_list);
440 spin_unlock(&gns_lock);
445 /* should we stop GNS control thread? */
446 static int inline ll_gns_check_stop(void)
449 return (gns_thread.t_flags & SVC_STOPPING) ? 1 : 0;
452 /* GNS control thread function. */
453 static int ll_gns_thread_main(void *arg)
455 struct ll_gns_ctl *ctl = arg;
460 char name[sizeof(current->comm)];
461 snprintf(name, sizeof(name) - 1, "ll_gns");
462 kportal_daemonize(name);
465 SIGNAL_MASK_LOCK(current, flags);
466 sigfillset(¤t->blocked);
468 SIGNAL_MASK_UNLOCK(current, flags);
471 * letting starting function know, that we are ready and control may be
474 gns_thread.t_flags = SVC_RUNNING;
475 complete(&ctl->gc_starting);
477 while (!ll_gns_check_stop()) {
478 struct l_wait_info lwi = { 0 };
480 l_wait_event(gns_thread.t_ctl_waitq,
481 (ll_gns_check_event() ||
482 ll_gns_check_stop()), &lwi);
484 spin_lock(&gns_lock);
485 while (!list_empty(&gns_sbi_list)) {
486 struct ll_sb_info *sbi;
488 sbi = list_entry(gns_sbi_list.prev,
492 list_del_init(&sbi->ll_gns_sbi_head);
493 spin_unlock(&gns_lock);
494 ll_gns_check_mounts(sbi, LL_GNS_CHECK);
495 spin_lock(&gns_lock);
497 spin_unlock(&gns_lock);
501 gns_thread.t_flags = SVC_STOPPED;
503 /* this is SMP-safe way to finish thread. */
504 complete_and_exit(&ctl->gc_finishing, 0);
507 void ll_gns_add_timer(struct ll_sb_info *sbi)
509 mod_timer(&sbi->ll_gns_timer,
510 jiffies + sbi->ll_gns_tick * HZ);
513 void ll_gns_del_timer(struct ll_sb_info *sbi)
515 del_timer(&sbi->ll_gns_timer);
519 * starts GNS control thread and waits for a signal it is up and work may be
522 int ll_gns_start_thread(void)
527 LASSERT(gns_thread.t_flags == 0);
528 init_completion(&gns_ctl.gc_starting);
529 init_completion(&gns_ctl.gc_finishing);
530 init_waitqueue_head(&gns_thread.t_ctl_waitq);
532 rc = kernel_thread(ll_gns_thread_main, &gns_ctl,
533 (CLONE_VM | CLONE_FILES));
535 CERROR("cannot start GNS control thread, "
539 wait_for_completion(&gns_ctl.gc_starting);
540 LASSERT(gns_thread.t_flags == SVC_RUNNING);
544 /* stops GNS control thread and waits its actual stop. */
545 void ll_gns_stop_thread(void)
548 gns_thread.t_flags = SVC_STOPPING;
549 wake_up(&gns_thread.t_ctl_waitq);
550 wait_for_completion(&gns_ctl.gc_finishing);
551 LASSERT(gns_thread.t_flags == SVC_STOPPED);
552 gns_thread.t_flags = 0;