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 char *path, *pathpage, *datapage, *argv[4];
81 struct file *mntinfo_fd = NULL;
82 int cleanup_phase = 0, rc = 0;
83 struct ll_sb_info *sbi;
84 struct dentry *dchild;
87 LASSERT(dentry->d_inode != NULL);
89 if (!S_ISDIR(dentry->d_inode->i_mode))
92 sbi = ll_i2sbi(dentry->d_inode);
95 spin_lock(&sbi->ll_gns_lock);
97 if (sbi->ll_gns_state == LL_GNS_DISABLED) {
98 spin_unlock(&sbi->ll_gns_lock);
103 * another thead is in progress or just finished mounting the
104 * dentry. Handling that.
106 if (sbi->ll_gns_state == LL_GNS_MOUNTING ||
107 sbi->ll_gns_state == LL_GNS_FINISHED) {
109 * check if another thread is trying to mount some GNS dentry
110 * too. Letting it know that we busy and make ll_lookup_it() to
111 * restart syscall and try again later.
113 spin_unlock(&sbi->ll_gns_lock);
116 LASSERT(sbi->ll_gns_state == LL_GNS_IDLE);
119 CERROR("suid directory found, but no "
120 "vfsmount available.\n");
124 CDEBUG(D_INODE, "mounting dentry %p\n", dentry);
126 /* mounting started */
127 sbi->ll_gns_state = LL_GNS_MOUNTING;
128 spin_unlock(&sbi->ll_gns_lock);
130 /* we need to build an absolute pathname to pass to mount */
131 pathpage = (char *)__get_free_page(GFP_KERNEL);
133 GOTO(cleanup, rc = -ENOMEM);
136 /* getting @dentry path stored in @pathpage. */
137 path = d_path(dentry, mnt, pathpage, PAGE_SIZE);
139 CERROR("can't build mount object path, err %d\n",
140 (int)PTR_ERR(dchild));
141 GOTO(cleanup, rc = PTR_ERR(dchild));
144 /* synchronizing with possible /proc/fs/...write */
145 down(&sbi->ll_gns_sem);
148 * mount object name is taken from sbi, where it is set in mount time or
149 * via /proc/fs... tunable. It may be ".mntinfo" or so.
153 * recursive lookup with trying to mount SUID bit marked directories on
154 * the way is not possible here, as lookup_one_len() does not pass @nd
155 * to ->lookup() and this is checked in ll_lookup_it(). So, do not
156 * handle possible -EAGAIN here.
158 dchild = ll_lookup_one_len(sbi->ll_gns_oname, dentry,
159 strlen(sbi->ll_gns_oname));
160 up(&sbi->ll_gns_sem);
164 if (IS_ERR(dchild)) {
165 rc = PTR_ERR(dchild);
166 CERROR("can't find mount object %*s/%*s err = %d.\n",
167 (int)dentry->d_name.len, dentry->d_name.name,
168 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
173 /* mount object is not found */
174 if (!dchild->d_inode)
175 GOTO(cleanup, rc = -ENOENT);
177 /* check if found child is regular file */
178 if (!S_ISREG(dchild->d_inode->i_mode))
179 GOTO(cleanup, rc = -EOPNOTSUPP);
183 /* ok, mount object if found, opening it. */
184 mntinfo_fd = dentry_open(dchild, mnt, 0);
185 if (IS_ERR(mntinfo_fd)) {
186 CERROR("can't open mount object %*s/%*s err = %d.\n",
187 (int)dentry->d_name.len, dentry->d_name.name,
188 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
189 (int)PTR_ERR(mntinfo_fd));
191 GOTO(cleanup, rc = PTR_ERR(mntinfo_fd));
195 if (mntinfo_fd->f_dentry->d_inode->i_size > PAGE_SIZE - 1) {
196 CERROR("mount object %*s/%*s is too big (%Ld)\n",
197 (int)dentry->d_name.len, dentry->d_name.name,
198 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
199 mntinfo_fd->f_dentry->d_inode->i_size);
200 GOTO(cleanup, rc = -EFBIG);
203 datapage = (char *)__get_free_page(GFP_KERNEL);
205 GOTO(cleanup, rc = -ENOMEM);
209 /* read data from mount object. */
210 rc = kernel_read(mntinfo_fd, 0, datapage, PAGE_SIZE - 1);
212 CERROR("can't read mount object %*s/%*s data, err %d\n",
213 (int)dentry->d_name.len, dentry->d_name.name,
214 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname,
219 /* no data in mount object? */
221 CERROR("mount object %*s/%*s is empty?\n",
222 (int)dentry->d_name.len, dentry->d_name.name,
223 strlen(sbi->ll_gns_oname), sbi->ll_gns_oname);
232 /* synchronizing with possible /proc/fs/...write */
233 down(&sbi->ll_gns_sem);
236 * upcall is initialized in mount time or via /proc/fs/... tuneable and
237 * may be /usr/lib/lustre/gns-upcall.sh
239 argv[0] = sbi->ll_gns_upcall;
244 up(&sbi->ll_gns_sem);
246 rc = USERMODEHELPER(argv[0], argv, NULL);
248 CERROR("failed to call GNS upcall %s, err = %d\n",
249 sbi->ll_gns_upcall, rc);
254 * wait for mount completion. This is actually not needed, because
255 * USERMODEHELPER() returns only when usermode process finishes. But we
256 * doing this just for case USERMODEHELPER() semantics will be changed
257 * or usermode upcall program will start mounting in backgound and
258 * return instantly. --umka
260 rc = ll_gns_wait_for_mount(dentry, 1, GNS_WAIT_ATTEMPTS);
262 struct dentry *rdentry;
263 struct vfsmount *rmnt;
265 /* mount is successful */
266 LASSERT(sbi->ll_gns_state == LL_GNS_FINISHED);
269 rdentry = dget(dentry);
271 if (follow_down(&rmnt, &rdentry)) {
273 * registering new mount in GNS mounts list and thus
274 * make it accessible from GNS control thread.
276 spin_lock(&dcache_lock);
277 LASSERT(list_empty(&rmnt->mnt_lustre_list));
278 list_add_tail(&rmnt->mnt_lustre_list,
280 spin_unlock(&dcache_lock);
281 rmnt->mnt_last_used = jiffies;
289 CERROR("usermode upcall %s failed to mount %s, err %d\n",
290 sbi->ll_gns_upcall, path, rc);
295 switch (cleanup_phase) {
297 free_page((unsigned long)datapage);
299 if (mntinfo_fd != NULL)
305 free_page((unsigned long)pathpage);
307 spin_lock(&sbi->ll_gns_lock);
308 sbi->ll_gns_state = LL_GNS_IDLE;
309 spin_unlock(&sbi->ll_gns_lock);
314 /* tries to umount passed @mnt. */
315 int ll_gns_umount_object(struct vfsmount *mnt)
320 CDEBUG(D_INODE, "unmounting mnt %p\n", mnt);
321 rc = do_umount(mnt, 0);
323 CDEBUG(D_INODE, "can't umount 0x%p, err = %d\n",
330 int ll_gns_check_mounts(struct ll_sb_info *sbi, int flags)
332 struct list_head check_list = LIST_HEAD_INIT(check_list);
333 struct vfsmount *mnt;
337 spin_lock(&dcache_lock);
338 list_splice_init(&sbi->ll_mnt_list, &check_list);
341 * walk the list in reverse order, and put them on the front of the sbi
342 * list each iteration; this avoids list-ordering problems if we race
343 * with another gns-mounting thread.
345 while (!list_empty(&check_list)) {
346 mnt = list_entry(check_list.prev,
352 list_del_init(&mnt->mnt_lustre_list);
354 list_add(&mnt->mnt_lustre_list,
357 /* check for timeout if needed */
358 pass = jiffies - mnt->mnt_last_used;
360 if (flags == LL_GNS_CHECK &&
361 pass < sbi->ll_gns_timeout * HZ)
366 spin_unlock(&dcache_lock);
369 ll_gns_umount_object(mnt);
372 spin_lock(&dcache_lock);
374 spin_unlock(&dcache_lock);
379 * GNS timer callback function. It restarts gns timer and wakes up GNS control
380 * thread to process mounts list.
382 void ll_gns_timer_callback(unsigned long data)
384 struct ll_sb_info *sbi = (void *)data;
387 spin_lock(&gns_lock);
388 if (list_empty(&sbi->ll_gns_sbi_head))
389 list_add(&sbi->ll_gns_sbi_head, &gns_sbi_list);
390 spin_unlock(&gns_lock);
392 wake_up(&gns_thread.t_ctl_waitq);
393 mod_timer(&sbi->ll_gns_timer,
394 jiffies + sbi->ll_gns_tick * HZ);
397 /* this function checks if something new happened to exist in gns list. */
398 static int inline ll_gns_check_event(void)
402 spin_lock(&gns_lock);
403 rc = !list_empty(&gns_sbi_list);
404 spin_unlock(&gns_lock);
409 /* should we stop GNS control thread? */
410 static int inline ll_gns_check_stop(void)
413 return (gns_thread.t_flags & SVC_STOPPING) ? 1 : 0;
416 /* GNS control thread function. */
417 static int ll_gns_thread_main(void *arg)
419 struct ll_gns_ctl *ctl = arg;
424 char name[sizeof(current->comm)];
425 snprintf(name, sizeof(name) - 1, "ll_gns");
426 kportal_daemonize(name);
429 SIGNAL_MASK_LOCK(current, flags);
430 sigfillset(¤t->blocked);
432 SIGNAL_MASK_UNLOCK(current, flags);
435 * letting starting function know, that we are ready and control may be
438 gns_thread.t_flags = SVC_RUNNING;
439 complete(&ctl->gc_starting);
441 while (!ll_gns_check_stop()) {
442 struct l_wait_info lwi = { 0 };
444 l_wait_event(gns_thread.t_ctl_waitq,
445 (ll_gns_check_event() ||
446 ll_gns_check_stop()), &lwi);
448 spin_lock(&gns_lock);
449 while (!list_empty(&gns_sbi_list)) {
450 struct ll_sb_info *sbi;
452 sbi = list_entry(gns_sbi_list.prev,
456 list_del_init(&sbi->ll_gns_sbi_head);
457 spin_unlock(&gns_lock);
458 ll_gns_check_mounts(sbi, LL_GNS_CHECK);
459 spin_lock(&gns_lock);
461 spin_unlock(&gns_lock);
465 gns_thread.t_flags = SVC_STOPPED;
467 /* this is SMP-safe way to finish thread. */
468 complete_and_exit(&ctl->gc_finishing, 0);
471 void ll_gns_add_timer(struct ll_sb_info *sbi)
473 mod_timer(&sbi->ll_gns_timer,
474 jiffies + sbi->ll_gns_tick * HZ);
477 void ll_gns_del_timer(struct ll_sb_info *sbi)
479 del_timer(&sbi->ll_gns_timer);
483 * starts GNS control thread and waits for a signal it is up and work may be
486 int ll_gns_start_thread(void)
491 LASSERT(gns_thread.t_flags == 0);
492 init_completion(&gns_ctl.gc_starting);
493 init_completion(&gns_ctl.gc_finishing);
494 init_waitqueue_head(&gns_thread.t_ctl_waitq);
496 rc = kernel_thread(ll_gns_thread_main, &gns_ctl,
497 (CLONE_VM | CLONE_FILES));
499 CERROR("cannot start GNS control thread, "
503 wait_for_completion(&gns_ctl.gc_starting);
504 LASSERT(gns_thread.t_flags == SVC_RUNNING);
508 /* stops GNS control thread and waits its actual stop. */
509 void ll_gns_stop_thread(void)
512 gns_thread.t_flags = SVC_STOPPING;
513 wake_up(&gns_thread.t_ctl_waitq);
514 wait_for_completion(&gns_ctl.gc_finishing);
515 LASSERT(gns_thread.t_flags == SVC_STOPPED);
516 gns_thread.t_flags = 0;