Root/fs/ocfs2/dlmfs/userdlm.c

1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * userdlm.c
5 *
6 * Code which implements the kernel side of a minimal userspace
7 * interface to our DLM.
8 *
9 * Many of the functions here are pared down versions of dlmglue.c
10 * functions.
11 *
12 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
13 *
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public
16 * License as published by the Free Software Foundation; either
17 * version 2 of the License, or (at your option) any later version.
18 *
19 * This program is distributed in the hope that it will be useful,
20 * but WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 * General Public License for more details.
23 *
24 * You should have received a copy of the GNU General Public
25 * License along with this program; if not, write to the
26 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
27 * Boston, MA 021110-1307, USA.
28 */
29
30#include <linux/signal.h>
31
32#include <linux/module.h>
33#include <linux/fs.h>
34#include <linux/types.h>
35#include <linux/crc32.h>
36
37#include "ocfs2_lockingver.h"
38#include "stackglue.h"
39#include "userdlm.h"
40
41#define MLOG_MASK_PREFIX ML_DLMFS
42#include "cluster/masklog.h"
43
44
45static inline struct user_lock_res *user_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
46{
47    return container_of(lksb, struct user_lock_res, l_lksb);
48}
49
50static inline int user_check_wait_flag(struct user_lock_res *lockres,
51                       int flag)
52{
53    int ret;
54
55    spin_lock(&lockres->l_lock);
56    ret = lockres->l_flags & flag;
57    spin_unlock(&lockres->l_lock);
58
59    return ret;
60}
61
62static inline void user_wait_on_busy_lock(struct user_lock_res *lockres)
63
64{
65    wait_event(lockres->l_event,
66           !user_check_wait_flag(lockres, USER_LOCK_BUSY));
67}
68
69static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres)
70
71{
72    wait_event(lockres->l_event,
73           !user_check_wait_flag(lockres, USER_LOCK_BLOCKED));
74}
75
76/* I heart container_of... */
77static inline struct ocfs2_cluster_connection *
78cluster_connection_from_user_lockres(struct user_lock_res *lockres)
79{
80    struct dlmfs_inode_private *ip;
81
82    ip = container_of(lockres,
83              struct dlmfs_inode_private,
84              ip_lockres);
85    return ip->ip_conn;
86}
87
88static struct inode *
89user_dlm_inode_from_user_lockres(struct user_lock_res *lockres)
90{
91    struct dlmfs_inode_private *ip;
92
93    ip = container_of(lockres,
94              struct dlmfs_inode_private,
95              ip_lockres);
96    return &ip->ip_vfs_inode;
97}
98
99static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
100{
101    spin_lock(&lockres->l_lock);
102    lockres->l_flags &= ~USER_LOCK_BUSY;
103    spin_unlock(&lockres->l_lock);
104}
105
106#define user_log_dlm_error(_func, _stat, _lockres) do { \
107    mlog(ML_ERROR, "Dlm error %d while calling %s on " \
108        "resource %.*s\n", _stat, _func, \
109        _lockres->l_namelen, _lockres->l_name); \
110} while (0)
111
112/* WARNING: This function lives in a world where the only three lock
113 * levels are EX, PR, and NL. It *will* have to be adjusted when more
114 * lock types are added. */
115static inline int user_highest_compat_lock_level(int level)
116{
117    int new_level = DLM_LOCK_EX;
118
119    if (level == DLM_LOCK_EX)
120        new_level = DLM_LOCK_NL;
121    else if (level == DLM_LOCK_PR)
122        new_level = DLM_LOCK_PR;
123    return new_level;
124}
125
126static void user_ast(struct ocfs2_dlm_lksb *lksb)
127{
128    struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
129    int status;
130
131    mlog(ML_BASTS, "AST fired for lockres %.*s, level %d => %d\n",
132         lockres->l_namelen, lockres->l_name, lockres->l_level,
133         lockres->l_requested);
134
135    spin_lock(&lockres->l_lock);
136
137    status = ocfs2_dlm_lock_status(&lockres->l_lksb);
138    if (status) {
139        mlog(ML_ERROR, "lksb status value of %u on lockres %.*s\n",
140             status, lockres->l_namelen, lockres->l_name);
141        spin_unlock(&lockres->l_lock);
142        return;
143    }
144
145    mlog_bug_on_msg(lockres->l_requested == DLM_LOCK_IV,
146            "Lockres %.*s, requested ivmode. flags 0x%x\n",
147            lockres->l_namelen, lockres->l_name, lockres->l_flags);
148
149    /* we're downconverting. */
150    if (lockres->l_requested < lockres->l_level) {
151        if (lockres->l_requested <=
152            user_highest_compat_lock_level(lockres->l_blocking)) {
153            lockres->l_blocking = DLM_LOCK_NL;
154            lockres->l_flags &= ~USER_LOCK_BLOCKED;
155        }
156    }
157
158    lockres->l_level = lockres->l_requested;
159    lockres->l_requested = DLM_LOCK_IV;
160    lockres->l_flags |= USER_LOCK_ATTACHED;
161    lockres->l_flags &= ~USER_LOCK_BUSY;
162
163    spin_unlock(&lockres->l_lock);
164
165    wake_up(&lockres->l_event);
166}
167
168static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres)
169{
170    struct inode *inode;
171    inode = user_dlm_inode_from_user_lockres(lockres);
172    if (!igrab(inode))
173        BUG();
174}
175
176static void user_dlm_unblock_lock(struct work_struct *work);
177
178static void __user_dlm_queue_lockres(struct user_lock_res *lockres)
179{
180    if (!(lockres->l_flags & USER_LOCK_QUEUED)) {
181        user_dlm_grab_inode_ref(lockres);
182
183        INIT_WORK(&lockres->l_work, user_dlm_unblock_lock);
184
185        queue_work(user_dlm_worker, &lockres->l_work);
186        lockres->l_flags |= USER_LOCK_QUEUED;
187    }
188}
189
190static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres)
191{
192    int queue = 0;
193
194    if (!(lockres->l_flags & USER_LOCK_BLOCKED))
195        return;
196
197    switch (lockres->l_blocking) {
198    case DLM_LOCK_EX:
199        if (!lockres->l_ex_holders && !lockres->l_ro_holders)
200            queue = 1;
201        break;
202    case DLM_LOCK_PR:
203        if (!lockres->l_ex_holders)
204            queue = 1;
205        break;
206    default:
207        BUG();
208    }
209
210    if (queue)
211        __user_dlm_queue_lockres(lockres);
212}
213
214static void user_bast(struct ocfs2_dlm_lksb *lksb, int level)
215{
216    struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
217
218    mlog(ML_BASTS, "BAST fired for lockres %.*s, blocking %d, level %d\n",
219         lockres->l_namelen, lockres->l_name, level, lockres->l_level);
220
221    spin_lock(&lockres->l_lock);
222    lockres->l_flags |= USER_LOCK_BLOCKED;
223    if (level > lockres->l_blocking)
224        lockres->l_blocking = level;
225
226    __user_dlm_queue_lockres(lockres);
227    spin_unlock(&lockres->l_lock);
228
229    wake_up(&lockres->l_event);
230}
231
232static void user_unlock_ast(struct ocfs2_dlm_lksb *lksb, int status)
233{
234    struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
235
236    mlog(ML_BASTS, "UNLOCK AST fired for lockres %.*s, flags 0x%x\n",
237         lockres->l_namelen, lockres->l_name, lockres->l_flags);
238
239    if (status)
240        mlog(ML_ERROR, "dlm returns status %d\n", status);
241
242    spin_lock(&lockres->l_lock);
243    /* The teardown flag gets set early during the unlock process,
244     * so test the cancel flag to make sure that this ast isn't
245     * for a concurrent cancel. */
246    if (lockres->l_flags & USER_LOCK_IN_TEARDOWN
247        && !(lockres->l_flags & USER_LOCK_IN_CANCEL)) {
248        lockres->l_level = DLM_LOCK_IV;
249    } else if (status == DLM_CANCELGRANT) {
250        /* We tried to cancel a convert request, but it was
251         * already granted. Don't clear the busy flag - the
252         * ast should've done this already. */
253        BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
254        lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
255        goto out_noclear;
256    } else {
257        BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
258        /* Cancel succeeded, we want to re-queue */
259        lockres->l_requested = DLM_LOCK_IV; /* cancel an
260                            * upconvert
261                            * request. */
262        lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
263        /* we want the unblock thread to look at it again
264         * now. */
265        if (lockres->l_flags & USER_LOCK_BLOCKED)
266            __user_dlm_queue_lockres(lockres);
267    }
268
269    lockres->l_flags &= ~USER_LOCK_BUSY;
270out_noclear:
271    spin_unlock(&lockres->l_lock);
272
273    wake_up(&lockres->l_event);
274}
275
276/*
277 * This is the userdlmfs locking protocol version.
278 *
279 * See fs/ocfs2/dlmglue.c for more details on locking versions.
280 */
281static struct ocfs2_locking_protocol user_dlm_lproto = {
282    .lp_max_version = {
283        .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
284        .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
285    },
286    .lp_lock_ast = user_ast,
287    .lp_blocking_ast = user_bast,
288    .lp_unlock_ast = user_unlock_ast,
289};
290
291static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres)
292{
293    struct inode *inode;
294    inode = user_dlm_inode_from_user_lockres(lockres);
295    iput(inode);
296}
297
298static void user_dlm_unblock_lock(struct work_struct *work)
299{
300    int new_level, status;
301    struct user_lock_res *lockres =
302        container_of(work, struct user_lock_res, l_work);
303    struct ocfs2_cluster_connection *conn =
304        cluster_connection_from_user_lockres(lockres);
305
306    mlog(0, "lockres %.*s\n", lockres->l_namelen, lockres->l_name);
307
308    spin_lock(&lockres->l_lock);
309
310    mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
311            "Lockres %.*s, flags 0x%x\n",
312            lockres->l_namelen, lockres->l_name, lockres->l_flags);
313
314    /* notice that we don't clear USER_LOCK_BLOCKED here. If it's
315     * set, we want user_ast clear it. */
316    lockres->l_flags &= ~USER_LOCK_QUEUED;
317
318    /* It's valid to get here and no longer be blocked - if we get
319     * several basts in a row, we might be queued by the first
320     * one, the unblock thread might run and clear the queued
321     * flag, and finally we might get another bast which re-queues
322     * us before our ast for the downconvert is called. */
323    if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
324        mlog(ML_BASTS, "lockres %.*s USER_LOCK_BLOCKED\n",
325             lockres->l_namelen, lockres->l_name);
326        spin_unlock(&lockres->l_lock);
327        goto drop_ref;
328    }
329
330    if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
331        mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_TEARDOWN\n",
332             lockres->l_namelen, lockres->l_name);
333        spin_unlock(&lockres->l_lock);
334        goto drop_ref;
335    }
336
337    if (lockres->l_flags & USER_LOCK_BUSY) {
338        if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
339            mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_CANCEL\n",
340                 lockres->l_namelen, lockres->l_name);
341            spin_unlock(&lockres->l_lock);
342            goto drop_ref;
343        }
344
345        lockres->l_flags |= USER_LOCK_IN_CANCEL;
346        spin_unlock(&lockres->l_lock);
347
348        status = ocfs2_dlm_unlock(conn, &lockres->l_lksb,
349                      DLM_LKF_CANCEL);
350        if (status)
351            user_log_dlm_error("ocfs2_dlm_unlock", status, lockres);
352        goto drop_ref;
353    }
354
355    /* If there are still incompat holders, we can exit safely
356     * without worrying about re-queueing this lock as that will
357     * happen on the last call to user_cluster_unlock. */
358    if ((lockres->l_blocking == DLM_LOCK_EX)
359        && (lockres->l_ex_holders || lockres->l_ro_holders)) {
360        spin_unlock(&lockres->l_lock);
361        mlog(ML_BASTS, "lockres %.*s, EX/PR Holders %u,%u\n",
362             lockres->l_namelen, lockres->l_name,
363             lockres->l_ex_holders, lockres->l_ro_holders);
364        goto drop_ref;
365    }
366
367    if ((lockres->l_blocking == DLM_LOCK_PR)
368        && lockres->l_ex_holders) {
369        spin_unlock(&lockres->l_lock);
370        mlog(ML_BASTS, "lockres %.*s, EX Holders %u\n",
371             lockres->l_namelen, lockres->l_name,
372             lockres->l_ex_holders);
373        goto drop_ref;
374    }
375
376    /* yay, we can downconvert now. */
377    new_level = user_highest_compat_lock_level(lockres->l_blocking);
378    lockres->l_requested = new_level;
379    lockres->l_flags |= USER_LOCK_BUSY;
380    mlog(ML_BASTS, "lockres %.*s, downconvert %d => %d\n",
381         lockres->l_namelen, lockres->l_name, lockres->l_level, new_level);
382    spin_unlock(&lockres->l_lock);
383
384    /* need lock downconvert request now... */
385    status = ocfs2_dlm_lock(conn, new_level, &lockres->l_lksb,
386                DLM_LKF_CONVERT|DLM_LKF_VALBLK,
387                lockres->l_name,
388                lockres->l_namelen);
389    if (status) {
390        user_log_dlm_error("ocfs2_dlm_lock", status, lockres);
391        user_recover_from_dlm_error(lockres);
392    }
393
394drop_ref:
395    user_dlm_drop_inode_ref(lockres);
396}
397
398static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
399                    int level)
400{
401    switch(level) {
402    case DLM_LOCK_EX:
403        lockres->l_ex_holders++;
404        break;
405    case DLM_LOCK_PR:
406        lockres->l_ro_holders++;
407        break;
408    default:
409        BUG();
410    }
411}
412
413/* predict what lock level we'll be dropping down to on behalf
414 * of another node, and return true if the currently wanted
415 * level will be compatible with it. */
416static inline int
417user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
418                  int wanted)
419{
420    BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
421
422    return wanted <= user_highest_compat_lock_level(lockres->l_blocking);
423}
424
425int user_dlm_cluster_lock(struct user_lock_res *lockres,
426              int level,
427              int lkm_flags)
428{
429    int status, local_flags;
430    struct ocfs2_cluster_connection *conn =
431        cluster_connection_from_user_lockres(lockres);
432
433    if (level != DLM_LOCK_EX &&
434        level != DLM_LOCK_PR) {
435        mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
436             lockres->l_namelen, lockres->l_name);
437        status = -EINVAL;
438        goto bail;
439    }
440
441    mlog(ML_BASTS, "lockres %.*s, level %d, flags = 0x%x\n",
442         lockres->l_namelen, lockres->l_name, level, lkm_flags);
443
444again:
445    if (signal_pending(current)) {
446        status = -ERESTARTSYS;
447        goto bail;
448    }
449
450    spin_lock(&lockres->l_lock);
451
452    /* We only compare against the currently granted level
453     * here. If the lock is blocked waiting on a downconvert,
454     * we'll get caught below. */
455    if ((lockres->l_flags & USER_LOCK_BUSY) &&
456        (level > lockres->l_level)) {
457        /* is someone sitting in dlm_lock? If so, wait on
458         * them. */
459        spin_unlock(&lockres->l_lock);
460
461        user_wait_on_busy_lock(lockres);
462        goto again;
463    }
464
465    if ((lockres->l_flags & USER_LOCK_BLOCKED) &&
466        (!user_may_continue_on_blocked_lock(lockres, level))) {
467        /* is the lock is currently blocked on behalf of
468         * another node */
469        spin_unlock(&lockres->l_lock);
470
471        user_wait_on_blocked_lock(lockres);
472        goto again;
473    }
474
475    if (level > lockres->l_level) {
476        local_flags = lkm_flags | DLM_LKF_VALBLK;
477        if (lockres->l_level != DLM_LOCK_IV)
478            local_flags |= DLM_LKF_CONVERT;
479
480        lockres->l_requested = level;
481        lockres->l_flags |= USER_LOCK_BUSY;
482        spin_unlock(&lockres->l_lock);
483
484        BUG_ON(level == DLM_LOCK_IV);
485        BUG_ON(level == DLM_LOCK_NL);
486
487        /* call dlm_lock to upgrade lock now */
488        status = ocfs2_dlm_lock(conn, level, &lockres->l_lksb,
489                    local_flags, lockres->l_name,
490                    lockres->l_namelen);
491        if (status) {
492            if ((lkm_flags & DLM_LKF_NOQUEUE) &&
493                (status != -EAGAIN))
494                user_log_dlm_error("ocfs2_dlm_lock",
495                           status, lockres);
496            user_recover_from_dlm_error(lockres);
497            goto bail;
498        }
499
500        user_wait_on_busy_lock(lockres);
501        goto again;
502    }
503
504    user_dlm_inc_holders(lockres, level);
505    spin_unlock(&lockres->l_lock);
506
507    status = 0;
508bail:
509    return status;
510}
511
512static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
513                    int level)
514{
515    switch(level) {
516    case DLM_LOCK_EX:
517        BUG_ON(!lockres->l_ex_holders);
518        lockres->l_ex_holders--;
519        break;
520    case DLM_LOCK_PR:
521        BUG_ON(!lockres->l_ro_holders);
522        lockres->l_ro_holders--;
523        break;
524    default:
525        BUG();
526    }
527}
528
529void user_dlm_cluster_unlock(struct user_lock_res *lockres,
530                 int level)
531{
532    if (level != DLM_LOCK_EX &&
533        level != DLM_LOCK_PR) {
534        mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
535             lockres->l_namelen, lockres->l_name);
536        return;
537    }
538
539    spin_lock(&lockres->l_lock);
540    user_dlm_dec_holders(lockres, level);
541    __user_dlm_cond_queue_lockres(lockres);
542    spin_unlock(&lockres->l_lock);
543}
544
545void user_dlm_write_lvb(struct inode *inode,
546            const char *val,
547            unsigned int len)
548{
549    struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
550    char *lvb;
551
552    BUG_ON(len > DLM_LVB_LEN);
553
554    spin_lock(&lockres->l_lock);
555
556    BUG_ON(lockres->l_level < DLM_LOCK_EX);
557    lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
558    memcpy(lvb, val, len);
559
560    spin_unlock(&lockres->l_lock);
561}
562
563ssize_t user_dlm_read_lvb(struct inode *inode,
564              char *val,
565              unsigned int len)
566{
567    struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
568    char *lvb;
569    ssize_t ret = len;
570
571    BUG_ON(len > DLM_LVB_LEN);
572
573    spin_lock(&lockres->l_lock);
574
575    BUG_ON(lockres->l_level < DLM_LOCK_PR);
576    if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)) {
577        lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
578        memcpy(val, lvb, len);
579    } else
580        ret = 0;
581
582    spin_unlock(&lockres->l_lock);
583    return ret;
584}
585
586void user_dlm_lock_res_init(struct user_lock_res *lockres,
587                struct dentry *dentry)
588{
589    memset(lockres, 0, sizeof(*lockres));
590
591    spin_lock_init(&lockres->l_lock);
592    init_waitqueue_head(&lockres->l_event);
593    lockres->l_level = DLM_LOCK_IV;
594    lockres->l_requested = DLM_LOCK_IV;
595    lockres->l_blocking = DLM_LOCK_IV;
596
597    /* should have been checked before getting here. */
598    BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);
599
600    memcpy(lockres->l_name,
601           dentry->d_name.name,
602           dentry->d_name.len);
603    lockres->l_namelen = dentry->d_name.len;
604}
605
606int user_dlm_destroy_lock(struct user_lock_res *lockres)
607{
608    int status = -EBUSY;
609    struct ocfs2_cluster_connection *conn =
610        cluster_connection_from_user_lockres(lockres);
611
612    mlog(ML_BASTS, "lockres %.*s\n", lockres->l_namelen, lockres->l_name);
613
614    spin_lock(&lockres->l_lock);
615    if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
616        spin_unlock(&lockres->l_lock);
617        return 0;
618    }
619
620    lockres->l_flags |= USER_LOCK_IN_TEARDOWN;
621
622    while (lockres->l_flags & USER_LOCK_BUSY) {
623        spin_unlock(&lockres->l_lock);
624
625        user_wait_on_busy_lock(lockres);
626
627        spin_lock(&lockres->l_lock);
628    }
629
630    if (lockres->l_ro_holders || lockres->l_ex_holders) {
631        spin_unlock(&lockres->l_lock);
632        goto bail;
633    }
634
635    status = 0;
636    if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
637        spin_unlock(&lockres->l_lock);
638        goto bail;
639    }
640
641    lockres->l_flags &= ~USER_LOCK_ATTACHED;
642    lockres->l_flags |= USER_LOCK_BUSY;
643    spin_unlock(&lockres->l_lock);
644
645    status = ocfs2_dlm_unlock(conn, &lockres->l_lksb, DLM_LKF_VALBLK);
646    if (status) {
647        user_log_dlm_error("ocfs2_dlm_unlock", status, lockres);
648        goto bail;
649    }
650
651    user_wait_on_busy_lock(lockres);
652
653    status = 0;
654bail:
655    return status;
656}
657
658static void user_dlm_recovery_handler_noop(int node_num,
659                       void *recovery_data)
660{
661    /* We ignore recovery events */
662    return;
663}
664
665void user_dlm_set_locking_protocol(void)
666{
667    ocfs2_stack_glue_set_max_proto_version(&user_dlm_lproto.lp_max_version);
668}
669
670struct ocfs2_cluster_connection *user_dlm_register(struct qstr *name)
671{
672    int rc;
673    struct ocfs2_cluster_connection *conn;
674
675    rc = ocfs2_cluster_connect_agnostic(name->name, name->len,
676                        &user_dlm_lproto,
677                        user_dlm_recovery_handler_noop,
678                        NULL, &conn);
679    if (rc)
680        mlog_errno(rc);
681
682    return rc ? ERR_PTR(rc) : conn;
683}
684
685void user_dlm_unregister(struct ocfs2_cluster_connection *conn)
686{
687    ocfs2_cluster_disconnect(conn, 0);
688}
689

Archive Download this file



interactive