Linux 5.8-rc4
[linux/fpc-iii.git] / fs / nfsd / nfs4recover.c
blob9e40dfecf1b1a6dd93c322452893c5e2ccf6af51
1 /*
2 * Copyright (c) 2004 The Regents of the University of Michigan.
3 * Copyright (c) 2012 Jeff Layton <jlayton@redhat.com>
4 * All rights reserved.
6 * Andy Adamson <andros@citi.umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <crypto/hash.h>
36 #include <linux/file.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/sched.h>
40 #include <linux/fs.h>
41 #include <linux/module.h>
42 #include <net/net_namespace.h>
43 #include <linux/sunrpc/rpc_pipe_fs.h>
44 #include <linux/sunrpc/clnt.h>
45 #include <linux/nfsd/cld.h>
47 #include "nfsd.h"
48 #include "state.h"
49 #include "vfs.h"
50 #include "netns.h"
52 #define NFSDDBG_FACILITY NFSDDBG_PROC
54 /* Declarations */
55 struct nfsd4_client_tracking_ops {
56 int (*init)(struct net *);
57 void (*exit)(struct net *);
58 void (*create)(struct nfs4_client *);
59 void (*remove)(struct nfs4_client *);
60 int (*check)(struct nfs4_client *);
61 void (*grace_done)(struct nfsd_net *);
62 uint8_t version;
63 size_t msglen;
66 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops;
67 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2;
69 /* Globals */
70 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
72 static int
73 nfs4_save_creds(const struct cred **original_creds)
75 struct cred *new;
77 new = prepare_creds();
78 if (!new)
79 return -ENOMEM;
81 new->fsuid = GLOBAL_ROOT_UID;
82 new->fsgid = GLOBAL_ROOT_GID;
83 *original_creds = override_creds(new);
84 put_cred(new);
85 return 0;
88 static void
89 nfs4_reset_creds(const struct cred *original)
91 revert_creds(original);
94 static void
95 md5_to_hex(char *out, char *md5)
97 int i;
99 for (i=0; i<16; i++) {
100 unsigned char c = md5[i];
102 *out++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
103 *out++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
105 *out = '\0';
108 static int
109 nfs4_make_rec_clidname(char *dname, const struct xdr_netobj *clname)
111 struct xdr_netobj cksum;
112 struct crypto_shash *tfm;
113 int status;
115 dprintk("NFSD: nfs4_make_rec_clidname for %.*s\n",
116 clname->len, clname->data);
117 tfm = crypto_alloc_shash("md5", 0, 0);
118 if (IS_ERR(tfm)) {
119 status = PTR_ERR(tfm);
120 goto out_no_tfm;
123 cksum.len = crypto_shash_digestsize(tfm);
124 cksum.data = kmalloc(cksum.len, GFP_KERNEL);
125 if (cksum.data == NULL) {
126 status = -ENOMEM;
127 goto out;
130 status = crypto_shash_tfm_digest(tfm, clname->data, clname->len,
131 cksum.data);
132 if (status)
133 goto out;
135 md5_to_hex(dname, cksum.data);
137 status = 0;
138 out:
139 kfree(cksum.data);
140 crypto_free_shash(tfm);
141 out_no_tfm:
142 return status;
146 * If we had an error generating the recdir name for the legacy tracker
147 * then warn the admin. If the error doesn't appear to be transient,
148 * then disable recovery tracking.
150 static void
151 legacy_recdir_name_error(struct nfs4_client *clp, int error)
153 printk(KERN_ERR "NFSD: unable to generate recoverydir "
154 "name (%d).\n", error);
157 * if the algorithm just doesn't exist, then disable the recovery
158 * tracker altogether. The crypto libs will generally return this if
159 * FIPS is enabled as well.
161 if (error == -ENOENT) {
162 printk(KERN_ERR "NFSD: disabling legacy clientid tracking. "
163 "Reboot recovery will not function correctly!\n");
164 nfsd4_client_tracking_exit(clp->net);
168 static void
169 __nfsd4_create_reclaim_record_grace(struct nfs4_client *clp,
170 const char *dname, int len, struct nfsd_net *nn)
172 struct xdr_netobj name;
173 struct xdr_netobj princhash = { .len = 0, .data = NULL };
174 struct nfs4_client_reclaim *crp;
176 name.data = kmemdup(dname, len, GFP_KERNEL);
177 if (!name.data) {
178 dprintk("%s: failed to allocate memory for name.data!\n",
179 __func__);
180 return;
182 name.len = len;
183 crp = nfs4_client_to_reclaim(name, princhash, nn);
184 if (!crp) {
185 kfree(name.data);
186 return;
188 crp->cr_clp = clp;
191 static void
192 nfsd4_create_clid_dir(struct nfs4_client *clp)
194 const struct cred *original_cred;
195 char dname[HEXDIR_LEN];
196 struct dentry *dir, *dentry;
197 int status;
198 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
200 if (test_and_set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
201 return;
202 if (!nn->rec_file)
203 return;
205 status = nfs4_make_rec_clidname(dname, &clp->cl_name);
206 if (status)
207 return legacy_recdir_name_error(clp, status);
209 status = nfs4_save_creds(&original_cred);
210 if (status < 0)
211 return;
213 status = mnt_want_write_file(nn->rec_file);
214 if (status)
215 goto out_creds;
217 dir = nn->rec_file->f_path.dentry;
218 /* lock the parent */
219 inode_lock(d_inode(dir));
221 dentry = lookup_one_len(dname, dir, HEXDIR_LEN-1);
222 if (IS_ERR(dentry)) {
223 status = PTR_ERR(dentry);
224 goto out_unlock;
226 if (d_really_is_positive(dentry))
228 * In the 4.1 case, where we're called from
229 * reclaim_complete(), records from the previous reboot
230 * may still be left, so this is OK.
232 * In the 4.0 case, we should never get here; but we may
233 * as well be forgiving and just succeed silently.
235 goto out_put;
236 status = vfs_mkdir(d_inode(dir), dentry, S_IRWXU);
237 out_put:
238 dput(dentry);
239 out_unlock:
240 inode_unlock(d_inode(dir));
241 if (status == 0) {
242 if (nn->in_grace)
243 __nfsd4_create_reclaim_record_grace(clp, dname,
244 HEXDIR_LEN, nn);
245 vfs_fsync(nn->rec_file, 0);
246 } else {
247 printk(KERN_ERR "NFSD: failed to write recovery record"
248 " (err %d); please check that %s exists"
249 " and is writeable", status,
250 user_recovery_dirname);
252 mnt_drop_write_file(nn->rec_file);
253 out_creds:
254 nfs4_reset_creds(original_cred);
257 typedef int (recdir_func)(struct dentry *, struct dentry *, struct nfsd_net *);
259 struct name_list {
260 char name[HEXDIR_LEN];
261 struct list_head list;
264 struct nfs4_dir_ctx {
265 struct dir_context ctx;
266 struct list_head names;
269 static int
270 nfsd4_build_namelist(struct dir_context *__ctx, const char *name, int namlen,
271 loff_t offset, u64 ino, unsigned int d_type)
273 struct nfs4_dir_ctx *ctx =
274 container_of(__ctx, struct nfs4_dir_ctx, ctx);
275 struct name_list *entry;
277 if (namlen != HEXDIR_LEN - 1)
278 return 0;
279 entry = kmalloc(sizeof(struct name_list), GFP_KERNEL);
280 if (entry == NULL)
281 return -ENOMEM;
282 memcpy(entry->name, name, HEXDIR_LEN - 1);
283 entry->name[HEXDIR_LEN - 1] = '\0';
284 list_add(&entry->list, &ctx->names);
285 return 0;
288 static int
289 nfsd4_list_rec_dir(recdir_func *f, struct nfsd_net *nn)
291 const struct cred *original_cred;
292 struct dentry *dir = nn->rec_file->f_path.dentry;
293 struct nfs4_dir_ctx ctx = {
294 .ctx.actor = nfsd4_build_namelist,
295 .names = LIST_HEAD_INIT(ctx.names)
297 struct name_list *entry, *tmp;
298 int status;
300 status = nfs4_save_creds(&original_cred);
301 if (status < 0)
302 return status;
304 status = vfs_llseek(nn->rec_file, 0, SEEK_SET);
305 if (status < 0) {
306 nfs4_reset_creds(original_cred);
307 return status;
310 status = iterate_dir(nn->rec_file, &ctx.ctx);
311 inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
313 list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
314 if (!status) {
315 struct dentry *dentry;
316 dentry = lookup_one_len(entry->name, dir, HEXDIR_LEN-1);
317 if (IS_ERR(dentry)) {
318 status = PTR_ERR(dentry);
319 break;
321 status = f(dir, dentry, nn);
322 dput(dentry);
324 list_del(&entry->list);
325 kfree(entry);
327 inode_unlock(d_inode(dir));
328 nfs4_reset_creds(original_cred);
330 list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
331 dprintk("NFSD: %s. Left entry %s\n", __func__, entry->name);
332 list_del(&entry->list);
333 kfree(entry);
335 return status;
338 static int
339 nfsd4_unlink_clid_dir(char *name, int namlen, struct nfsd_net *nn)
341 struct dentry *dir, *dentry;
342 int status;
344 dprintk("NFSD: nfsd4_unlink_clid_dir. name %.*s\n", namlen, name);
346 dir = nn->rec_file->f_path.dentry;
347 inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
348 dentry = lookup_one_len(name, dir, namlen);
349 if (IS_ERR(dentry)) {
350 status = PTR_ERR(dentry);
351 goto out_unlock;
353 status = -ENOENT;
354 if (d_really_is_negative(dentry))
355 goto out;
356 status = vfs_rmdir(d_inode(dir), dentry);
357 out:
358 dput(dentry);
359 out_unlock:
360 inode_unlock(d_inode(dir));
361 return status;
364 static void
365 __nfsd4_remove_reclaim_record_grace(const char *dname, int len,
366 struct nfsd_net *nn)
368 struct xdr_netobj name;
369 struct nfs4_client_reclaim *crp;
371 name.data = kmemdup(dname, len, GFP_KERNEL);
372 if (!name.data) {
373 dprintk("%s: failed to allocate memory for name.data!\n",
374 __func__);
375 return;
377 name.len = len;
378 crp = nfsd4_find_reclaim_client(name, nn);
379 kfree(name.data);
380 if (crp)
381 nfs4_remove_reclaim_record(crp, nn);
384 static void
385 nfsd4_remove_clid_dir(struct nfs4_client *clp)
387 const struct cred *original_cred;
388 char dname[HEXDIR_LEN];
389 int status;
390 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
392 if (!nn->rec_file || !test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
393 return;
395 status = nfs4_make_rec_clidname(dname, &clp->cl_name);
396 if (status)
397 return legacy_recdir_name_error(clp, status);
399 status = mnt_want_write_file(nn->rec_file);
400 if (status)
401 goto out;
402 clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
404 status = nfs4_save_creds(&original_cred);
405 if (status < 0)
406 goto out_drop_write;
408 status = nfsd4_unlink_clid_dir(dname, HEXDIR_LEN-1, nn);
409 nfs4_reset_creds(original_cred);
410 if (status == 0) {
411 vfs_fsync(nn->rec_file, 0);
412 if (nn->in_grace)
413 __nfsd4_remove_reclaim_record_grace(dname,
414 HEXDIR_LEN, nn);
416 out_drop_write:
417 mnt_drop_write_file(nn->rec_file);
418 out:
419 if (status)
420 printk("NFSD: Failed to remove expired client state directory"
421 " %.*s\n", HEXDIR_LEN, dname);
424 static int
425 purge_old(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
427 int status;
428 struct xdr_netobj name;
430 if (child->d_name.len != HEXDIR_LEN - 1) {
431 printk("%s: illegal name %pd in recovery directory\n",
432 __func__, child);
433 /* Keep trying; maybe the others are OK: */
434 return 0;
436 name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
437 if (!name.data) {
438 dprintk("%s: failed to allocate memory for name.data!\n",
439 __func__);
440 goto out;
442 name.len = HEXDIR_LEN;
443 if (nfs4_has_reclaimed_state(name, nn))
444 goto out_free;
446 status = vfs_rmdir(d_inode(parent), child);
447 if (status)
448 printk("failed to remove client recovery directory %pd\n",
449 child);
450 out_free:
451 kfree(name.data);
452 out:
453 /* Keep trying, success or failure: */
454 return 0;
457 static void
458 nfsd4_recdir_purge_old(struct nfsd_net *nn)
460 int status;
462 nn->in_grace = false;
463 if (!nn->rec_file)
464 return;
465 status = mnt_want_write_file(nn->rec_file);
466 if (status)
467 goto out;
468 status = nfsd4_list_rec_dir(purge_old, nn);
469 if (status == 0)
470 vfs_fsync(nn->rec_file, 0);
471 mnt_drop_write_file(nn->rec_file);
472 out:
473 nfs4_release_reclaim(nn);
474 if (status)
475 printk("nfsd4: failed to purge old clients from recovery"
476 " directory %pD\n", nn->rec_file);
479 static int
480 load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
482 struct xdr_netobj name;
483 struct xdr_netobj princhash = { .len = 0, .data = NULL };
485 if (child->d_name.len != HEXDIR_LEN - 1) {
486 printk("%s: illegal name %pd in recovery directory\n",
487 __func__, child);
488 /* Keep trying; maybe the others are OK: */
489 return 0;
491 name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
492 if (!name.data) {
493 dprintk("%s: failed to allocate memory for name.data!\n",
494 __func__);
495 goto out;
497 name.len = HEXDIR_LEN;
498 if (!nfs4_client_to_reclaim(name, princhash, nn))
499 kfree(name.data);
500 out:
501 return 0;
504 static int
505 nfsd4_recdir_load(struct net *net) {
506 int status;
507 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
509 if (!nn->rec_file)
510 return 0;
512 status = nfsd4_list_rec_dir(load_recdir, nn);
513 if (status)
514 printk("nfsd4: failed loading clients from recovery"
515 " directory %pD\n", nn->rec_file);
516 return status;
520 * Hold reference to the recovery directory.
523 static int
524 nfsd4_init_recdir(struct net *net)
526 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
527 const struct cred *original_cred;
528 int status;
530 printk("NFSD: Using %s as the NFSv4 state recovery directory\n",
531 user_recovery_dirname);
533 BUG_ON(nn->rec_file);
535 status = nfs4_save_creds(&original_cred);
536 if (status < 0) {
537 printk("NFSD: Unable to change credentials to find recovery"
538 " directory: error %d\n",
539 status);
540 return status;
543 nn->rec_file = filp_open(user_recovery_dirname, O_RDONLY | O_DIRECTORY, 0);
544 if (IS_ERR(nn->rec_file)) {
545 printk("NFSD: unable to find recovery directory %s\n",
546 user_recovery_dirname);
547 status = PTR_ERR(nn->rec_file);
548 nn->rec_file = NULL;
551 nfs4_reset_creds(original_cred);
552 if (!status)
553 nn->in_grace = true;
554 return status;
557 static void
558 nfsd4_shutdown_recdir(struct net *net)
560 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
562 if (!nn->rec_file)
563 return;
564 fput(nn->rec_file);
565 nn->rec_file = NULL;
568 static int
569 nfs4_legacy_state_init(struct net *net)
571 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
572 int i;
574 nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
575 sizeof(struct list_head),
576 GFP_KERNEL);
577 if (!nn->reclaim_str_hashtbl)
578 return -ENOMEM;
580 for (i = 0; i < CLIENT_HASH_SIZE; i++)
581 INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
582 nn->reclaim_str_hashtbl_size = 0;
584 return 0;
587 static void
588 nfs4_legacy_state_shutdown(struct net *net)
590 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
592 kfree(nn->reclaim_str_hashtbl);
595 static int
596 nfsd4_load_reboot_recovery_data(struct net *net)
598 int status;
600 status = nfsd4_init_recdir(net);
601 if (status)
602 return status;
604 status = nfsd4_recdir_load(net);
605 if (status)
606 nfsd4_shutdown_recdir(net);
608 return status;
611 static int
612 nfsd4_legacy_tracking_init(struct net *net)
614 int status;
616 /* XXX: The legacy code won't work in a container */
617 if (net != &init_net) {
618 pr_warn("NFSD: attempt to initialize legacy client tracking in a container ignored.\n");
619 return -EINVAL;
622 status = nfs4_legacy_state_init(net);
623 if (status)
624 return status;
626 status = nfsd4_load_reboot_recovery_data(net);
627 if (status)
628 goto err;
629 printk("NFSD: Using legacy client tracking operations.\n");
630 return 0;
632 err:
633 nfs4_legacy_state_shutdown(net);
634 return status;
637 static void
638 nfsd4_legacy_tracking_exit(struct net *net)
640 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
642 nfs4_release_reclaim(nn);
643 nfsd4_shutdown_recdir(net);
644 nfs4_legacy_state_shutdown(net);
648 * Change the NFSv4 recovery directory to recdir.
651 nfs4_reset_recoverydir(char *recdir)
653 int status;
654 struct path path;
656 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
657 if (status)
658 return status;
659 status = -ENOTDIR;
660 if (d_is_dir(path.dentry)) {
661 strcpy(user_recovery_dirname, recdir);
662 status = 0;
664 path_put(&path);
665 return status;
668 char *
669 nfs4_recoverydir(void)
671 return user_recovery_dirname;
674 static int
675 nfsd4_check_legacy_client(struct nfs4_client *clp)
677 int status;
678 char dname[HEXDIR_LEN];
679 struct nfs4_client_reclaim *crp;
680 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
681 struct xdr_netobj name;
683 /* did we already find that this client is stable? */
684 if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
685 return 0;
687 status = nfs4_make_rec_clidname(dname, &clp->cl_name);
688 if (status) {
689 legacy_recdir_name_error(clp, status);
690 return status;
693 /* look for it in the reclaim hashtable otherwise */
694 name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
695 if (!name.data) {
696 dprintk("%s: failed to allocate memory for name.data!\n",
697 __func__);
698 goto out_enoent;
700 name.len = HEXDIR_LEN;
701 crp = nfsd4_find_reclaim_client(name, nn);
702 kfree(name.data);
703 if (crp) {
704 set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
705 crp->cr_clp = clp;
706 return 0;
709 out_enoent:
710 return -ENOENT;
713 static const struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
714 .init = nfsd4_legacy_tracking_init,
715 .exit = nfsd4_legacy_tracking_exit,
716 .create = nfsd4_create_clid_dir,
717 .remove = nfsd4_remove_clid_dir,
718 .check = nfsd4_check_legacy_client,
719 .grace_done = nfsd4_recdir_purge_old,
720 .version = 1,
721 .msglen = 0,
724 /* Globals */
725 #define NFSD_PIPE_DIR "nfsd"
726 #define NFSD_CLD_PIPE "cld"
728 /* per-net-ns structure for holding cld upcall info */
729 struct cld_net {
730 struct rpc_pipe *cn_pipe;
731 spinlock_t cn_lock;
732 struct list_head cn_list;
733 unsigned int cn_xid;
734 bool cn_has_legacy;
735 struct crypto_shash *cn_tfm;
738 struct cld_upcall {
739 struct list_head cu_list;
740 struct cld_net *cu_net;
741 struct completion cu_done;
742 union {
743 struct cld_msg_hdr cu_hdr;
744 struct cld_msg cu_msg;
745 struct cld_msg_v2 cu_msg_v2;
746 } cu_u;
749 static int
750 __cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg)
752 int ret;
753 struct rpc_pipe_msg msg;
754 struct cld_upcall *cup = container_of(cmsg, struct cld_upcall, cu_u);
755 struct nfsd_net *nn = net_generic(pipe->dentry->d_sb->s_fs_info,
756 nfsd_net_id);
758 memset(&msg, 0, sizeof(msg));
759 msg.data = cmsg;
760 msg.len = nn->client_tracking_ops->msglen;
762 ret = rpc_queue_upcall(pipe, &msg);
763 if (ret < 0) {
764 goto out;
767 wait_for_completion(&cup->cu_done);
769 if (msg.errno < 0)
770 ret = msg.errno;
771 out:
772 return ret;
775 static int
776 cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg)
778 int ret;
781 * -EAGAIN occurs when pipe is closed and reopened while there are
782 * upcalls queued.
784 do {
785 ret = __cld_pipe_upcall(pipe, cmsg);
786 } while (ret == -EAGAIN);
788 return ret;
791 static ssize_t
792 __cld_pipe_inprogress_downcall(const struct cld_msg_v2 __user *cmsg,
793 struct nfsd_net *nn)
795 uint8_t cmd, princhashlen;
796 struct xdr_netobj name, princhash = { .len = 0, .data = NULL };
797 uint16_t namelen;
798 struct cld_net *cn = nn->cld_net;
800 if (get_user(cmd, &cmsg->cm_cmd)) {
801 dprintk("%s: error when copying cmd from userspace", __func__);
802 return -EFAULT;
804 if (cmd == Cld_GraceStart) {
805 if (nn->client_tracking_ops->version >= 2) {
806 const struct cld_clntinfo __user *ci;
808 ci = &cmsg->cm_u.cm_clntinfo;
809 if (get_user(namelen, &ci->cc_name.cn_len))
810 return -EFAULT;
811 name.data = memdup_user(&ci->cc_name.cn_id, namelen);
812 if (IS_ERR_OR_NULL(name.data))
813 return -EFAULT;
814 name.len = namelen;
815 get_user(princhashlen, &ci->cc_princhash.cp_len);
816 if (princhashlen > 0) {
817 princhash.data = memdup_user(
818 &ci->cc_princhash.cp_data,
819 princhashlen);
820 if (IS_ERR_OR_NULL(princhash.data))
821 return -EFAULT;
822 princhash.len = princhashlen;
823 } else
824 princhash.len = 0;
825 } else {
826 const struct cld_name __user *cnm;
828 cnm = &cmsg->cm_u.cm_name;
829 if (get_user(namelen, &cnm->cn_len))
830 return -EFAULT;
831 name.data = memdup_user(&cnm->cn_id, namelen);
832 if (IS_ERR_OR_NULL(name.data))
833 return -EFAULT;
834 name.len = namelen;
836 if (name.len > 5 && memcmp(name.data, "hash:", 5) == 0) {
837 name.len = name.len - 5;
838 memmove(name.data, name.data + 5, name.len);
839 cn->cn_has_legacy = true;
841 if (!nfs4_client_to_reclaim(name, princhash, nn)) {
842 kfree(name.data);
843 kfree(princhash.data);
844 return -EFAULT;
846 return nn->client_tracking_ops->msglen;
848 return -EFAULT;
851 static ssize_t
852 cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
854 struct cld_upcall *tmp, *cup;
855 struct cld_msg_hdr __user *hdr = (struct cld_msg_hdr __user *)src;
856 struct cld_msg_v2 __user *cmsg = (struct cld_msg_v2 __user *)src;
857 uint32_t xid;
858 struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
859 nfsd_net_id);
860 struct cld_net *cn = nn->cld_net;
861 int16_t status;
863 if (mlen != nn->client_tracking_ops->msglen) {
864 dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
865 nn->client_tracking_ops->msglen);
866 return -EINVAL;
869 /* copy just the xid so we can try to find that */
870 if (copy_from_user(&xid, &hdr->cm_xid, sizeof(xid)) != 0) {
871 dprintk("%s: error when copying xid from userspace", __func__);
872 return -EFAULT;
876 * copy the status so we know whether to remove the upcall from the
877 * list (for -EINPROGRESS, we just want to make sure the xid is
878 * valid, not remove the upcall from the list)
880 if (get_user(status, &hdr->cm_status)) {
881 dprintk("%s: error when copying status from userspace", __func__);
882 return -EFAULT;
885 /* walk the list and find corresponding xid */
886 cup = NULL;
887 spin_lock(&cn->cn_lock);
888 list_for_each_entry(tmp, &cn->cn_list, cu_list) {
889 if (get_unaligned(&tmp->cu_u.cu_hdr.cm_xid) == xid) {
890 cup = tmp;
891 if (status != -EINPROGRESS)
892 list_del_init(&cup->cu_list);
893 break;
896 spin_unlock(&cn->cn_lock);
898 /* couldn't find upcall? */
899 if (!cup) {
900 dprintk("%s: couldn't find upcall -- xid=%u\n", __func__, xid);
901 return -EINVAL;
904 if (status == -EINPROGRESS)
905 return __cld_pipe_inprogress_downcall(cmsg, nn);
907 if (copy_from_user(&cup->cu_u.cu_msg_v2, src, mlen) != 0)
908 return -EFAULT;
910 complete(&cup->cu_done);
911 return mlen;
914 static void
915 cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
917 struct cld_msg *cmsg = msg->data;
918 struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
919 cu_u.cu_msg);
921 /* errno >= 0 means we got a downcall */
922 if (msg->errno >= 0)
923 return;
925 complete(&cup->cu_done);
928 static const struct rpc_pipe_ops cld_upcall_ops = {
929 .upcall = rpc_pipe_generic_upcall,
930 .downcall = cld_pipe_downcall,
931 .destroy_msg = cld_pipe_destroy_msg,
934 static struct dentry *
935 nfsd4_cld_register_sb(struct super_block *sb, struct rpc_pipe *pipe)
937 struct dentry *dir, *dentry;
939 dir = rpc_d_lookup_sb(sb, NFSD_PIPE_DIR);
940 if (dir == NULL)
941 return ERR_PTR(-ENOENT);
942 dentry = rpc_mkpipe_dentry(dir, NFSD_CLD_PIPE, NULL, pipe);
943 dput(dir);
944 return dentry;
947 static void
948 nfsd4_cld_unregister_sb(struct rpc_pipe *pipe)
950 if (pipe->dentry)
951 rpc_unlink(pipe->dentry);
954 static struct dentry *
955 nfsd4_cld_register_net(struct net *net, struct rpc_pipe *pipe)
957 struct super_block *sb;
958 struct dentry *dentry;
960 sb = rpc_get_sb_net(net);
961 if (!sb)
962 return NULL;
963 dentry = nfsd4_cld_register_sb(sb, pipe);
964 rpc_put_sb_net(net);
965 return dentry;
968 static void
969 nfsd4_cld_unregister_net(struct net *net, struct rpc_pipe *pipe)
971 struct super_block *sb;
973 sb = rpc_get_sb_net(net);
974 if (sb) {
975 nfsd4_cld_unregister_sb(pipe);
976 rpc_put_sb_net(net);
980 /* Initialize rpc_pipefs pipe for communication with client tracking daemon */
981 static int
982 __nfsd4_init_cld_pipe(struct net *net)
984 int ret;
985 struct dentry *dentry;
986 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
987 struct cld_net *cn;
989 if (nn->cld_net)
990 return 0;
992 cn = kzalloc(sizeof(*cn), GFP_KERNEL);
993 if (!cn) {
994 ret = -ENOMEM;
995 goto err;
998 cn->cn_pipe = rpc_mkpipe_data(&cld_upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
999 if (IS_ERR(cn->cn_pipe)) {
1000 ret = PTR_ERR(cn->cn_pipe);
1001 goto err;
1003 spin_lock_init(&cn->cn_lock);
1004 INIT_LIST_HEAD(&cn->cn_list);
1006 dentry = nfsd4_cld_register_net(net, cn->cn_pipe);
1007 if (IS_ERR(dentry)) {
1008 ret = PTR_ERR(dentry);
1009 goto err_destroy_data;
1012 cn->cn_pipe->dentry = dentry;
1013 cn->cn_has_legacy = false;
1014 nn->cld_net = cn;
1015 return 0;
1017 err_destroy_data:
1018 rpc_destroy_pipe_data(cn->cn_pipe);
1019 err:
1020 kfree(cn);
1021 printk(KERN_ERR "NFSD: unable to create nfsdcld upcall pipe (%d)\n",
1022 ret);
1023 return ret;
1026 static int
1027 nfsd4_init_cld_pipe(struct net *net)
1029 int status;
1031 status = __nfsd4_init_cld_pipe(net);
1032 if (!status)
1033 printk("NFSD: Using old nfsdcld client tracking operations.\n");
1034 return status;
1037 static void
1038 nfsd4_remove_cld_pipe(struct net *net)
1040 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1041 struct cld_net *cn = nn->cld_net;
1043 nfsd4_cld_unregister_net(net, cn->cn_pipe);
1044 rpc_destroy_pipe_data(cn->cn_pipe);
1045 if (cn->cn_tfm)
1046 crypto_free_shash(cn->cn_tfm);
1047 kfree(nn->cld_net);
1048 nn->cld_net = NULL;
1051 static struct cld_upcall *
1052 alloc_cld_upcall(struct nfsd_net *nn)
1054 struct cld_upcall *new, *tmp;
1055 struct cld_net *cn = nn->cld_net;
1057 new = kzalloc(sizeof(*new), GFP_KERNEL);
1058 if (!new)
1059 return new;
1061 /* FIXME: hard cap on number in flight? */
1062 restart_search:
1063 spin_lock(&cn->cn_lock);
1064 list_for_each_entry(tmp, &cn->cn_list, cu_list) {
1065 if (tmp->cu_u.cu_msg.cm_xid == cn->cn_xid) {
1066 cn->cn_xid++;
1067 spin_unlock(&cn->cn_lock);
1068 goto restart_search;
1071 init_completion(&new->cu_done);
1072 new->cu_u.cu_msg.cm_vers = nn->client_tracking_ops->version;
1073 put_unaligned(cn->cn_xid++, &new->cu_u.cu_msg.cm_xid);
1074 new->cu_net = cn;
1075 list_add(&new->cu_list, &cn->cn_list);
1076 spin_unlock(&cn->cn_lock);
1078 dprintk("%s: allocated xid %u\n", __func__, new->cu_u.cu_msg.cm_xid);
1080 return new;
1083 static void
1084 free_cld_upcall(struct cld_upcall *victim)
1086 struct cld_net *cn = victim->cu_net;
1088 spin_lock(&cn->cn_lock);
1089 list_del(&victim->cu_list);
1090 spin_unlock(&cn->cn_lock);
1091 kfree(victim);
1094 /* Ask daemon to create a new record */
1095 static void
1096 nfsd4_cld_create(struct nfs4_client *clp)
1098 int ret;
1099 struct cld_upcall *cup;
1100 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1101 struct cld_net *cn = nn->cld_net;
1103 /* Don't upcall if it's already stored */
1104 if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1105 return;
1107 cup = alloc_cld_upcall(nn);
1108 if (!cup) {
1109 ret = -ENOMEM;
1110 goto out_err;
1113 cup->cu_u.cu_msg.cm_cmd = Cld_Create;
1114 cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1115 memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1116 clp->cl_name.len);
1118 ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1119 if (!ret) {
1120 ret = cup->cu_u.cu_msg.cm_status;
1121 set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1124 free_cld_upcall(cup);
1125 out_err:
1126 if (ret)
1127 printk(KERN_ERR "NFSD: Unable to create client "
1128 "record on stable storage: %d\n", ret);
1131 /* Ask daemon to create a new record */
1132 static void
1133 nfsd4_cld_create_v2(struct nfs4_client *clp)
1135 int ret;
1136 struct cld_upcall *cup;
1137 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1138 struct cld_net *cn = nn->cld_net;
1139 struct cld_msg_v2 *cmsg;
1140 struct crypto_shash *tfm = cn->cn_tfm;
1141 struct xdr_netobj cksum;
1142 char *principal = NULL;
1144 /* Don't upcall if it's already stored */
1145 if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1146 return;
1148 cup = alloc_cld_upcall(nn);
1149 if (!cup) {
1150 ret = -ENOMEM;
1151 goto out_err;
1154 cmsg = &cup->cu_u.cu_msg_v2;
1155 cmsg->cm_cmd = Cld_Create;
1156 cmsg->cm_u.cm_clntinfo.cc_name.cn_len = clp->cl_name.len;
1157 memcpy(cmsg->cm_u.cm_clntinfo.cc_name.cn_id, clp->cl_name.data,
1158 clp->cl_name.len);
1159 if (clp->cl_cred.cr_raw_principal)
1160 principal = clp->cl_cred.cr_raw_principal;
1161 else if (clp->cl_cred.cr_principal)
1162 principal = clp->cl_cred.cr_principal;
1163 if (principal) {
1164 cksum.len = crypto_shash_digestsize(tfm);
1165 cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1166 if (cksum.data == NULL) {
1167 ret = -ENOMEM;
1168 goto out;
1170 ret = crypto_shash_tfm_digest(tfm, principal, strlen(principal),
1171 cksum.data);
1172 if (ret) {
1173 kfree(cksum.data);
1174 goto out;
1176 cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = cksum.len;
1177 memcpy(cmsg->cm_u.cm_clntinfo.cc_princhash.cp_data,
1178 cksum.data, cksum.len);
1179 kfree(cksum.data);
1180 } else
1181 cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = 0;
1183 ret = cld_pipe_upcall(cn->cn_pipe, cmsg);
1184 if (!ret) {
1185 ret = cmsg->cm_status;
1186 set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1189 out:
1190 free_cld_upcall(cup);
1191 out_err:
1192 if (ret)
1193 pr_err("NFSD: Unable to create client record on stable storage: %d\n",
1194 ret);
1197 /* Ask daemon to create a new record */
1198 static void
1199 nfsd4_cld_remove(struct nfs4_client *clp)
1201 int ret;
1202 struct cld_upcall *cup;
1203 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1204 struct cld_net *cn = nn->cld_net;
1206 /* Don't upcall if it's already removed */
1207 if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1208 return;
1210 cup = alloc_cld_upcall(nn);
1211 if (!cup) {
1212 ret = -ENOMEM;
1213 goto out_err;
1216 cup->cu_u.cu_msg.cm_cmd = Cld_Remove;
1217 cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1218 memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1219 clp->cl_name.len);
1221 ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1222 if (!ret) {
1223 ret = cup->cu_u.cu_msg.cm_status;
1224 clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1227 free_cld_upcall(cup);
1228 out_err:
1229 if (ret)
1230 printk(KERN_ERR "NFSD: Unable to remove client "
1231 "record from stable storage: %d\n", ret);
1235 * For older nfsdcld's that do not allow us to "slurp" the clients
1236 * from the tracking database during startup.
1238 * Check for presence of a record, and update its timestamp
1240 static int
1241 nfsd4_cld_check_v0(struct nfs4_client *clp)
1243 int ret;
1244 struct cld_upcall *cup;
1245 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1246 struct cld_net *cn = nn->cld_net;
1248 /* Don't upcall if one was already stored during this grace pd */
1249 if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1250 return 0;
1252 cup = alloc_cld_upcall(nn);
1253 if (!cup) {
1254 printk(KERN_ERR "NFSD: Unable to check client record on "
1255 "stable storage: %d\n", -ENOMEM);
1256 return -ENOMEM;
1259 cup->cu_u.cu_msg.cm_cmd = Cld_Check;
1260 cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1261 memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1262 clp->cl_name.len);
1264 ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1265 if (!ret) {
1266 ret = cup->cu_u.cu_msg.cm_status;
1267 set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1270 free_cld_upcall(cup);
1271 return ret;
1275 * For newer nfsdcld's that allow us to "slurp" the clients
1276 * from the tracking database during startup.
1278 * Check for presence of a record in the reclaim_str_hashtbl
1280 static int
1281 nfsd4_cld_check(struct nfs4_client *clp)
1283 struct nfs4_client_reclaim *crp;
1284 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1285 struct cld_net *cn = nn->cld_net;
1286 int status;
1287 char dname[HEXDIR_LEN];
1288 struct xdr_netobj name;
1290 /* did we already find that this client is stable? */
1291 if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1292 return 0;
1294 /* look for it in the reclaim hashtable otherwise */
1295 crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1296 if (crp)
1297 goto found;
1299 if (cn->cn_has_legacy) {
1300 status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1301 if (status)
1302 return -ENOENT;
1304 name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1305 if (!name.data) {
1306 dprintk("%s: failed to allocate memory for name.data!\n",
1307 __func__);
1308 return -ENOENT;
1310 name.len = HEXDIR_LEN;
1311 crp = nfsd4_find_reclaim_client(name, nn);
1312 kfree(name.data);
1313 if (crp)
1314 goto found;
1317 return -ENOENT;
1318 found:
1319 crp->cr_clp = clp;
1320 return 0;
1323 static int
1324 nfsd4_cld_check_v2(struct nfs4_client *clp)
1326 struct nfs4_client_reclaim *crp;
1327 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1328 struct cld_net *cn = nn->cld_net;
1329 int status;
1330 char dname[HEXDIR_LEN];
1331 struct xdr_netobj name;
1332 struct crypto_shash *tfm = cn->cn_tfm;
1333 struct xdr_netobj cksum;
1334 char *principal = NULL;
1336 /* did we already find that this client is stable? */
1337 if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1338 return 0;
1340 /* look for it in the reclaim hashtable otherwise */
1341 crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1342 if (crp)
1343 goto found;
1345 if (cn->cn_has_legacy) {
1346 status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1347 if (status)
1348 return -ENOENT;
1350 name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1351 if (!name.data) {
1352 dprintk("%s: failed to allocate memory for name.data\n",
1353 __func__);
1354 return -ENOENT;
1356 name.len = HEXDIR_LEN;
1357 crp = nfsd4_find_reclaim_client(name, nn);
1358 kfree(name.data);
1359 if (crp)
1360 goto found;
1363 return -ENOENT;
1364 found:
1365 if (crp->cr_princhash.len) {
1366 if (clp->cl_cred.cr_raw_principal)
1367 principal = clp->cl_cred.cr_raw_principal;
1368 else if (clp->cl_cred.cr_principal)
1369 principal = clp->cl_cred.cr_principal;
1370 if (principal == NULL)
1371 return -ENOENT;
1372 cksum.len = crypto_shash_digestsize(tfm);
1373 cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1374 if (cksum.data == NULL)
1375 return -ENOENT;
1376 status = crypto_shash_tfm_digest(tfm, principal,
1377 strlen(principal), cksum.data);
1378 if (status) {
1379 kfree(cksum.data);
1380 return -ENOENT;
1382 if (memcmp(crp->cr_princhash.data, cksum.data,
1383 crp->cr_princhash.len)) {
1384 kfree(cksum.data);
1385 return -ENOENT;
1387 kfree(cksum.data);
1389 crp->cr_clp = clp;
1390 return 0;
1393 static int
1394 nfsd4_cld_grace_start(struct nfsd_net *nn)
1396 int ret;
1397 struct cld_upcall *cup;
1398 struct cld_net *cn = nn->cld_net;
1400 cup = alloc_cld_upcall(nn);
1401 if (!cup) {
1402 ret = -ENOMEM;
1403 goto out_err;
1406 cup->cu_u.cu_msg.cm_cmd = Cld_GraceStart;
1407 ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1408 if (!ret)
1409 ret = cup->cu_u.cu_msg.cm_status;
1411 free_cld_upcall(cup);
1412 out_err:
1413 if (ret)
1414 dprintk("%s: Unable to get clients from userspace: %d\n",
1415 __func__, ret);
1416 return ret;
1419 /* For older nfsdcld's that need cm_gracetime */
1420 static void
1421 nfsd4_cld_grace_done_v0(struct nfsd_net *nn)
1423 int ret;
1424 struct cld_upcall *cup;
1425 struct cld_net *cn = nn->cld_net;
1427 cup = alloc_cld_upcall(nn);
1428 if (!cup) {
1429 ret = -ENOMEM;
1430 goto out_err;
1433 cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1434 cup->cu_u.cu_msg.cm_u.cm_gracetime = nn->boot_time;
1435 ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1436 if (!ret)
1437 ret = cup->cu_u.cu_msg.cm_status;
1439 free_cld_upcall(cup);
1440 out_err:
1441 if (ret)
1442 printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1446 * For newer nfsdcld's that do not need cm_gracetime. We also need to call
1447 * nfs4_release_reclaim() to clear out the reclaim_str_hashtbl.
1449 static void
1450 nfsd4_cld_grace_done(struct nfsd_net *nn)
1452 int ret;
1453 struct cld_upcall *cup;
1454 struct cld_net *cn = nn->cld_net;
1456 cup = alloc_cld_upcall(nn);
1457 if (!cup) {
1458 ret = -ENOMEM;
1459 goto out_err;
1462 cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1463 ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1464 if (!ret)
1465 ret = cup->cu_u.cu_msg.cm_status;
1467 free_cld_upcall(cup);
1468 out_err:
1469 nfs4_release_reclaim(nn);
1470 if (ret)
1471 printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1474 static int
1475 nfs4_cld_state_init(struct net *net)
1477 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1478 int i;
1480 nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
1481 sizeof(struct list_head),
1482 GFP_KERNEL);
1483 if (!nn->reclaim_str_hashtbl)
1484 return -ENOMEM;
1486 for (i = 0; i < CLIENT_HASH_SIZE; i++)
1487 INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
1488 nn->reclaim_str_hashtbl_size = 0;
1489 nn->track_reclaim_completes = true;
1490 atomic_set(&nn->nr_reclaim_complete, 0);
1492 return 0;
1495 static void
1496 nfs4_cld_state_shutdown(struct net *net)
1498 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1500 nn->track_reclaim_completes = false;
1501 kfree(nn->reclaim_str_hashtbl);
1504 static bool
1505 cld_running(struct nfsd_net *nn)
1507 struct cld_net *cn = nn->cld_net;
1508 struct rpc_pipe *pipe = cn->cn_pipe;
1510 return pipe->nreaders || pipe->nwriters;
1513 static int
1514 nfsd4_cld_get_version(struct nfsd_net *nn)
1516 int ret = 0;
1517 struct cld_upcall *cup;
1518 struct cld_net *cn = nn->cld_net;
1519 uint8_t version;
1521 cup = alloc_cld_upcall(nn);
1522 if (!cup) {
1523 ret = -ENOMEM;
1524 goto out_err;
1526 cup->cu_u.cu_msg.cm_cmd = Cld_GetVersion;
1527 ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1528 if (!ret) {
1529 ret = cup->cu_u.cu_msg.cm_status;
1530 if (ret)
1531 goto out_free;
1532 version = cup->cu_u.cu_msg.cm_u.cm_version;
1533 dprintk("%s: userspace returned version %u\n",
1534 __func__, version);
1535 if (version < 1)
1536 version = 1;
1537 else if (version > CLD_UPCALL_VERSION)
1538 version = CLD_UPCALL_VERSION;
1540 switch (version) {
1541 case 1:
1542 nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
1543 break;
1544 case 2:
1545 nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v2;
1546 break;
1547 default:
1548 break;
1551 out_free:
1552 free_cld_upcall(cup);
1553 out_err:
1554 if (ret)
1555 dprintk("%s: Unable to get version from userspace: %d\n",
1556 __func__, ret);
1557 return ret;
1560 static int
1561 nfsd4_cld_tracking_init(struct net *net)
1563 int status;
1564 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1565 bool running;
1566 int retries = 10;
1567 struct crypto_shash *tfm;
1569 status = nfs4_cld_state_init(net);
1570 if (status)
1571 return status;
1573 status = __nfsd4_init_cld_pipe(net);
1574 if (status)
1575 goto err_shutdown;
1578 * rpc pipe upcalls take 30 seconds to time out, so we don't want to
1579 * queue an upcall unless we know that nfsdcld is running (because we
1580 * want this to fail fast so that nfsd4_client_tracking_init() can try
1581 * the next client tracking method). nfsdcld should already be running
1582 * before nfsd is started, so the wait here is for nfsdcld to open the
1583 * pipefs file we just created.
1585 while (!(running = cld_running(nn)) && retries--)
1586 msleep(100);
1588 if (!running) {
1589 status = -ETIMEDOUT;
1590 goto err_remove;
1592 tfm = crypto_alloc_shash("sha256", 0, 0);
1593 if (IS_ERR(tfm)) {
1594 status = PTR_ERR(tfm);
1595 goto err_remove;
1597 nn->cld_net->cn_tfm = tfm;
1599 status = nfsd4_cld_get_version(nn);
1600 if (status == -EOPNOTSUPP)
1601 pr_warn("NFSD: nfsdcld GetVersion upcall failed. Please upgrade nfsdcld.\n");
1603 status = nfsd4_cld_grace_start(nn);
1604 if (status) {
1605 if (status == -EOPNOTSUPP)
1606 pr_warn("NFSD: nfsdcld GraceStart upcall failed. Please upgrade nfsdcld.\n");
1607 nfs4_release_reclaim(nn);
1608 goto err_remove;
1609 } else
1610 printk("NFSD: Using nfsdcld client tracking operations.\n");
1611 return 0;
1613 err_remove:
1614 nfsd4_remove_cld_pipe(net);
1615 err_shutdown:
1616 nfs4_cld_state_shutdown(net);
1617 return status;
1620 static void
1621 nfsd4_cld_tracking_exit(struct net *net)
1623 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1625 nfs4_release_reclaim(nn);
1626 nfsd4_remove_cld_pipe(net);
1627 nfs4_cld_state_shutdown(net);
1630 /* For older nfsdcld's */
1631 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v0 = {
1632 .init = nfsd4_init_cld_pipe,
1633 .exit = nfsd4_remove_cld_pipe,
1634 .create = nfsd4_cld_create,
1635 .remove = nfsd4_cld_remove,
1636 .check = nfsd4_cld_check_v0,
1637 .grace_done = nfsd4_cld_grace_done_v0,
1638 .version = 1,
1639 .msglen = sizeof(struct cld_msg),
1642 /* For newer nfsdcld's */
1643 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
1644 .init = nfsd4_cld_tracking_init,
1645 .exit = nfsd4_cld_tracking_exit,
1646 .create = nfsd4_cld_create,
1647 .remove = nfsd4_cld_remove,
1648 .check = nfsd4_cld_check,
1649 .grace_done = nfsd4_cld_grace_done,
1650 .version = 1,
1651 .msglen = sizeof(struct cld_msg),
1654 /* v2 create/check ops include the principal, if available */
1655 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2 = {
1656 .init = nfsd4_cld_tracking_init,
1657 .exit = nfsd4_cld_tracking_exit,
1658 .create = nfsd4_cld_create_v2,
1659 .remove = nfsd4_cld_remove,
1660 .check = nfsd4_cld_check_v2,
1661 .grace_done = nfsd4_cld_grace_done,
1662 .version = 2,
1663 .msglen = sizeof(struct cld_msg_v2),
1666 /* upcall via usermodehelper */
1667 static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
1668 module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
1669 S_IRUGO|S_IWUSR);
1670 MODULE_PARM_DESC(cltrack_prog, "Path to the nfsdcltrack upcall program");
1672 static bool cltrack_legacy_disable;
1673 module_param(cltrack_legacy_disable, bool, S_IRUGO|S_IWUSR);
1674 MODULE_PARM_DESC(cltrack_legacy_disable,
1675 "Disable legacy recoverydir conversion. Default: false");
1677 #define LEGACY_TOPDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_TOPDIR="
1678 #define LEGACY_RECDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_RECDIR="
1679 #define HAS_SESSION_ENV_PREFIX "NFSDCLTRACK_CLIENT_HAS_SESSION="
1680 #define GRACE_START_ENV_PREFIX "NFSDCLTRACK_GRACE_START="
1682 static char *
1683 nfsd4_cltrack_legacy_topdir(void)
1685 int copied;
1686 size_t len;
1687 char *result;
1689 if (cltrack_legacy_disable)
1690 return NULL;
1692 len = strlen(LEGACY_TOPDIR_ENV_PREFIX) +
1693 strlen(nfs4_recoverydir()) + 1;
1695 result = kmalloc(len, GFP_KERNEL);
1696 if (!result)
1697 return result;
1699 copied = snprintf(result, len, LEGACY_TOPDIR_ENV_PREFIX "%s",
1700 nfs4_recoverydir());
1701 if (copied >= len) {
1702 /* just return nothing if output was truncated */
1703 kfree(result);
1704 return NULL;
1707 return result;
1710 static char *
1711 nfsd4_cltrack_legacy_recdir(const struct xdr_netobj *name)
1713 int copied;
1714 size_t len;
1715 char *result;
1717 if (cltrack_legacy_disable)
1718 return NULL;
1720 /* +1 is for '/' between "topdir" and "recdir" */
1721 len = strlen(LEGACY_RECDIR_ENV_PREFIX) +
1722 strlen(nfs4_recoverydir()) + 1 + HEXDIR_LEN;
1724 result = kmalloc(len, GFP_KERNEL);
1725 if (!result)
1726 return result;
1728 copied = snprintf(result, len, LEGACY_RECDIR_ENV_PREFIX "%s/",
1729 nfs4_recoverydir());
1730 if (copied > (len - HEXDIR_LEN)) {
1731 /* just return nothing if output will be truncated */
1732 kfree(result);
1733 return NULL;
1736 copied = nfs4_make_rec_clidname(result + copied, name);
1737 if (copied) {
1738 kfree(result);
1739 return NULL;
1742 return result;
1745 static char *
1746 nfsd4_cltrack_client_has_session(struct nfs4_client *clp)
1748 int copied;
1749 size_t len;
1750 char *result;
1752 /* prefix + Y/N character + terminating NULL */
1753 len = strlen(HAS_SESSION_ENV_PREFIX) + 1 + 1;
1755 result = kmalloc(len, GFP_KERNEL);
1756 if (!result)
1757 return result;
1759 copied = snprintf(result, len, HAS_SESSION_ENV_PREFIX "%c",
1760 clp->cl_minorversion ? 'Y' : 'N');
1761 if (copied >= len) {
1762 /* just return nothing if output was truncated */
1763 kfree(result);
1764 return NULL;
1767 return result;
1770 static char *
1771 nfsd4_cltrack_grace_start(time64_t grace_start)
1773 int copied;
1774 size_t len;
1775 char *result;
1777 /* prefix + max width of int64_t string + terminating NULL */
1778 len = strlen(GRACE_START_ENV_PREFIX) + 22 + 1;
1780 result = kmalloc(len, GFP_KERNEL);
1781 if (!result)
1782 return result;
1784 copied = snprintf(result, len, GRACE_START_ENV_PREFIX "%lld",
1785 grace_start);
1786 if (copied >= len) {
1787 /* just return nothing if output was truncated */
1788 kfree(result);
1789 return NULL;
1792 return result;
1795 static int
1796 nfsd4_umh_cltrack_upcall(char *cmd, char *arg, char *env0, char *env1)
1798 char *envp[3];
1799 char *argv[4];
1800 int ret;
1802 if (unlikely(!cltrack_prog[0])) {
1803 dprintk("%s: cltrack_prog is disabled\n", __func__);
1804 return -EACCES;
1807 dprintk("%s: cmd: %s\n", __func__, cmd);
1808 dprintk("%s: arg: %s\n", __func__, arg ? arg : "(null)");
1809 dprintk("%s: env0: %s\n", __func__, env0 ? env0 : "(null)");
1810 dprintk("%s: env1: %s\n", __func__, env1 ? env1 : "(null)");
1812 envp[0] = env0;
1813 envp[1] = env1;
1814 envp[2] = NULL;
1816 argv[0] = (char *)cltrack_prog;
1817 argv[1] = cmd;
1818 argv[2] = arg;
1819 argv[3] = NULL;
1821 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1823 * Disable the upcall mechanism if we're getting an ENOENT or EACCES
1824 * error. The admin can re-enable it on the fly by using sysfs
1825 * once the problem has been fixed.
1827 if (ret == -ENOENT || ret == -EACCES) {
1828 dprintk("NFSD: %s was not found or isn't executable (%d). "
1829 "Setting cltrack_prog to blank string!",
1830 cltrack_prog, ret);
1831 cltrack_prog[0] = '\0';
1833 dprintk("%s: %s return value: %d\n", __func__, cltrack_prog, ret);
1835 return ret;
1838 static char *
1839 bin_to_hex_dup(const unsigned char *src, int srclen)
1841 char *buf;
1843 /* +1 for terminating NULL */
1844 buf = kzalloc((srclen * 2) + 1, GFP_KERNEL);
1845 if (!buf)
1846 return buf;
1848 bin2hex(buf, src, srclen);
1849 return buf;
1852 static int
1853 nfsd4_umh_cltrack_init(struct net *net)
1855 int ret;
1856 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1857 char *grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1859 /* XXX: The usermode helper s not working in container yet. */
1860 if (net != &init_net) {
1861 pr_warn("NFSD: attempt to initialize umh client tracking in a container ignored.\n");
1862 kfree(grace_start);
1863 return -EINVAL;
1866 ret = nfsd4_umh_cltrack_upcall("init", NULL, grace_start, NULL);
1867 kfree(grace_start);
1868 if (!ret)
1869 printk("NFSD: Using UMH upcall client tracking operations.\n");
1870 return ret;
1873 static void
1874 nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)
1876 wait_on_bit_lock(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK,
1877 TASK_UNINTERRUPTIBLE);
1880 static void
1881 nfsd4_cltrack_upcall_unlock(struct nfs4_client *clp)
1883 smp_mb__before_atomic();
1884 clear_bit(NFSD4_CLIENT_UPCALL_LOCK, &clp->cl_flags);
1885 smp_mb__after_atomic();
1886 wake_up_bit(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK);
1889 static void
1890 nfsd4_umh_cltrack_create(struct nfs4_client *clp)
1892 char *hexid, *has_session, *grace_start;
1893 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1896 * With v4.0 clients, there's little difference in outcome between a
1897 * create and check operation, and we can end up calling into this
1898 * function multiple times per client (once for each openowner). So,
1899 * for v4.0 clients skip upcalling once the client has been recorded
1900 * on stable storage.
1902 * For v4.1+ clients, the outcome of the two operations is different,
1903 * so we must ensure that we upcall for the create operation. v4.1+
1904 * clients call this on RECLAIM_COMPLETE though, so we should only end
1905 * up doing a single create upcall per client.
1907 if (clp->cl_minorversion == 0 &&
1908 test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1909 return;
1911 hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1912 if (!hexid) {
1913 dprintk("%s: can't allocate memory for upcall!\n", __func__);
1914 return;
1917 has_session = nfsd4_cltrack_client_has_session(clp);
1918 grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1920 nfsd4_cltrack_upcall_lock(clp);
1921 if (!nfsd4_umh_cltrack_upcall("create", hexid, has_session, grace_start))
1922 set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1923 nfsd4_cltrack_upcall_unlock(clp);
1925 kfree(has_session);
1926 kfree(grace_start);
1927 kfree(hexid);
1930 static void
1931 nfsd4_umh_cltrack_remove(struct nfs4_client *clp)
1933 char *hexid;
1935 if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1936 return;
1938 hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1939 if (!hexid) {
1940 dprintk("%s: can't allocate memory for upcall!\n", __func__);
1941 return;
1944 nfsd4_cltrack_upcall_lock(clp);
1945 if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags) &&
1946 nfsd4_umh_cltrack_upcall("remove", hexid, NULL, NULL) == 0)
1947 clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1948 nfsd4_cltrack_upcall_unlock(clp);
1950 kfree(hexid);
1953 static int
1954 nfsd4_umh_cltrack_check(struct nfs4_client *clp)
1956 int ret;
1957 char *hexid, *has_session, *legacy;
1959 if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1960 return 0;
1962 hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1963 if (!hexid) {
1964 dprintk("%s: can't allocate memory for upcall!\n", __func__);
1965 return -ENOMEM;
1968 has_session = nfsd4_cltrack_client_has_session(clp);
1969 legacy = nfsd4_cltrack_legacy_recdir(&clp->cl_name);
1971 nfsd4_cltrack_upcall_lock(clp);
1972 if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags)) {
1973 ret = 0;
1974 } else {
1975 ret = nfsd4_umh_cltrack_upcall("check", hexid, has_session, legacy);
1976 if (ret == 0)
1977 set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1979 nfsd4_cltrack_upcall_unlock(clp);
1980 kfree(has_session);
1981 kfree(legacy);
1982 kfree(hexid);
1984 return ret;
1987 static void
1988 nfsd4_umh_cltrack_grace_done(struct nfsd_net *nn)
1990 char *legacy;
1991 char timestr[22]; /* FIXME: better way to determine max size? */
1993 sprintf(timestr, "%lld", nn->boot_time);
1994 legacy = nfsd4_cltrack_legacy_topdir();
1995 nfsd4_umh_cltrack_upcall("gracedone", timestr, legacy, NULL);
1996 kfree(legacy);
1999 static const struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
2000 .init = nfsd4_umh_cltrack_init,
2001 .exit = NULL,
2002 .create = nfsd4_umh_cltrack_create,
2003 .remove = nfsd4_umh_cltrack_remove,
2004 .check = nfsd4_umh_cltrack_check,
2005 .grace_done = nfsd4_umh_cltrack_grace_done,
2006 .version = 1,
2007 .msglen = 0,
2011 nfsd4_client_tracking_init(struct net *net)
2013 int status;
2014 struct path path;
2015 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2017 /* just run the init if it the method is already decided */
2018 if (nn->client_tracking_ops)
2019 goto do_init;
2021 /* First, try to use nfsdcld */
2022 nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
2023 status = nn->client_tracking_ops->init(net);
2024 if (!status)
2025 return status;
2026 if (status != -ETIMEDOUT) {
2027 nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v0;
2028 status = nn->client_tracking_ops->init(net);
2029 if (!status)
2030 return status;
2034 * Next, try the UMH upcall.
2036 nn->client_tracking_ops = &nfsd4_umh_tracking_ops;
2037 status = nn->client_tracking_ops->init(net);
2038 if (!status)
2039 return status;
2042 * Finally, See if the recoverydir exists and is a directory.
2043 * If it is, then use the legacy ops.
2045 nn->client_tracking_ops = &nfsd4_legacy_tracking_ops;
2046 status = kern_path(nfs4_recoverydir(), LOOKUP_FOLLOW, &path);
2047 if (!status) {
2048 status = d_is_dir(path.dentry);
2049 path_put(&path);
2050 if (!status) {
2051 status = -EINVAL;
2052 goto out;
2056 do_init:
2057 status = nn->client_tracking_ops->init(net);
2058 out:
2059 if (status) {
2060 printk(KERN_WARNING "NFSD: Unable to initialize client "
2061 "recovery tracking! (%d)\n", status);
2062 nn->client_tracking_ops = NULL;
2064 return status;
2067 void
2068 nfsd4_client_tracking_exit(struct net *net)
2070 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2072 if (nn->client_tracking_ops) {
2073 if (nn->client_tracking_ops->exit)
2074 nn->client_tracking_ops->exit(net);
2075 nn->client_tracking_ops = NULL;
2079 void
2080 nfsd4_client_record_create(struct nfs4_client *clp)
2082 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2084 if (nn->client_tracking_ops)
2085 nn->client_tracking_ops->create(clp);
2088 void
2089 nfsd4_client_record_remove(struct nfs4_client *clp)
2091 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2093 if (nn->client_tracking_ops)
2094 nn->client_tracking_ops->remove(clp);
2098 nfsd4_client_record_check(struct nfs4_client *clp)
2100 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2102 if (nn->client_tracking_ops)
2103 return nn->client_tracking_ops->check(clp);
2105 return -EOPNOTSUPP;
2108 void
2109 nfsd4_record_grace_done(struct nfsd_net *nn)
2111 if (nn->client_tracking_ops)
2112 nn->client_tracking_ops->grace_done(nn);
2115 static int
2116 rpc_pipefs_event(struct notifier_block *nb, unsigned long event, void *ptr)
2118 struct super_block *sb = ptr;
2119 struct net *net = sb->s_fs_info;
2120 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2121 struct cld_net *cn = nn->cld_net;
2122 struct dentry *dentry;
2123 int ret = 0;
2125 if (!try_module_get(THIS_MODULE))
2126 return 0;
2128 if (!cn) {
2129 module_put(THIS_MODULE);
2130 return 0;
2133 switch (event) {
2134 case RPC_PIPEFS_MOUNT:
2135 dentry = nfsd4_cld_register_sb(sb, cn->cn_pipe);
2136 if (IS_ERR(dentry)) {
2137 ret = PTR_ERR(dentry);
2138 break;
2140 cn->cn_pipe->dentry = dentry;
2141 break;
2142 case RPC_PIPEFS_UMOUNT:
2143 if (cn->cn_pipe->dentry)
2144 nfsd4_cld_unregister_sb(cn->cn_pipe);
2145 break;
2146 default:
2147 ret = -ENOTSUPP;
2148 break;
2150 module_put(THIS_MODULE);
2151 return ret;
2154 static struct notifier_block nfsd4_cld_block = {
2155 .notifier_call = rpc_pipefs_event,
2159 register_cld_notifier(void)
2161 return rpc_pipefs_notifier_register(&nfsd4_cld_block);
2164 void
2165 unregister_cld_notifier(void)
2167 rpc_pipefs_notifier_unregister(&nfsd4_cld_block);