2 * Device operations for the pnfs nfs4 file layout driver.
5 * The Regents of the University of Michigan
8 * Dean Hildebrand <dhildebz@umich.edu>
9 * Garth Goodson <Garth.Goodson@netapp.com>
11 * Permission is granted to use, copy, create derivative works, and
12 * redistribute this software and such derivative works for any purpose,
13 * so long as the name of the University of Michigan is not used in
14 * any advertising or publicity pertaining to the use or distribution
15 * of this software without specific, written prior authorization. If
16 * the above copyright notice or any other identification of the
17 * University of Michigan is included in any copy of any portion of
18 * this software, then the disclaimer below must also be included.
20 * This software is provided as is, without representation or warranty
21 * of any kind either express or implied, including without limitation
22 * the implied warranties of merchantability, fitness for a particular
23 * purpose, or noninfringement. The Regents of the University of
24 * Michigan shall not be liable for any damages, including special,
25 * indirect, incidental, or consequential damages, with respect to any
26 * claim arising out of or in connection with the use of the software,
27 * even if it has been or is hereafter advised of the possibility of
31 #include <linux/nfs_fs.h>
32 #include <linux/vmalloc.h>
33 #include <linux/module.h>
34 #include <linux/sunrpc/addr.h>
37 #include "nfs4session.h"
38 #include "nfs4filelayout.h"
40 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
42 static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
43 static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
48 * Data servers can be mapped to different device ids.
49 * nfs4_pnfs_ds reference counting
50 * - set to 1 on allocation
51 * - incremented when a device id maps a data server already in the cache.
52 * - decremented when deviceid is removed from the cache.
54 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
55 static LIST_HEAD(nfs4_data_server_cache);
59 print_ds(struct nfs4_pnfs_ds *ds)
62 printk("%s NULL device\n", __func__);
68 " cl_exchange_flags %x\n",
70 atomic_read(&ds->ds_count), ds->ds_clp,
71 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
75 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
77 struct sockaddr_in *a, *b;
78 struct sockaddr_in6 *a6, *b6;
80 if (addr1->sa_family != addr2->sa_family)
83 switch (addr1->sa_family) {
85 a = (struct sockaddr_in *)addr1;
86 b = (struct sockaddr_in *)addr2;
88 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
89 a->sin_port == b->sin_port)
94 a6 = (struct sockaddr_in6 *)addr1;
95 b6 = (struct sockaddr_in6 *)addr2;
97 /* LINKLOCAL addresses must have matching scope_id */
98 if (ipv6_addr_scope(&a6->sin6_addr) ==
99 IPV6_ADDR_SCOPE_LINKLOCAL &&
100 a6->sin6_scope_id != b6->sin6_scope_id)
103 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
104 a6->sin6_port == b6->sin6_port)
109 dprintk("%s: unhandled address family: %u\n",
110 __func__, addr1->sa_family);
118 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
119 const struct list_head *dsaddrs2)
121 struct nfs4_pnfs_ds_addr *da1, *da2;
123 /* step through both lists, comparing as we go */
124 for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node),
125 da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node);
126 da1 != NULL && da2 != NULL;
127 da1 = list_entry(da1->da_node.next, typeof(*da1), da_node),
128 da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) {
129 if (!same_sockaddr((struct sockaddr *)&da1->da_addr,
130 (struct sockaddr *)&da2->da_addr))
133 if (da1 == NULL && da2 == NULL)
140 * Lookup DS by addresses. nfs4_ds_cache_lock is held
142 static struct nfs4_pnfs_ds *
143 _data_server_lookup_locked(const struct list_head *dsaddrs)
145 struct nfs4_pnfs_ds *ds;
147 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
148 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
154 * Create an rpc connection to the nfs4_pnfs_ds data server
155 * Currently only supports IPv4 and IPv6 addresses
158 nfs4_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds)
160 struct nfs_client *clp = ERR_PTR(-EIO);
161 struct nfs4_pnfs_ds_addr *da;
164 dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr,
165 mds_srv->nfs_client->cl_rpcclient->cl_auth->au_flavor);
167 list_for_each_entry(da, &ds->ds_addrs, da_node) {
168 dprintk("%s: DS %s: trying address %s\n",
169 __func__, ds->ds_remotestr, da->da_remotestr);
171 clp = nfs4_set_ds_client(mds_srv->nfs_client,
172 (struct sockaddr *)&da->da_addr,
173 da->da_addrlen, IPPROTO_TCP,
174 dataserver_timeo, dataserver_retrans);
180 status = PTR_ERR(clp);
184 status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time);
190 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
199 destroy_ds(struct nfs4_pnfs_ds *ds)
201 struct nfs4_pnfs_ds_addr *da;
203 dprintk("--> %s\n", __func__);
208 nfs_put_client(ds->ds_clp);
210 while (!list_empty(&ds->ds_addrs)) {
211 da = list_first_entry(&ds->ds_addrs,
212 struct nfs4_pnfs_ds_addr,
214 list_del_init(&da->da_node);
215 kfree(da->da_remotestr);
219 kfree(ds->ds_remotestr);
224 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
226 struct nfs4_pnfs_ds *ds;
229 nfs4_print_deviceid(&dsaddr->id_node.deviceid);
231 for (i = 0; i < dsaddr->ds_num; i++) {
232 ds = dsaddr->ds_list[i];
234 if (atomic_dec_and_lock(&ds->ds_count,
235 &nfs4_ds_cache_lock)) {
236 list_del_init(&ds->ds_node);
237 spin_unlock(&nfs4_ds_cache_lock);
242 kfree(dsaddr->stripe_indices);
247 * Create a string with a human readable address and port to avoid
248 * complicated setup around many dprinks.
251 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
253 struct nfs4_pnfs_ds_addr *da;
258 len = 3; /* '{', '}' and eol */
259 list_for_each_entry(da, dsaddrs, da_node) {
260 len += strlen(da->da_remotestr) + 1; /* string plus comma */
263 remotestr = kzalloc(len, gfp_flags);
270 list_for_each_entry(da, dsaddrs, da_node) {
271 size_t ll = strlen(da->da_remotestr);
276 memcpy(p, da->da_remotestr, ll);
295 static struct nfs4_pnfs_ds *
296 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
298 struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
301 if (list_empty(dsaddrs)) {
302 dprintk("%s: no addresses defined\n", __func__);
306 ds = kzalloc(sizeof(*ds), gfp_flags);
310 /* this is only used for debugging, so it's ok if its NULL */
311 remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
313 spin_lock(&nfs4_ds_cache_lock);
314 tmp_ds = _data_server_lookup_locked(dsaddrs);
315 if (tmp_ds == NULL) {
316 INIT_LIST_HEAD(&ds->ds_addrs);
317 list_splice_init(dsaddrs, &ds->ds_addrs);
318 ds->ds_remotestr = remotestr;
319 atomic_set(&ds->ds_count, 1);
320 INIT_LIST_HEAD(&ds->ds_node);
322 list_add(&ds->ds_node, &nfs4_data_server_cache);
323 dprintk("%s add new data server %s\n", __func__,
328 atomic_inc(&tmp_ds->ds_count);
329 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
330 __func__, tmp_ds->ds_remotestr,
331 atomic_read(&tmp_ds->ds_count));
334 spin_unlock(&nfs4_ds_cache_lock);
340 * Currently only supports ipv4, ipv6 and one multi-path address.
342 static struct nfs4_pnfs_ds_addr *
343 decode_ds_addr(struct net *net, struct xdr_stream *streamp, gfp_t gfp_flags)
345 struct nfs4_pnfs_ds_addr *da = NULL;
351 char *netid, *match_netid;
352 size_t len, match_netid_len;
358 p = xdr_inline_decode(streamp, 4);
361 nlen = be32_to_cpup(p++);
363 p = xdr_inline_decode(streamp, nlen);
367 netid = kmalloc(nlen+1, gfp_flags);
368 if (unlikely(!netid))
372 memcpy(netid, p, nlen);
374 /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
375 p = xdr_inline_decode(streamp, 4);
378 rlen = be32_to_cpup(p);
380 p = xdr_inline_decode(streamp, rlen);
384 /* port is ".ABC.DEF", 8 chars max */
385 if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
386 dprintk("%s: Invalid address, length %d\n", __func__,
390 buf = kmalloc(rlen + 1, gfp_flags);
392 dprintk("%s: Not enough memory\n", __func__);
396 memcpy(buf, p, rlen);
398 /* replace port '.' with '-' */
399 portstr = strrchr(buf, '.');
401 dprintk("%s: Failed finding expected dot in port\n",
407 /* find '.' between address and port */
408 portstr = strrchr(buf, '.');
410 dprintk("%s: Failed finding expected dot between address and "
416 da = kzalloc(sizeof(*da), gfp_flags);
420 INIT_LIST_HEAD(&da->da_node);
422 if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
423 sizeof(da->da_addr))) {
424 dprintk("%s: error parsing address %s\n", __func__, buf);
429 sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
430 port = htons((tmp[0] << 8) | (tmp[1]));
432 switch (da->da_addr.ss_family) {
434 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
435 da->da_addrlen = sizeof(struct sockaddr_in);
441 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
442 da->da_addrlen = sizeof(struct sockaddr_in6);
443 match_netid = "tcp6";
450 dprintk("%s: unsupported address family: %u\n",
451 __func__, da->da_addr.ss_family);
455 if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
456 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
457 __func__, netid, match_netid);
461 /* save human readable address */
462 len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
463 da->da_remotestr = kzalloc(len, gfp_flags);
465 /* NULL is ok, only used for dprintk */
466 if (da->da_remotestr)
467 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
468 buf, endsep, ntohs(port));
470 dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
478 dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
486 /* Decode opaque device data and return the result */
487 static struct nfs4_file_layout_dsaddr*
488 decode_device(struct inode *ino, struct pnfs_device *pdev, gfp_t gfp_flags)
496 struct nfs4_file_layout_dsaddr *dsaddr = NULL;
497 struct xdr_stream stream;
499 struct page *scratch;
500 struct list_head dsaddrs;
501 struct nfs4_pnfs_ds_addr *da;
503 /* set up xdr stream */
504 scratch = alloc_page(gfp_flags);
508 xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
509 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
511 /* Get the stripe count (number of stripe index) */
512 p = xdr_inline_decode(&stream, 4);
514 goto out_err_free_scratch;
516 cnt = be32_to_cpup(p);
517 dprintk("%s stripe count %d\n", __func__, cnt);
518 if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
519 printk(KERN_WARNING "NFS: %s: stripe count %d greater than "
520 "supported maximum %d\n", __func__,
521 cnt, NFS4_PNFS_MAX_STRIPE_CNT);
522 goto out_err_free_scratch;
525 /* read stripe indices */
526 stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
528 goto out_err_free_scratch;
530 p = xdr_inline_decode(&stream, cnt << 2);
532 goto out_err_free_stripe_indices;
534 indexp = &stripe_indices[0];
535 max_stripe_index = 0;
536 for (i = 0; i < cnt; i++) {
537 *indexp = be32_to_cpup(p++);
538 max_stripe_index = max(max_stripe_index, *indexp);
542 /* Check the multipath list count */
543 p = xdr_inline_decode(&stream, 4);
545 goto out_err_free_stripe_indices;
547 num = be32_to_cpup(p);
548 dprintk("%s ds_num %u\n", __func__, num);
549 if (num > NFS4_PNFS_MAX_MULTI_CNT) {
550 printk(KERN_WARNING "NFS: %s: multipath count %d greater than "
551 "supported maximum %d\n", __func__,
552 num, NFS4_PNFS_MAX_MULTI_CNT);
553 goto out_err_free_stripe_indices;
556 /* validate stripe indices are all < num */
557 if (max_stripe_index >= num) {
558 printk(KERN_WARNING "NFS: %s: stripe index %u >= num ds %u\n",
559 __func__, max_stripe_index, num);
560 goto out_err_free_stripe_indices;
563 dsaddr = kzalloc(sizeof(*dsaddr) +
564 (sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
567 goto out_err_free_stripe_indices;
569 dsaddr->stripe_count = cnt;
570 dsaddr->stripe_indices = stripe_indices;
571 stripe_indices = NULL;
572 dsaddr->ds_num = num;
573 nfs4_init_deviceid_node(&dsaddr->id_node,
574 NFS_SERVER(ino)->pnfs_curr_ld,
575 NFS_SERVER(ino)->nfs_client,
578 INIT_LIST_HEAD(&dsaddrs);
580 for (i = 0; i < dsaddr->ds_num; i++) {
584 p = xdr_inline_decode(&stream, 4);
586 goto out_err_free_deviceid;
588 mp_count = be32_to_cpup(p); /* multipath count */
589 for (j = 0; j < mp_count; j++) {
590 da = decode_ds_addr(NFS_SERVER(ino)->nfs_client->cl_net,
593 list_add_tail(&da->da_node, &dsaddrs);
595 if (list_empty(&dsaddrs)) {
596 dprintk("%s: no suitable DS addresses found\n",
598 goto out_err_free_deviceid;
601 dsaddr->ds_list[i] = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
602 if (!dsaddr->ds_list[i])
603 goto out_err_drain_dsaddrs;
605 /* If DS was already in cache, free ds addrs */
606 while (!list_empty(&dsaddrs)) {
607 da = list_first_entry(&dsaddrs,
608 struct nfs4_pnfs_ds_addr,
610 list_del_init(&da->da_node);
611 kfree(da->da_remotestr);
616 __free_page(scratch);
619 out_err_drain_dsaddrs:
620 while (!list_empty(&dsaddrs)) {
621 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
623 list_del_init(&da->da_node);
624 kfree(da->da_remotestr);
627 out_err_free_deviceid:
628 nfs4_fl_free_deviceid(dsaddr);
629 /* stripe_indicies was part of dsaddr */
630 goto out_err_free_scratch;
631 out_err_free_stripe_indices:
632 kfree(stripe_indices);
633 out_err_free_scratch:
634 __free_page(scratch);
636 dprintk("%s ERROR: returning NULL\n", __func__);
641 * Decode the opaque device specified in 'dev' and add it to the cache of
644 static struct nfs4_file_layout_dsaddr *
645 decode_and_add_device(struct inode *inode, struct pnfs_device *dev, gfp_t gfp_flags)
647 struct nfs4_deviceid_node *d;
648 struct nfs4_file_layout_dsaddr *n, *new;
650 new = decode_device(inode, dev, gfp_flags);
652 printk(KERN_WARNING "NFS: %s: Could not decode or add device\n",
657 d = nfs4_insert_deviceid_node(&new->id_node);
658 n = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
660 nfs4_fl_free_deviceid(new);
668 * Retrieve the information for dev_id, add it to the list
669 * of available devices, and return it.
671 struct nfs4_file_layout_dsaddr *
672 filelayout_get_device_info(struct inode *inode,
673 struct nfs4_deviceid *dev_id,
674 struct rpc_cred *cred,
677 struct pnfs_device *pdev = NULL;
680 struct page **pages = NULL;
681 struct nfs4_file_layout_dsaddr *dsaddr = NULL;
683 struct nfs_server *server = NFS_SERVER(inode);
686 * Use the session max response size as the basis for setting
687 * GETDEVICEINFO's maxcount
689 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
690 max_pages = nfs_page_array_len(0, max_resp_sz);
691 dprintk("%s inode %p max_resp_sz %u max_pages %d\n",
692 __func__, inode, max_resp_sz, max_pages);
694 pdev = kzalloc(sizeof(struct pnfs_device), gfp_flags);
698 pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags);
703 for (i = 0; i < max_pages; i++) {
704 pages[i] = alloc_page(gfp_flags);
709 memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id));
710 pdev->layout_type = LAYOUT_NFSV4_1_FILES;
713 pdev->pglen = max_resp_sz;
715 pdev->maxcount = max_resp_sz - nfs41_maxgetdevinfo_overhead;
717 rc = nfs4_proc_getdeviceinfo(server, pdev, cred);
718 dprintk("%s getdevice info returns %d\n", __func__, rc);
723 * Found new device, need to decode it and then add it to the
724 * list of known devices for this mountpoint.
726 dsaddr = decode_and_add_device(inode, pdev, gfp_flags);
728 for (i = 0; i < max_pages; i++)
729 __free_page(pages[i]);
732 dprintk("<-- %s dsaddr %p\n", __func__, dsaddr);
737 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
739 nfs4_put_deviceid_node(&dsaddr->id_node);
743 * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
744 * Then: ((res + fsi) % dsaddr->stripe_count)
747 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
749 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
752 tmp = offset - flseg->pattern_offset;
753 do_div(tmp, flseg->stripe_unit);
754 tmp += flseg->first_stripe_index;
755 return do_div(tmp, flseg->dsaddr->stripe_count);
759 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
761 return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
765 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
767 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
770 if (flseg->stripe_type == STRIPE_SPARSE) {
771 if (flseg->num_fh == 1)
773 else if (flseg->num_fh == 0)
774 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
777 i = nfs4_fl_calc_ds_index(lseg, j);
780 return flseg->fh_array[i];
783 static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
786 wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING,
787 nfs_wait_bit_killable, TASK_KILLABLE);
790 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
792 smp_mb__before_clear_bit();
793 clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
794 smp_mb__after_clear_bit();
795 wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
799 struct nfs4_pnfs_ds *
800 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
802 struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
803 struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
804 struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
805 struct nfs4_pnfs_ds *ret = ds;
808 printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
810 filelayout_mark_devid_invalid(devid);
817 if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
818 struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
821 err = nfs4_ds_connect(s, ds);
823 nfs4_mark_deviceid_unavailable(devid);
824 nfs4_clear_ds_conn_bit(ds);
826 /* Either ds is connected, or ds is NULL */
827 nfs4_wait_ds_connect(ds);
830 if (filelayout_test_devid_unavailable(devid))
836 module_param(dataserver_retrans, uint, 0644);
837 MODULE_PARM_DESC(dataserver_retrans, "The number of times the NFSv4.1 client "
838 "retries a request before it attempts further "
839 " recovery action.");
840 module_param(dataserver_timeo, uint, 0644);
841 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
842 "NFSv4.1 client waits for a response from a "
843 " data server before it retries an NFS request.");