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>
35 #include "nfs4filelayout.h"
37 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
42 * Data servers can be mapped to different device ids.
43 * nfs4_pnfs_ds reference counting
44 * - set to 1 on allocation
45 * - incremented when a device id maps a data server already in the cache.
46 * - decremented when deviceid is removed from the cache.
48 DEFINE_SPINLOCK(nfs4_ds_cache_lock);
49 static LIST_HEAD(nfs4_data_server_cache);
53 print_ds(struct nfs4_pnfs_ds *ds)
56 printk("%s NULL device\n", __func__);
62 " cl_exchange_flags %x\n",
64 atomic_read(&ds->ds_count), ds->ds_clp,
65 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
69 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
71 struct sockaddr_in *a, *b;
72 struct sockaddr_in6 *a6, *b6;
74 if (addr1->sa_family != addr2->sa_family)
77 switch (addr1->sa_family) {
79 a = (struct sockaddr_in *)addr1;
80 b = (struct sockaddr_in *)addr2;
82 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
83 a->sin_port == b->sin_port)
88 a6 = (struct sockaddr_in6 *)addr1;
89 b6 = (struct sockaddr_in6 *)addr2;
91 /* LINKLOCAL addresses must have matching scope_id */
92 if (ipv6_addr_scope(&a6->sin6_addr) ==
93 IPV6_ADDR_SCOPE_LINKLOCAL &&
94 a6->sin6_scope_id != b6->sin6_scope_id)
97 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
98 a6->sin6_port == b6->sin6_port)
103 dprintk("%s: unhandled address family: %u\n",
104 __func__, addr1->sa_family);
112 * Lookup DS by addresses. The first matching address returns true.
113 * nfs4_ds_cache_lock is held
115 static struct nfs4_pnfs_ds *
116 _data_server_lookup_locked(struct list_head *dsaddrs)
118 struct nfs4_pnfs_ds *ds;
119 struct nfs4_pnfs_ds_addr *da1, *da2;
121 list_for_each_entry(da1, dsaddrs, da_node) {
122 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node) {
123 list_for_each_entry(da2, &ds->ds_addrs, da_node) {
125 (struct sockaddr *)&da1->da_addr,
126 (struct sockaddr *)&da2->da_addr))
135 * Compare two lists of addresses.
138 _data_server_match_all_addrs_locked(struct list_head *dsaddrs1,
139 struct list_head *dsaddrs2)
141 struct nfs4_pnfs_ds_addr *da1, *da2;
145 list_for_each_entry(da1, dsaddrs1, da_node)
148 list_for_each_entry(da2, dsaddrs2, da_node) {
151 list_for_each_entry(da1, dsaddrs1, da_node) {
152 if (same_sockaddr((struct sockaddr *)&da1->da_addr,
153 (struct sockaddr *)&da2->da_addr)) {
162 return (count1 == count2);
166 * Create an rpc connection to the nfs4_pnfs_ds data server
167 * Currently only supports IPv4 and IPv6 addresses
170 nfs4_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds)
172 struct nfs_client *clp = ERR_PTR(-EIO);
173 struct nfs4_pnfs_ds_addr *da;
176 dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr,
177 mds_srv->nfs_client->cl_rpcclient->cl_auth->au_flavor);
179 BUG_ON(list_empty(&ds->ds_addrs));
181 list_for_each_entry(da, &ds->ds_addrs, da_node) {
182 dprintk("%s: DS %s: trying address %s\n",
183 __func__, ds->ds_remotestr, da->da_remotestr);
185 clp = nfs4_set_ds_client(mds_srv->nfs_client,
186 (struct sockaddr *)&da->da_addr,
187 da->da_addrlen, IPPROTO_TCP);
193 status = PTR_ERR(clp);
197 if ((clp->cl_exchange_flags & EXCHGID4_FLAG_MASK_PNFS) != 0) {
198 if (!is_ds_client(clp)) {
203 dprintk("%s [existing] server=%s\n", __func__,
209 * Do not set NFS_CS_CHECK_LEASE_TIME instead set the DS lease to
210 * be equal to the MDS lease. Renewal is scheduled in create_session.
212 spin_lock(&mds_srv->nfs_client->cl_lock);
213 clp->cl_lease_time = mds_srv->nfs_client->cl_lease_time;
214 spin_unlock(&mds_srv->nfs_client->cl_lock);
215 clp->cl_last_renewal = jiffies;
218 status = nfs4_init_ds_session(clp);
223 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
232 destroy_ds(struct nfs4_pnfs_ds *ds)
234 struct nfs4_pnfs_ds_addr *da;
236 dprintk("--> %s\n", __func__);
241 nfs_put_client(ds->ds_clp);
243 while (!list_empty(&ds->ds_addrs)) {
244 da = list_first_entry(&ds->ds_addrs,
245 struct nfs4_pnfs_ds_addr,
247 list_del_init(&da->da_node);
248 kfree(da->da_remotestr);
252 kfree(ds->ds_remotestr);
257 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
259 struct nfs4_pnfs_ds *ds;
262 nfs4_print_deviceid(&dsaddr->id_node.deviceid);
264 for (i = 0; i < dsaddr->ds_num; i++) {
265 ds = dsaddr->ds_list[i];
267 if (atomic_dec_and_lock(&ds->ds_count,
268 &nfs4_ds_cache_lock)) {
269 list_del_init(&ds->ds_node);
270 spin_unlock(&nfs4_ds_cache_lock);
275 kfree(dsaddr->stripe_indices);
280 * Create a string with a human readable address and port to avoid
281 * complicated setup around many dprinks.
284 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
286 struct nfs4_pnfs_ds_addr *da;
291 len = 3; /* '{', '}' and eol */
292 list_for_each_entry(da, dsaddrs, da_node) {
293 len += strlen(da->da_remotestr) + 1; /* string plus comma */
296 remotestr = kzalloc(len, gfp_flags);
303 list_for_each_entry(da, dsaddrs, da_node) {
304 size_t ll = strlen(da->da_remotestr);
309 memcpy(p, da->da_remotestr, ll);
328 static struct nfs4_pnfs_ds *
329 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
331 struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
334 if (list_empty(dsaddrs)) {
335 dprintk("%s: no addresses defined\n", __func__);
339 ds = kzalloc(sizeof(*ds), gfp_flags);
343 /* this is only used for debugging, so it's ok if its NULL */
344 remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
346 spin_lock(&nfs4_ds_cache_lock);
347 tmp_ds = _data_server_lookup_locked(dsaddrs);
348 if (tmp_ds == NULL) {
349 INIT_LIST_HEAD(&ds->ds_addrs);
350 list_splice_init(dsaddrs, &ds->ds_addrs);
351 ds->ds_remotestr = remotestr;
352 atomic_set(&ds->ds_count, 1);
353 INIT_LIST_HEAD(&ds->ds_node);
355 list_add(&ds->ds_node, &nfs4_data_server_cache);
356 dprintk("%s add new data server %s\n", __func__,
359 if (!_data_server_match_all_addrs_locked(&tmp_ds->ds_addrs,
361 dprintk("%s: multipath address mismatch: %s != %s",
362 __func__, tmp_ds->ds_remotestr, remotestr);
366 atomic_inc(&tmp_ds->ds_count);
367 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
368 __func__, tmp_ds->ds_remotestr,
369 atomic_read(&tmp_ds->ds_count));
372 spin_unlock(&nfs4_ds_cache_lock);
378 * Currently only supports ipv4, ipv6 and one multi-path address.
380 static struct nfs4_pnfs_ds_addr *
381 decode_ds_addr(struct xdr_stream *streamp, gfp_t gfp_flags)
383 struct nfs4_pnfs_ds_addr *da = NULL;
389 char *netid, *match_netid;
390 size_t len, match_netid_len;
396 p = xdr_inline_decode(streamp, 4);
399 nlen = be32_to_cpup(p++);
401 p = xdr_inline_decode(streamp, nlen);
405 netid = kmalloc(nlen+1, gfp_flags);
406 if (unlikely(!netid))
410 memcpy(netid, p, nlen);
412 /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
413 p = xdr_inline_decode(streamp, 4);
416 rlen = be32_to_cpup(p);
418 p = xdr_inline_decode(streamp, rlen);
422 /* port is ".ABC.DEF", 8 chars max */
423 if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
424 dprintk("%s: Invalid address, length %d\n", __func__,
428 buf = kmalloc(rlen + 1, gfp_flags);
430 dprintk("%s: Not enough memory\n", __func__);
434 memcpy(buf, p, rlen);
436 /* replace port '.' with '-' */
437 portstr = strrchr(buf, '.');
439 dprintk("%s: Failed finding expected dot in port\n",
445 /* find '.' between address and port */
446 portstr = strrchr(buf, '.');
448 dprintk("%s: Failed finding expected dot between address and "
454 da = kzalloc(sizeof(*da), gfp_flags);
458 INIT_LIST_HEAD(&da->da_node);
460 if (!rpc_pton(buf, portstr-buf, (struct sockaddr *)&da->da_addr,
461 sizeof(da->da_addr))) {
462 dprintk("%s: error parsing address %s\n", __func__, buf);
467 sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
468 port = htons((tmp[0] << 8) | (tmp[1]));
470 switch (da->da_addr.ss_family) {
472 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
473 da->da_addrlen = sizeof(struct sockaddr_in);
479 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
480 da->da_addrlen = sizeof(struct sockaddr_in6);
481 match_netid = "tcp6";
488 dprintk("%s: unsupported address family: %u\n",
489 __func__, da->da_addr.ss_family);
493 if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
494 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
495 __func__, netid, match_netid);
499 /* save human readable address */
500 len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
501 da->da_remotestr = kzalloc(len, gfp_flags);
503 /* NULL is ok, only used for dprintk */
504 if (da->da_remotestr)
505 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
506 buf, endsep, ntohs(port));
508 dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
516 dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
524 /* Decode opaque device data and return the result */
525 static struct nfs4_file_layout_dsaddr*
526 decode_device(struct inode *ino, struct pnfs_device *pdev, gfp_t gfp_flags)
534 struct nfs4_file_layout_dsaddr *dsaddr = NULL;
535 struct xdr_stream stream;
537 struct page *scratch;
538 struct list_head dsaddrs;
539 struct nfs4_pnfs_ds_addr *da;
541 /* set up xdr stream */
542 scratch = alloc_page(gfp_flags);
546 xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
547 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
549 /* Get the stripe count (number of stripe index) */
550 p = xdr_inline_decode(&stream, 4);
552 goto out_err_free_scratch;
554 cnt = be32_to_cpup(p);
555 dprintk("%s stripe count %d\n", __func__, cnt);
556 if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
557 printk(KERN_WARNING "%s: stripe count %d greater than "
558 "supported maximum %d\n", __func__,
559 cnt, NFS4_PNFS_MAX_STRIPE_CNT);
560 goto out_err_free_scratch;
563 /* read stripe indices */
564 stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
566 goto out_err_free_scratch;
568 p = xdr_inline_decode(&stream, cnt << 2);
570 goto out_err_free_stripe_indices;
572 indexp = &stripe_indices[0];
573 max_stripe_index = 0;
574 for (i = 0; i < cnt; i++) {
575 *indexp = be32_to_cpup(p++);
576 max_stripe_index = max(max_stripe_index, *indexp);
580 /* Check the multipath list count */
581 p = xdr_inline_decode(&stream, 4);
583 goto out_err_free_stripe_indices;
585 num = be32_to_cpup(p);
586 dprintk("%s ds_num %u\n", __func__, num);
587 if (num > NFS4_PNFS_MAX_MULTI_CNT) {
588 printk(KERN_WARNING "%s: multipath count %d greater than "
589 "supported maximum %d\n", __func__,
590 num, NFS4_PNFS_MAX_MULTI_CNT);
591 goto out_err_free_stripe_indices;
594 /* validate stripe indices are all < num */
595 if (max_stripe_index >= num) {
596 printk(KERN_WARNING "%s: stripe index %u >= num ds %u\n",
597 __func__, max_stripe_index, num);
598 goto out_err_free_stripe_indices;
601 dsaddr = kzalloc(sizeof(*dsaddr) +
602 (sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
605 goto out_err_free_stripe_indices;
607 dsaddr->stripe_count = cnt;
608 dsaddr->stripe_indices = stripe_indices;
609 stripe_indices = NULL;
610 dsaddr->ds_num = num;
611 nfs4_init_deviceid_node(&dsaddr->id_node,
612 NFS_SERVER(ino)->pnfs_curr_ld,
613 NFS_SERVER(ino)->nfs_client,
616 INIT_LIST_HEAD(&dsaddrs);
618 for (i = 0; i < dsaddr->ds_num; i++) {
622 p = xdr_inline_decode(&stream, 4);
624 goto out_err_free_deviceid;
626 mp_count = be32_to_cpup(p); /* multipath count */
627 for (j = 0; j < mp_count; j++) {
628 da = decode_ds_addr(&stream, gfp_flags);
630 list_add_tail(&da->da_node, &dsaddrs);
632 if (list_empty(&dsaddrs)) {
633 dprintk("%s: no suitable DS addresses found\n",
635 goto out_err_free_deviceid;
638 dsaddr->ds_list[i] = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
639 if (!dsaddr->ds_list[i])
640 goto out_err_drain_dsaddrs;
642 /* If DS was already in cache, free ds addrs */
643 while (!list_empty(&dsaddrs)) {
644 da = list_first_entry(&dsaddrs,
645 struct nfs4_pnfs_ds_addr,
647 list_del_init(&da->da_node);
648 kfree(da->da_remotestr);
653 __free_page(scratch);
656 out_err_drain_dsaddrs:
657 while (!list_empty(&dsaddrs)) {
658 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
660 list_del_init(&da->da_node);
661 kfree(da->da_remotestr);
664 out_err_free_deviceid:
665 nfs4_fl_free_deviceid(dsaddr);
666 /* stripe_indicies was part of dsaddr */
667 goto out_err_free_scratch;
668 out_err_free_stripe_indices:
669 kfree(stripe_indices);
670 out_err_free_scratch:
671 __free_page(scratch);
673 dprintk("%s ERROR: returning NULL\n", __func__);
678 * Decode the opaque device specified in 'dev' and add it to the cache of
681 static struct nfs4_file_layout_dsaddr *
682 decode_and_add_device(struct inode *inode, struct pnfs_device *dev, gfp_t gfp_flags)
684 struct nfs4_deviceid_node *d;
685 struct nfs4_file_layout_dsaddr *n, *new;
687 new = decode_device(inode, dev, gfp_flags);
689 printk(KERN_WARNING "%s: Could not decode or add device\n",
694 d = nfs4_insert_deviceid_node(&new->id_node);
695 n = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
697 nfs4_fl_free_deviceid(new);
705 * Retrieve the information for dev_id, add it to the list
706 * of available devices, and return it.
708 struct nfs4_file_layout_dsaddr *
709 get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id, gfp_t gfp_flags)
711 struct pnfs_device *pdev = NULL;
714 struct page **pages = NULL;
715 struct nfs4_file_layout_dsaddr *dsaddr = NULL;
717 struct nfs_server *server = NFS_SERVER(inode);
720 * Use the session max response size as the basis for setting
721 * GETDEVICEINFO's maxcount
723 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
724 max_pages = max_resp_sz >> PAGE_SHIFT;
725 dprintk("%s inode %p max_resp_sz %u max_pages %d\n",
726 __func__, inode, max_resp_sz, max_pages);
728 pdev = kzalloc(sizeof(struct pnfs_device), gfp_flags);
732 pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags);
737 for (i = 0; i < max_pages; i++) {
738 pages[i] = alloc_page(gfp_flags);
743 memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id));
744 pdev->layout_type = LAYOUT_NFSV4_1_FILES;
747 pdev->pglen = PAGE_SIZE * max_pages;
750 rc = nfs4_proc_getdeviceinfo(server, pdev);
751 dprintk("%s getdevice info returns %d\n", __func__, rc);
756 * Found new device, need to decode it and then add it to the
757 * list of known devices for this mountpoint.
759 dsaddr = decode_and_add_device(inode, pdev, gfp_flags);
761 for (i = 0; i < max_pages; i++)
762 __free_page(pages[i]);
765 dprintk("<-- %s dsaddr %p\n", __func__, dsaddr);
770 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
772 nfs4_put_deviceid_node(&dsaddr->id_node);
776 * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
777 * Then: ((res + fsi) % dsaddr->stripe_count)
780 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
782 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
785 tmp = offset - flseg->pattern_offset;
786 do_div(tmp, flseg->stripe_unit);
787 tmp += flseg->first_stripe_index;
788 return do_div(tmp, flseg->dsaddr->stripe_count);
792 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
794 return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
798 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
800 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
803 if (flseg->stripe_type == STRIPE_SPARSE) {
804 if (flseg->num_fh == 1)
806 else if (flseg->num_fh == 0)
807 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
810 i = nfs4_fl_calc_ds_index(lseg, j);
813 return flseg->fh_array[i];
817 filelayout_mark_devid_negative(struct nfs4_file_layout_dsaddr *dsaddr,
818 int err, const char *ds_remotestr)
820 u32 *p = (u32 *)&dsaddr->id_node.deviceid;
822 printk(KERN_ERR "NFS: data server %s connection error %d."
823 " Deviceid [%x%x%x%x] marked out of use.\n",
824 ds_remotestr, err, p[0], p[1], p[2], p[3]);
826 spin_lock(&nfs4_ds_cache_lock);
827 dsaddr->flags |= NFS4_DEVICE_ID_NEG_ENTRY;
828 spin_unlock(&nfs4_ds_cache_lock);
831 struct nfs4_pnfs_ds *
832 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
834 struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
835 struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
838 printk(KERN_ERR "%s: No data server for offset index %d\n",
844 struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
847 if (dsaddr->flags & NFS4_DEVICE_ID_NEG_ENTRY) {
848 /* Already tried to connect, don't try again */
849 dprintk("%s Deviceid marked out of use\n", __func__);
852 err = nfs4_ds_connect(s, ds);
854 filelayout_mark_devid_negative(dsaddr, err,