offb: Fix bug in calculating requested vram size
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / nfs / nfs4filelayoutdev.c
blob3b7bf13772641ac8409aef9e4e70d6eca99e7ec3
1 /*
2 * Device operations for the pnfs nfs4 file layout driver.
4 * Copyright (c) 2002
5 * The Regents of the University of Michigan
6 * All Rights Reserved
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
28 * such damages.
31 #include <linux/nfs_fs.h>
32 #include <linux/vmalloc.h>
34 #include "internal.h"
35 #include "nfs4filelayout.h"
37 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
40 * Data server cache
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);
51 /* Debug routines */
52 void
53 print_ds(struct nfs4_pnfs_ds *ds)
55 if (ds == NULL) {
56 printk("%s NULL device\n", __func__);
57 return;
59 printk(" ip_addr %x port %hu\n"
60 " ref count %d\n"
61 " client %p\n"
62 " cl_exchange_flags %x\n",
63 ntohl(ds->ds_ip_addr), ntohs(ds->ds_port),
64 atomic_read(&ds->ds_count), ds->ds_clp,
65 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
68 /* nfs4_ds_cache_lock is held */
69 static struct nfs4_pnfs_ds *
70 _data_server_lookup_locked(u32 ip_addr, u32 port)
72 struct nfs4_pnfs_ds *ds;
74 dprintk("_data_server_lookup: ip_addr=%x port=%hu\n",
75 ntohl(ip_addr), ntohs(port));
77 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node) {
78 if (ds->ds_ip_addr == ip_addr &&
79 ds->ds_port == port) {
80 return ds;
83 return NULL;
87 * Create an rpc connection to the nfs4_pnfs_ds data server
88 * Currently only support IPv4
90 static int
91 nfs4_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds)
93 struct nfs_client *clp;
94 struct sockaddr_in sin;
95 int status = 0;
97 dprintk("--> %s ip:port %x:%hu au_flavor %d\n", __func__,
98 ntohl(ds->ds_ip_addr), ntohs(ds->ds_port),
99 mds_srv->nfs_client->cl_rpcclient->cl_auth->au_flavor);
101 sin.sin_family = AF_INET;
102 sin.sin_addr.s_addr = ds->ds_ip_addr;
103 sin.sin_port = ds->ds_port;
105 clp = nfs4_set_ds_client(mds_srv->nfs_client, (struct sockaddr *)&sin,
106 sizeof(sin), IPPROTO_TCP);
107 if (IS_ERR(clp)) {
108 status = PTR_ERR(clp);
109 goto out;
112 if ((clp->cl_exchange_flags & EXCHGID4_FLAG_MASK_PNFS) != 0) {
113 if (!is_ds_client(clp)) {
114 status = -ENODEV;
115 goto out_put;
117 ds->ds_clp = clp;
118 dprintk("%s [existing] ip=%x, port=%hu\n", __func__,
119 ntohl(ds->ds_ip_addr), ntohs(ds->ds_port));
120 goto out;
124 * Do not set NFS_CS_CHECK_LEASE_TIME instead set the DS lease to
125 * be equal to the MDS lease. Renewal is scheduled in create_session.
127 spin_lock(&mds_srv->nfs_client->cl_lock);
128 clp->cl_lease_time = mds_srv->nfs_client->cl_lease_time;
129 spin_unlock(&mds_srv->nfs_client->cl_lock);
130 clp->cl_last_renewal = jiffies;
132 /* New nfs_client */
133 status = nfs4_init_ds_session(clp);
134 if (status)
135 goto out_put;
137 ds->ds_clp = clp;
138 dprintk("%s [new] ip=%x, port=%hu\n", __func__, ntohl(ds->ds_ip_addr),
139 ntohs(ds->ds_port));
140 out:
141 return status;
142 out_put:
143 nfs_put_client(clp);
144 goto out;
147 static void
148 destroy_ds(struct nfs4_pnfs_ds *ds)
150 dprintk("--> %s\n", __func__);
151 ifdebug(FACILITY)
152 print_ds(ds);
154 if (ds->ds_clp)
155 nfs_put_client(ds->ds_clp);
156 kfree(ds);
159 void
160 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
162 struct nfs4_pnfs_ds *ds;
163 int i;
165 nfs4_print_deviceid(&dsaddr->id_node.deviceid);
167 for (i = 0; i < dsaddr->ds_num; i++) {
168 ds = dsaddr->ds_list[i];
169 if (ds != NULL) {
170 if (atomic_dec_and_lock(&ds->ds_count,
171 &nfs4_ds_cache_lock)) {
172 list_del_init(&ds->ds_node);
173 spin_unlock(&nfs4_ds_cache_lock);
174 destroy_ds(ds);
178 kfree(dsaddr->stripe_indices);
179 kfree(dsaddr);
182 static struct nfs4_pnfs_ds *
183 nfs4_pnfs_ds_add(struct inode *inode, u32 ip_addr, u32 port, gfp_t gfp_flags)
185 struct nfs4_pnfs_ds *tmp_ds, *ds;
187 ds = kzalloc(sizeof(*tmp_ds), gfp_flags);
188 if (!ds)
189 goto out;
191 spin_lock(&nfs4_ds_cache_lock);
192 tmp_ds = _data_server_lookup_locked(ip_addr, port);
193 if (tmp_ds == NULL) {
194 ds->ds_ip_addr = ip_addr;
195 ds->ds_port = port;
196 atomic_set(&ds->ds_count, 1);
197 INIT_LIST_HEAD(&ds->ds_node);
198 ds->ds_clp = NULL;
199 list_add(&ds->ds_node, &nfs4_data_server_cache);
200 dprintk("%s add new data server ip 0x%x\n", __func__,
201 ds->ds_ip_addr);
202 } else {
203 kfree(ds);
204 atomic_inc(&tmp_ds->ds_count);
205 dprintk("%s data server found ip 0x%x, inc'ed ds_count to %d\n",
206 __func__, tmp_ds->ds_ip_addr,
207 atomic_read(&tmp_ds->ds_count));
208 ds = tmp_ds;
210 spin_unlock(&nfs4_ds_cache_lock);
211 out:
212 return ds;
216 * Currently only support ipv4, and one multi-path address.
218 static struct nfs4_pnfs_ds *
219 decode_and_add_ds(struct xdr_stream *streamp, struct inode *inode, gfp_t gfp_flags)
221 struct nfs4_pnfs_ds *ds = NULL;
222 char *buf;
223 const char *ipend, *pstr;
224 u32 ip_addr, port;
225 int nlen, rlen, i;
226 int tmp[2];
227 __be32 *p;
229 /* r_netid */
230 p = xdr_inline_decode(streamp, 4);
231 if (unlikely(!p))
232 goto out_err;
233 nlen = be32_to_cpup(p++);
235 p = xdr_inline_decode(streamp, nlen);
236 if (unlikely(!p))
237 goto out_err;
239 /* Check that netid is "tcp" */
240 if (nlen != 3 || memcmp((char *)p, "tcp", 3)) {
241 dprintk("%s: ERROR: non ipv4 TCP r_netid\n", __func__);
242 goto out_err;
245 /* r_addr */
246 p = xdr_inline_decode(streamp, 4);
247 if (unlikely(!p))
248 goto out_err;
249 rlen = be32_to_cpup(p);
251 p = xdr_inline_decode(streamp, rlen);
252 if (unlikely(!p))
253 goto out_err;
255 /* ipv6 length plus port is legal */
256 if (rlen > INET6_ADDRSTRLEN + 8) {
257 dprintk("%s: Invalid address, length %d\n", __func__,
258 rlen);
259 goto out_err;
261 buf = kmalloc(rlen + 1, gfp_flags);
262 if (!buf) {
263 dprintk("%s: Not enough memory\n", __func__);
264 goto out_err;
266 buf[rlen] = '\0';
267 memcpy(buf, p, rlen);
269 /* replace the port dots with dashes for the in4_pton() delimiter*/
270 for (i = 0; i < 2; i++) {
271 char *res = strrchr(buf, '.');
272 if (!res) {
273 dprintk("%s: Failed finding expected dots in port\n",
274 __func__);
275 goto out_free;
277 *res = '-';
280 /* Currently only support ipv4 address */
281 if (in4_pton(buf, rlen, (u8 *)&ip_addr, '-', &ipend) == 0) {
282 dprintk("%s: Only ipv4 addresses supported\n", __func__);
283 goto out_free;
286 /* port */
287 pstr = ipend;
288 sscanf(pstr, "-%d-%d", &tmp[0], &tmp[1]);
289 port = htons((tmp[0] << 8) | (tmp[1]));
291 ds = nfs4_pnfs_ds_add(inode, ip_addr, port, gfp_flags);
292 dprintk("%s: Decoded address and port %s\n", __func__, buf);
293 out_free:
294 kfree(buf);
295 out_err:
296 return ds;
299 /* Decode opaque device data and return the result */
300 static struct nfs4_file_layout_dsaddr*
301 decode_device(struct inode *ino, struct pnfs_device *pdev, gfp_t gfp_flags)
303 int i;
304 u32 cnt, num;
305 u8 *indexp;
306 __be32 *p;
307 u8 *stripe_indices;
308 u8 max_stripe_index;
309 struct nfs4_file_layout_dsaddr *dsaddr = NULL;
310 struct xdr_stream stream;
311 struct xdr_buf buf;
312 struct page *scratch;
314 /* set up xdr stream */
315 scratch = alloc_page(gfp_flags);
316 if (!scratch)
317 goto out_err;
319 xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
320 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
322 /* Get the stripe count (number of stripe index) */
323 p = xdr_inline_decode(&stream, 4);
324 if (unlikely(!p))
325 goto out_err_free_scratch;
327 cnt = be32_to_cpup(p);
328 dprintk("%s stripe count %d\n", __func__, cnt);
329 if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
330 printk(KERN_WARNING "%s: stripe count %d greater than "
331 "supported maximum %d\n", __func__,
332 cnt, NFS4_PNFS_MAX_STRIPE_CNT);
333 goto out_err_free_scratch;
336 /* read stripe indices */
337 stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
338 if (!stripe_indices)
339 goto out_err_free_scratch;
341 p = xdr_inline_decode(&stream, cnt << 2);
342 if (unlikely(!p))
343 goto out_err_free_stripe_indices;
345 indexp = &stripe_indices[0];
346 max_stripe_index = 0;
347 for (i = 0; i < cnt; i++) {
348 *indexp = be32_to_cpup(p++);
349 max_stripe_index = max(max_stripe_index, *indexp);
350 indexp++;
353 /* Check the multipath list count */
354 p = xdr_inline_decode(&stream, 4);
355 if (unlikely(!p))
356 goto out_err_free_stripe_indices;
358 num = be32_to_cpup(p);
359 dprintk("%s ds_num %u\n", __func__, num);
360 if (num > NFS4_PNFS_MAX_MULTI_CNT) {
361 printk(KERN_WARNING "%s: multipath count %d greater than "
362 "supported maximum %d\n", __func__,
363 num, NFS4_PNFS_MAX_MULTI_CNT);
364 goto out_err_free_stripe_indices;
367 /* validate stripe indices are all < num */
368 if (max_stripe_index >= num) {
369 printk(KERN_WARNING "%s: stripe index %u >= num ds %u\n",
370 __func__, max_stripe_index, num);
371 goto out_err_free_stripe_indices;
374 dsaddr = kzalloc(sizeof(*dsaddr) +
375 (sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
376 gfp_flags);
377 if (!dsaddr)
378 goto out_err_free_stripe_indices;
380 dsaddr->stripe_count = cnt;
381 dsaddr->stripe_indices = stripe_indices;
382 stripe_indices = NULL;
383 dsaddr->ds_num = num;
384 nfs4_init_deviceid_node(&dsaddr->id_node,
385 NFS_SERVER(ino)->pnfs_curr_ld,
386 NFS_SERVER(ino)->nfs_client,
387 &pdev->dev_id);
389 for (i = 0; i < dsaddr->ds_num; i++) {
390 int j;
391 u32 mp_count;
393 p = xdr_inline_decode(&stream, 4);
394 if (unlikely(!p))
395 goto out_err_free_deviceid;
397 mp_count = be32_to_cpup(p); /* multipath count */
398 if (mp_count > 1) {
399 printk(KERN_WARNING
400 "%s: Multipath count %d not supported, "
401 "skipping all greater than 1\n", __func__,
402 mp_count);
404 for (j = 0; j < mp_count; j++) {
405 if (j == 0) {
406 dsaddr->ds_list[i] = decode_and_add_ds(&stream,
407 ino, gfp_flags);
408 if (dsaddr->ds_list[i] == NULL)
409 goto out_err_free_deviceid;
410 } else {
411 u32 len;
412 /* skip extra multipath */
414 /* read len, skip */
415 p = xdr_inline_decode(&stream, 4);
416 if (unlikely(!p))
417 goto out_err_free_deviceid;
418 len = be32_to_cpup(p);
420 p = xdr_inline_decode(&stream, len);
421 if (unlikely(!p))
422 goto out_err_free_deviceid;
424 /* read len, skip */
425 p = xdr_inline_decode(&stream, 4);
426 if (unlikely(!p))
427 goto out_err_free_deviceid;
428 len = be32_to_cpup(p);
430 p = xdr_inline_decode(&stream, len);
431 if (unlikely(!p))
432 goto out_err_free_deviceid;
437 __free_page(scratch);
438 return dsaddr;
440 out_err_free_deviceid:
441 nfs4_fl_free_deviceid(dsaddr);
442 /* stripe_indicies was part of dsaddr */
443 goto out_err_free_scratch;
444 out_err_free_stripe_indices:
445 kfree(stripe_indices);
446 out_err_free_scratch:
447 __free_page(scratch);
448 out_err:
449 dprintk("%s ERROR: returning NULL\n", __func__);
450 return NULL;
454 * Decode the opaque device specified in 'dev' and add it to the cache of
455 * available devices.
457 static struct nfs4_file_layout_dsaddr *
458 decode_and_add_device(struct inode *inode, struct pnfs_device *dev, gfp_t gfp_flags)
460 struct nfs4_deviceid_node *d;
461 struct nfs4_file_layout_dsaddr *n, *new;
463 new = decode_device(inode, dev, gfp_flags);
464 if (!new) {
465 printk(KERN_WARNING "%s: Could not decode or add device\n",
466 __func__);
467 return NULL;
470 d = nfs4_insert_deviceid_node(&new->id_node);
471 n = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
472 if (n != new) {
473 nfs4_fl_free_deviceid(new);
474 return n;
477 return new;
481 * Retrieve the information for dev_id, add it to the list
482 * of available devices, and return it.
484 struct nfs4_file_layout_dsaddr *
485 get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id, gfp_t gfp_flags)
487 struct pnfs_device *pdev = NULL;
488 u32 max_resp_sz;
489 int max_pages;
490 struct page **pages = NULL;
491 struct nfs4_file_layout_dsaddr *dsaddr = NULL;
492 int rc, i;
493 struct nfs_server *server = NFS_SERVER(inode);
496 * Use the session max response size as the basis for setting
497 * GETDEVICEINFO's maxcount
499 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
500 max_pages = max_resp_sz >> PAGE_SHIFT;
501 dprintk("%s inode %p max_resp_sz %u max_pages %d\n",
502 __func__, inode, max_resp_sz, max_pages);
504 pdev = kzalloc(sizeof(struct pnfs_device), gfp_flags);
505 if (pdev == NULL)
506 return NULL;
508 pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags);
509 if (pages == NULL) {
510 kfree(pdev);
511 return NULL;
513 for (i = 0; i < max_pages; i++) {
514 pages[i] = alloc_page(gfp_flags);
515 if (!pages[i])
516 goto out_free;
519 memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id));
520 pdev->layout_type = LAYOUT_NFSV4_1_FILES;
521 pdev->pages = pages;
522 pdev->pgbase = 0;
523 pdev->pglen = PAGE_SIZE * max_pages;
524 pdev->mincount = 0;
526 rc = nfs4_proc_getdeviceinfo(server, pdev);
527 dprintk("%s getdevice info returns %d\n", __func__, rc);
528 if (rc)
529 goto out_free;
532 * Found new device, need to decode it and then add it to the
533 * list of known devices for this mountpoint.
535 dsaddr = decode_and_add_device(inode, pdev, gfp_flags);
536 out_free:
537 for (i = 0; i < max_pages; i++)
538 __free_page(pages[i]);
539 kfree(pages);
540 kfree(pdev);
541 dprintk("<-- %s dsaddr %p\n", __func__, dsaddr);
542 return dsaddr;
545 void
546 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
548 nfs4_put_deviceid_node(&dsaddr->id_node);
552 * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
553 * Then: ((res + fsi) % dsaddr->stripe_count)
556 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
558 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
559 u64 tmp;
561 tmp = offset - flseg->pattern_offset;
562 do_div(tmp, flseg->stripe_unit);
563 tmp += flseg->first_stripe_index;
564 return do_div(tmp, flseg->dsaddr->stripe_count);
568 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
570 return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
573 struct nfs_fh *
574 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
576 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
577 u32 i;
579 if (flseg->stripe_type == STRIPE_SPARSE) {
580 if (flseg->num_fh == 1)
581 i = 0;
582 else if (flseg->num_fh == 0)
583 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
584 return NULL;
585 else
586 i = nfs4_fl_calc_ds_index(lseg, j);
587 } else
588 i = j;
589 return flseg->fh_array[i];
592 static void
593 filelayout_mark_devid_negative(struct nfs4_file_layout_dsaddr *dsaddr,
594 int err, u32 ds_addr)
596 u32 *p = (u32 *)&dsaddr->id_node.deviceid;
598 printk(KERN_ERR "NFS: data server %x connection error %d."
599 " Deviceid [%x%x%x%x] marked out of use.\n",
600 ds_addr, err, p[0], p[1], p[2], p[3]);
602 spin_lock(&nfs4_ds_cache_lock);
603 dsaddr->flags |= NFS4_DEVICE_ID_NEG_ENTRY;
604 spin_unlock(&nfs4_ds_cache_lock);
607 struct nfs4_pnfs_ds *
608 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
610 struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
611 struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
613 if (ds == NULL) {
614 printk(KERN_ERR "%s: No data server for offset index %d\n",
615 __func__, ds_idx);
616 return NULL;
619 if (!ds->ds_clp) {
620 struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
621 int err;
623 if (dsaddr->flags & NFS4_DEVICE_ID_NEG_ENTRY) {
624 /* Already tried to connect, don't try again */
625 dprintk("%s Deviceid marked out of use\n", __func__);
626 return NULL;
628 err = nfs4_ds_connect(s, ds);
629 if (err) {
630 filelayout_mark_devid_negative(dsaddr, err,
631 ntohl(ds->ds_ip_addr));
632 return NULL;
635 return ds;