5914659c8ec507d047a7340d92dafb0d913a6371
[deliverable/linux.git] / fs / nfs / nfs4filelayoutdev.c
1 /*
2 * Device operations for the pnfs nfs4 file layout driver.
3 *
4 * Copyright (c) 2002
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 * Garth Goodson <Garth.Goodson@netapp.com>
10 *
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.
19 *
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.
29 */
30
31 #include <linux/nfs_fs.h>
32 #include <linux/vmalloc.h>
33
34 #include "internal.h"
35 #include "nfs4filelayout.h"
36
37 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
38
39 /*
40 * Data server cache
41 *
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.
47 */
48 DEFINE_SPINLOCK(nfs4_ds_cache_lock);
49 static LIST_HEAD(nfs4_data_server_cache);
50
51 /* Debug routines */
52 void
53 print_ds(struct nfs4_pnfs_ds *ds)
54 {
55 if (ds == NULL) {
56 printk("%s NULL device\n", __func__);
57 return;
58 }
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);
66 }
67
68 /* nfs4_ds_cache_lock is held */
69 static struct nfs4_pnfs_ds *
70 _data_server_lookup_locked(u32 ip_addr, u32 port)
71 {
72 struct nfs4_pnfs_ds *ds;
73
74 dprintk("_data_server_lookup: ip_addr=%x port=%hu\n",
75 ntohl(ip_addr), ntohs(port));
76
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;
81 }
82 }
83 return NULL;
84 }
85
86 /*
87 * Create an rpc connection to the nfs4_pnfs_ds data server
88 * Currently only support IPv4
89 */
90 static int
91 nfs4_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds)
92 {
93 struct nfs_client *clp;
94 struct sockaddr_in sin;
95 int status = 0;
96
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);
100
101 sin.sin_family = AF_INET;
102 sin.sin_addr.s_addr = ds->ds_ip_addr;
103 sin.sin_port = ds->ds_port;
104
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;
110 }
111
112 if ((clp->cl_exchange_flags & EXCHGID4_FLAG_MASK_PNFS) != 0) {
113 if (!is_ds_client(clp)) {
114 status = -ENODEV;
115 goto out_put;
116 }
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;
121 }
122
123 /*
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.
126 */
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;
131
132 /* New nfs_client */
133 status = nfs4_init_ds_session(clp);
134 if (status)
135 goto out_put;
136
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;
145 }
146
147 static void
148 destroy_ds(struct nfs4_pnfs_ds *ds)
149 {
150 dprintk("--> %s\n", __func__);
151 ifdebug(FACILITY)
152 print_ds(ds);
153
154 if (ds->ds_clp)
155 nfs_put_client(ds->ds_clp);
156 kfree(ds);
157 }
158
159 void
160 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
161 {
162 struct nfs4_pnfs_ds *ds;
163 int i;
164
165 nfs4_print_deviceid(&dsaddr->id_node.deviceid);
166
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);
175 }
176 }
177 }
178 kfree(dsaddr->stripe_indices);
179 kfree(dsaddr);
180 }
181
182 static struct nfs4_pnfs_ds *
183 nfs4_pnfs_ds_add(struct inode *inode, u32 ip_addr, u32 port, gfp_t gfp_flags)
184 {
185 struct nfs4_pnfs_ds *tmp_ds, *ds;
186
187 ds = kzalloc(sizeof(*tmp_ds), gfp_flags);
188 if (!ds)
189 goto out;
190
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;
209 }
210 spin_unlock(&nfs4_ds_cache_lock);
211 out:
212 return ds;
213 }
214
215 /*
216 * Currently only support ipv4, and one multi-path address.
217 */
218 static struct nfs4_pnfs_ds *
219 decode_and_add_ds(struct xdr_stream *streamp, struct inode *inode, gfp_t gfp_flags)
220 {
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;
228
229 /* r_netid */
230 p = xdr_inline_decode(streamp, 4);
231 if (unlikely(!p))
232 goto out_err;
233 nlen = be32_to_cpup(p++);
234
235 p = xdr_inline_decode(streamp, nlen);
236 if (unlikely(!p))
237 goto out_err;
238
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;
243 }
244
245 /* r_addr */
246 p = xdr_inline_decode(streamp, 4);
247 if (unlikely(!p))
248 goto out_err;
249 rlen = be32_to_cpup(p);
250
251 p = xdr_inline_decode(streamp, rlen);
252 if (unlikely(!p))
253 goto out_err;
254
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;
260 }
261 buf = kmalloc(rlen + 1, gfp_flags);
262 if (!buf) {
263 dprintk("%s: Not enough memory\n", __func__);
264 goto out_err;
265 }
266 buf[rlen] = '\0';
267 memcpy(buf, p, rlen);
268
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;
276 }
277 *res = '-';
278 }
279
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;
284 }
285
286 /* port */
287 pstr = ipend;
288 sscanf(pstr, "-%d-%d", &tmp[0], &tmp[1]);
289 port = htons((tmp[0] << 8) | (tmp[1]));
290
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;
297 }
298
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)
302 {
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 .pages = pdev->pages,
313 .page_len = pdev->pglen,
314 .buflen = pdev->pglen,
315 .len = pdev->pglen,
316 };
317 struct page *scratch;
318
319 /* set up xdr stream */
320 scratch = alloc_page(gfp_flags);
321 if (!scratch)
322 goto out_err;
323
324 xdr_init_decode(&stream, &buf, NULL);
325 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
326
327 /* Get the stripe count (number of stripe index) */
328 p = xdr_inline_decode(&stream, 4);
329 if (unlikely(!p))
330 goto out_err_free_scratch;
331
332 cnt = be32_to_cpup(p);
333 dprintk("%s stripe count %d\n", __func__, cnt);
334 if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
335 printk(KERN_WARNING "%s: stripe count %d greater than "
336 "supported maximum %d\n", __func__,
337 cnt, NFS4_PNFS_MAX_STRIPE_CNT);
338 goto out_err_free_scratch;
339 }
340
341 /* read stripe indices */
342 stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
343 if (!stripe_indices)
344 goto out_err_free_scratch;
345
346 p = xdr_inline_decode(&stream, cnt << 2);
347 if (unlikely(!p))
348 goto out_err_free_stripe_indices;
349
350 indexp = &stripe_indices[0];
351 max_stripe_index = 0;
352 for (i = 0; i < cnt; i++) {
353 *indexp = be32_to_cpup(p++);
354 max_stripe_index = max(max_stripe_index, *indexp);
355 indexp++;
356 }
357
358 /* Check the multipath list count */
359 p = xdr_inline_decode(&stream, 4);
360 if (unlikely(!p))
361 goto out_err_free_stripe_indices;
362
363 num = be32_to_cpup(p);
364 dprintk("%s ds_num %u\n", __func__, num);
365 if (num > NFS4_PNFS_MAX_MULTI_CNT) {
366 printk(KERN_WARNING "%s: multipath count %d greater than "
367 "supported maximum %d\n", __func__,
368 num, NFS4_PNFS_MAX_MULTI_CNT);
369 goto out_err_free_stripe_indices;
370 }
371
372 /* validate stripe indices are all < num */
373 if (max_stripe_index >= num) {
374 printk(KERN_WARNING "%s: stripe index %u >= num ds %u\n",
375 __func__, max_stripe_index, num);
376 goto out_err_free_stripe_indices;
377 }
378
379 dsaddr = kzalloc(sizeof(*dsaddr) +
380 (sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
381 gfp_flags);
382 if (!dsaddr)
383 goto out_err_free_stripe_indices;
384
385 dsaddr->stripe_count = cnt;
386 dsaddr->stripe_indices = stripe_indices;
387 stripe_indices = NULL;
388 dsaddr->ds_num = num;
389 nfs4_init_deviceid_node(&dsaddr->id_node,
390 NFS_SERVER(ino)->pnfs_curr_ld,
391 NFS_SERVER(ino)->nfs_client,
392 &pdev->dev_id);
393
394 for (i = 0; i < dsaddr->ds_num; i++) {
395 int j;
396 u32 mp_count;
397
398 p = xdr_inline_decode(&stream, 4);
399 if (unlikely(!p))
400 goto out_err_free_deviceid;
401
402 mp_count = be32_to_cpup(p); /* multipath count */
403 if (mp_count > 1) {
404 printk(KERN_WARNING
405 "%s: Multipath count %d not supported, "
406 "skipping all greater than 1\n", __func__,
407 mp_count);
408 }
409 for (j = 0; j < mp_count; j++) {
410 if (j == 0) {
411 dsaddr->ds_list[i] = decode_and_add_ds(&stream,
412 ino, gfp_flags);
413 if (dsaddr->ds_list[i] == NULL)
414 goto out_err_free_deviceid;
415 } else {
416 u32 len;
417 /* skip extra multipath */
418
419 /* read len, skip */
420 p = xdr_inline_decode(&stream, 4);
421 if (unlikely(!p))
422 goto out_err_free_deviceid;
423 len = be32_to_cpup(p);
424
425 p = xdr_inline_decode(&stream, len);
426 if (unlikely(!p))
427 goto out_err_free_deviceid;
428
429 /* read len, skip */
430 p = xdr_inline_decode(&stream, 4);
431 if (unlikely(!p))
432 goto out_err_free_deviceid;
433 len = be32_to_cpup(p);
434
435 p = xdr_inline_decode(&stream, len);
436 if (unlikely(!p))
437 goto out_err_free_deviceid;
438 }
439 }
440 }
441
442 __free_page(scratch);
443 return dsaddr;
444
445 out_err_free_deviceid:
446 nfs4_fl_free_deviceid(dsaddr);
447 /* stripe_indicies was part of dsaddr */
448 goto out_err_free_scratch;
449 out_err_free_stripe_indices:
450 kfree(stripe_indices);
451 out_err_free_scratch:
452 __free_page(scratch);
453 out_err:
454 dprintk("%s ERROR: returning NULL\n", __func__);
455 return NULL;
456 }
457
458 /*
459 * Decode the opaque device specified in 'dev' and add it to the cache of
460 * available devices.
461 */
462 static struct nfs4_file_layout_dsaddr *
463 decode_and_add_device(struct inode *inode, struct pnfs_device *dev, gfp_t gfp_flags)
464 {
465 struct nfs4_deviceid_node *d;
466 struct nfs4_file_layout_dsaddr *n, *new;
467
468 new = decode_device(inode, dev, gfp_flags);
469 if (!new) {
470 printk(KERN_WARNING "%s: Could not decode or add device\n",
471 __func__);
472 return NULL;
473 }
474
475 d = nfs4_insert_deviceid_node(&new->id_node);
476 n = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
477 if (n != new) {
478 nfs4_fl_free_deviceid(new);
479 return n;
480 }
481
482 return new;
483 }
484
485 /*
486 * Retrieve the information for dev_id, add it to the list
487 * of available devices, and return it.
488 */
489 struct nfs4_file_layout_dsaddr *
490 get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id, gfp_t gfp_flags)
491 {
492 struct pnfs_device *pdev = NULL;
493 u32 max_resp_sz;
494 int max_pages;
495 struct page **pages = NULL;
496 struct nfs4_file_layout_dsaddr *dsaddr = NULL;
497 int rc, i;
498 struct nfs_server *server = NFS_SERVER(inode);
499
500 /*
501 * Use the session max response size as the basis for setting
502 * GETDEVICEINFO's maxcount
503 */
504 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
505 max_pages = max_resp_sz >> PAGE_SHIFT;
506 dprintk("%s inode %p max_resp_sz %u max_pages %d\n",
507 __func__, inode, max_resp_sz, max_pages);
508
509 pdev = kzalloc(sizeof(struct pnfs_device), gfp_flags);
510 if (pdev == NULL)
511 return NULL;
512
513 pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags);
514 if (pages == NULL) {
515 kfree(pdev);
516 return NULL;
517 }
518 for (i = 0; i < max_pages; i++) {
519 pages[i] = alloc_page(gfp_flags);
520 if (!pages[i])
521 goto out_free;
522 }
523
524 memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id));
525 pdev->layout_type = LAYOUT_NFSV4_1_FILES;
526 pdev->pages = pages;
527 pdev->pgbase = 0;
528 pdev->pglen = PAGE_SIZE * max_pages;
529 pdev->mincount = 0;
530
531 rc = nfs4_proc_getdeviceinfo(server, pdev);
532 dprintk("%s getdevice info returns %d\n", __func__, rc);
533 if (rc)
534 goto out_free;
535
536 /*
537 * Found new device, need to decode it and then add it to the
538 * list of known devices for this mountpoint.
539 */
540 dsaddr = decode_and_add_device(inode, pdev, gfp_flags);
541 out_free:
542 for (i = 0; i < max_pages; i++)
543 __free_page(pages[i]);
544 kfree(pages);
545 kfree(pdev);
546 dprintk("<-- %s dsaddr %p\n", __func__, dsaddr);
547 return dsaddr;
548 }
549
550 void
551 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
552 {
553 nfs4_put_deviceid_node(&dsaddr->id_node);
554 }
555
556 /*
557 * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
558 * Then: ((res + fsi) % dsaddr->stripe_count)
559 */
560 u32
561 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
562 {
563 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
564 u64 tmp;
565
566 tmp = offset - flseg->pattern_offset;
567 do_div(tmp, flseg->stripe_unit);
568 tmp += flseg->first_stripe_index;
569 return do_div(tmp, flseg->dsaddr->stripe_count);
570 }
571
572 u32
573 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
574 {
575 return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
576 }
577
578 struct nfs_fh *
579 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
580 {
581 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
582 u32 i;
583
584 if (flseg->stripe_type == STRIPE_SPARSE) {
585 if (flseg->num_fh == 1)
586 i = 0;
587 else if (flseg->num_fh == 0)
588 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
589 return NULL;
590 else
591 i = nfs4_fl_calc_ds_index(lseg, j);
592 } else
593 i = j;
594 return flseg->fh_array[i];
595 }
596
597 static void
598 filelayout_mark_devid_negative(struct nfs4_file_layout_dsaddr *dsaddr,
599 int err, u32 ds_addr)
600 {
601 u32 *p = (u32 *)&dsaddr->id_node.deviceid;
602
603 printk(KERN_ERR "NFS: data server %x connection error %d."
604 " Deviceid [%x%x%x%x] marked out of use.\n",
605 ds_addr, err, p[0], p[1], p[2], p[3]);
606
607 spin_lock(&nfs4_ds_cache_lock);
608 dsaddr->flags |= NFS4_DEVICE_ID_NEG_ENTRY;
609 spin_unlock(&nfs4_ds_cache_lock);
610 }
611
612 struct nfs4_pnfs_ds *
613 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
614 {
615 struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
616 struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
617
618 if (ds == NULL) {
619 printk(KERN_ERR "%s: No data server for offset index %d\n",
620 __func__, ds_idx);
621 return NULL;
622 }
623
624 if (!ds->ds_clp) {
625 struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
626 int err;
627
628 if (dsaddr->flags & NFS4_DEVICE_ID_NEG_ENTRY) {
629 /* Already tried to connect, don't try again */
630 dprintk("%s Deviceid marked out of use\n", __func__);
631 return NULL;
632 }
633 err = nfs4_ds_connect(s, ds);
634 if (err) {
635 filelayout_mark_devid_negative(dsaddr, err,
636 ntohl(ds->ds_ip_addr));
637 return NULL;
638 }
639 }
640 return ds;
641 }
This page took 0.086862 seconds and 4 git commands to generate.