blob: 0f5b66f90d1740c8b44e570d4f6ec98bddebc7d3 [file] [log] [blame]
Ricardo Labiaga85e174b2010-10-20 00:17:58 -04001/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
Andy Adamson974cec82010-10-20 00:18:02 -040031#include "internal.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040032#include "pnfs.h"
33
34#define NFSDBG_FACILITY NFSDBG_PNFS
35
Fred Isaman02c35fc2010-10-20 00:17:59 -040036/* Locking:
37 *
38 * pnfs_spinlock:
39 * protects pnfs_modules_tbl.
40 */
41static DEFINE_SPINLOCK(pnfs_spinlock);
42
43/*
44 * pnfs_modules_tbl holds all pnfs modules
45 */
46static LIST_HEAD(pnfs_modules_tbl);
47
48/* Return the registered pnfs layout driver module matching given id */
49static struct pnfs_layoutdriver_type *
50find_pnfs_driver_locked(u32 id)
51{
52 struct pnfs_layoutdriver_type *local;
53
54 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
55 if (local->id == id)
56 goto out;
57 local = NULL;
58out:
59 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
60 return local;
61}
62
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040063static struct pnfs_layoutdriver_type *
64find_pnfs_driver(u32 id)
65{
Fred Isaman02c35fc2010-10-20 00:17:59 -040066 struct pnfs_layoutdriver_type *local;
67
68 spin_lock(&pnfs_spinlock);
69 local = find_pnfs_driver_locked(id);
70 spin_unlock(&pnfs_spinlock);
71 return local;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040072}
73
74void
75unset_pnfs_layoutdriver(struct nfs_server *nfss)
76{
Fred Isaman02c35fc2010-10-20 00:17:59 -040077 if (nfss->pnfs_curr_ld) {
Trond Myklebust1c787092010-10-21 16:56:48 -040078 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
Fred Isaman02c35fc2010-10-20 00:17:59 -040079 module_put(nfss->pnfs_curr_ld->owner);
80 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040081 nfss->pnfs_curr_ld = NULL;
82}
83
84/*
85 * Try to set the server's pnfs module to the pnfs layout type specified by id.
86 * Currently only one pNFS layout driver per filesystem is supported.
87 *
88 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
89 */
90void
91set_pnfs_layoutdriver(struct nfs_server *server, u32 id)
92{
93 struct pnfs_layoutdriver_type *ld_type = NULL;
94
95 if (id == 0)
96 goto out_no_driver;
97 if (!(server->nfs_client->cl_exchange_flags &
98 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
99 printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__,
100 id, server->nfs_client->cl_exchange_flags);
101 goto out_no_driver;
102 }
103 ld_type = find_pnfs_driver(id);
104 if (!ld_type) {
105 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
106 ld_type = find_pnfs_driver(id);
107 if (!ld_type) {
108 dprintk("%s: No pNFS module found for %u.\n",
109 __func__, id);
110 goto out_no_driver;
111 }
112 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400113 if (!try_module_get(ld_type->owner)) {
114 dprintk("%s: Could not grab reference on module\n", __func__);
115 goto out_no_driver;
116 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400117 server->pnfs_curr_ld = ld_type;
Trond Myklebust1c787092010-10-21 16:56:48 -0400118 if (ld_type->set_layoutdriver(server)) {
Fred Isaman02c35fc2010-10-20 00:17:59 -0400119 printk(KERN_ERR
120 "%s: Error initializing mount point for layout driver %u.\n",
121 __func__, id);
122 module_put(ld_type->owner);
123 goto out_no_driver;
124 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400125 dprintk("%s: pNFS module for %u set\n", __func__, id);
126 return;
127
128out_no_driver:
129 dprintk("%s: Using NFSv4 I/O\n", __func__);
130 server->pnfs_curr_ld = NULL;
131}
Fred Isaman02c35fc2010-10-20 00:17:59 -0400132
133int
134pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
135{
136 int status = -EINVAL;
137 struct pnfs_layoutdriver_type *tmp;
138
139 if (ld_type->id == 0) {
140 printk(KERN_ERR "%s id 0 is reserved\n", __func__);
141 return status;
142 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400143 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
144 printk(KERN_ERR "%s Layout driver must provide "
145 "alloc_lseg and free_lseg.\n", __func__);
146 return status;
147 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400148
149 spin_lock(&pnfs_spinlock);
150 tmp = find_pnfs_driver_locked(ld_type->id);
151 if (!tmp) {
152 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
153 status = 0;
154 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
155 ld_type->name);
156 } else {
157 printk(KERN_ERR "%s Module with id %d already loaded!\n",
158 __func__, ld_type->id);
159 }
160 spin_unlock(&pnfs_spinlock);
161
162 return status;
163}
164EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
165
166void
167pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
168{
169 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
170 spin_lock(&pnfs_spinlock);
171 list_del(&ld_type->pnfs_tblid);
172 spin_unlock(&pnfs_spinlock);
173}
174EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
Benny Halevye5e94012010-10-20 00:18:01 -0400175
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400176/*
177 * pNFS client layout cache
178 */
179
Fred Isamancc6e5342011-01-06 11:36:28 +0000180/* Need to hold i_lock if caller does not already hold reference */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000181void
Fred Isamancc6e5342011-01-06 11:36:28 +0000182get_layout_hdr(struct pnfs_layout_hdr *lo)
Benny Halevye5e94012010-10-20 00:18:01 -0400183{
Fred Isamancc6e5342011-01-06 11:36:28 +0000184 atomic_inc(&lo->plh_refcount);
185}
186
187static void
188destroy_layout_hdr(struct pnfs_layout_hdr *lo)
189{
190 dprintk("%s: freeing layout cache %p\n", __func__, lo);
191 BUG_ON(!list_empty(&lo->plh_layouts));
192 NFS_I(lo->plh_inode)->layout = NULL;
193 kfree(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400194}
195
196static void
197put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
198{
Fred Isamancc6e5342011-01-06 11:36:28 +0000199 if (atomic_dec_and_test(&lo->plh_refcount))
200 destroy_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400201}
202
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400203void
Fred Isamancc6e5342011-01-06 11:36:28 +0000204put_layout_hdr(struct pnfs_layout_hdr *lo)
Andy Adamson974cec82010-10-20 00:18:02 -0400205{
Fred Isamancc6e5342011-01-06 11:36:28 +0000206 struct inode *inode = lo->plh_inode;
207
208 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
209 destroy_layout_hdr(lo);
210 spin_unlock(&inode->i_lock);
211 }
Andy Adamson974cec82010-10-20 00:18:02 -0400212}
213
214static void
215init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
216{
Fred Isaman566052c2011-01-06 11:36:20 +0000217 INIT_LIST_HEAD(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000218 atomic_set(&lseg->pls_refcount, 1);
219 smp_mb();
220 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
Fred Isaman566052c2011-01-06 11:36:20 +0000221 lseg->pls_layout = lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400222}
223
Fred Isaman4541d162011-01-06 11:36:23 +0000224static void free_lseg(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400225{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000226 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400227
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400228 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Fred Isaman52fabd72011-01-06 11:36:18 +0000229 /* Matched by get_layout_hdr in pnfs_insert_layout */
Fred Isamancc6e5342011-01-06 11:36:28 +0000230 put_layout_hdr(NFS_I(ino)->layout);
Andy Adamson974cec82010-10-20 00:18:02 -0400231}
232
Fred Isaman4541d162011-01-06 11:36:23 +0000233/* The use of tmp_list is necessary because pnfs_curr_ld->free_lseg
234 * could sleep, so must be called outside of the lock.
235 * Returns 1 if object was removed, otherwise return 0.
236 */
237static int
238put_lseg_locked(struct pnfs_layout_segment *lseg,
239 struct list_head *tmp_list)
Andy Adamson974cec82010-10-20 00:18:02 -0400240{
Fred Isaman4541d162011-01-06 11:36:23 +0000241 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
242 atomic_read(&lseg->pls_refcount),
243 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
244 if (atomic_dec_and_test(&lseg->pls_refcount)) {
245 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400246
Fred Isaman4541d162011-01-06 11:36:23 +0000247 BUG_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
248 list_del(&lseg->pls_list);
249 if (list_empty(&lseg->pls_layout->plh_segs)) {
Fred Isaman38511722011-02-03 18:28:50 +0000250 set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
251 /* Matched by initial refcount set in alloc_init_layout_hdr */
252 put_layout_hdr_locked(lseg->pls_layout);
Fred Isaman4541d162011-01-06 11:36:23 +0000253 }
Fred Isamanf7e89172011-01-06 11:36:32 +0000254 rpc_wake_up(&NFS_SERVER(ino)->roc_rpcwaitq);
Fred Isaman4541d162011-01-06 11:36:23 +0000255 list_add(&lseg->pls_list, tmp_list);
256 return 1;
257 }
258 return 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400259}
260
Fred Isaman4541d162011-01-06 11:36:23 +0000261static bool
262should_free_lseg(u32 lseg_iomode, u32 recall_iomode)
263{
264 return (recall_iomode == IOMODE_ANY ||
265 lseg_iomode == recall_iomode);
266}
267
268/* Returns 1 if lseg is removed from list, 0 otherwise */
269static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
270 struct list_head *tmp_list)
271{
272 int rv = 0;
273
274 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
275 /* Remove the reference keeping the lseg in the
276 * list. It will now be removed when all
277 * outstanding io is finished.
278 */
279 rv = put_lseg_locked(lseg, tmp_list);
280 }
281 return rv;
282}
283
284/* Returns count of number of matching invalid lsegs remaining in list
285 * after call.
286 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000287int
Fred Isaman4541d162011-01-06 11:36:23 +0000288mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
289 struct list_head *tmp_list,
290 u32 iomode)
Andy Adamson974cec82010-10-20 00:18:02 -0400291{
292 struct pnfs_layout_segment *lseg, *next;
Fred Isaman4541d162011-01-06 11:36:23 +0000293 int invalid = 0, removed = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400294
295 dprintk("%s:Begin lo %p\n", __func__, lo);
296
Fred Isaman38511722011-02-03 18:28:50 +0000297 if (list_empty(&lo->plh_segs)) {
298 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
299 put_layout_hdr_locked(lo);
300 return 0;
301 }
Fred Isaman4541d162011-01-06 11:36:23 +0000302 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
303 if (should_free_lseg(lseg->pls_range.iomode, iomode)) {
304 dprintk("%s: freeing lseg %p iomode %d "
305 "offset %llu length %llu\n", __func__,
306 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
307 lseg->pls_range.length);
308 invalid++;
309 removed += mark_lseg_invalid(lseg, tmp_list);
310 }
311 dprintk("%s:Return %i\n", __func__, invalid - removed);
312 return invalid - removed;
Andy Adamson974cec82010-10-20 00:18:02 -0400313}
314
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000315/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000316void
Fred Isaman4541d162011-01-06 11:36:23 +0000317pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400318{
Fred Isaman4541d162011-01-06 11:36:23 +0000319 struct pnfs_layout_segment *lseg, *tmp;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000320 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400321
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000322 if (list_empty(free_me))
323 return;
324
325 lo = list_first_entry(free_me, struct pnfs_layout_segment,
326 pls_list)->pls_layout;
327
328 if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
329 struct nfs_client *clp;
330
331 clp = NFS_SERVER(lo->plh_inode)->nfs_client;
332 spin_lock(&clp->cl_lock);
333 list_del_init(&lo->plh_layouts);
334 spin_unlock(&clp->cl_lock);
335 }
Fred Isaman4541d162011-01-06 11:36:23 +0000336 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000337 list_del(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000338 free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400339 }
340}
341
Benny Halevye5e94012010-10-20 00:18:01 -0400342void
343pnfs_destroy_layout(struct nfs_inode *nfsi)
344{
345 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400346 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400347
348 spin_lock(&nfsi->vfs_inode.i_lock);
349 lo = nfsi->layout;
350 if (lo) {
Fred Isaman38511722011-02-03 18:28:50 +0000351 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
Fred Isaman4541d162011-01-06 11:36:23 +0000352 mark_matching_lsegs_invalid(lo, &tmp_list, IOMODE_ANY);
Benny Halevye5e94012010-10-20 00:18:01 -0400353 }
354 spin_unlock(&nfsi->vfs_inode.i_lock);
Andy Adamson974cec82010-10-20 00:18:02 -0400355 pnfs_free_lseg_list(&tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400356}
357
Andy Adamson974cec82010-10-20 00:18:02 -0400358/*
359 * Called by the state manger to remove all layouts established under an
360 * expired lease.
361 */
362void
363pnfs_destroy_all_layouts(struct nfs_client *clp)
364{
365 struct pnfs_layout_hdr *lo;
366 LIST_HEAD(tmp_list);
367
368 spin_lock(&clp->cl_lock);
369 list_splice_init(&clp->cl_layouts, &tmp_list);
370 spin_unlock(&clp->cl_lock);
371
372 while (!list_empty(&tmp_list)) {
373 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000374 plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400375 dprintk("%s freeing layout for inode %lu\n", __func__,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000376 lo->plh_inode->i_ino);
377 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400378 }
379}
380
Fred Isamanfd6002e2011-01-06 11:36:22 +0000381/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000382void
383pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
384 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400385{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000386 u32 oldseq, newseq;
Andy Adamson974cec82010-10-20 00:18:02 -0400387
Fred Isamanfd6002e2011-01-06 11:36:22 +0000388 oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid);
389 newseq = be32_to_cpu(new->stateid.seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000390 if ((int)(newseq - oldseq) > 0) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000391 memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid));
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000392 if (update_barrier) {
393 u32 new_barrier = be32_to_cpu(new->stateid.seqid);
394
395 if ((int)(new_barrier - lo->plh_barrier))
396 lo->plh_barrier = new_barrier;
397 } else {
398 /* Because of wraparound, we want to keep the barrier
399 * "close" to the current seqids. It needs to be
400 * within 2**31 to count as "behind", so if it
401 * gets too near that limit, give us a litle leeway
402 * and bring it to within 2**30.
403 * NOTE - and yes, this is all unsigned arithmetic.
404 */
405 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
406 lo->plh_barrier = newseq - (1 << 30);
407 }
408 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400409}
410
Fred Isamancf7d63f2011-01-06 11:36:25 +0000411/* lget is set to 1 if called from inside send_layoutget call chain */
412static bool
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000413pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
414 int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000415{
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000416 if ((stateid) &&
417 (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0)
418 return true;
Fred Isamanf7e89172011-01-06 11:36:32 +0000419 return lo->plh_block_lgets ||
Fred Isaman38511722011-02-03 18:28:50 +0000420 test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
Fred Isamanf7e89172011-01-06 11:36:32 +0000421 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000422 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000423 (atomic_read(&lo->plh_outstanding) > lget));
424}
425
Fred Isamanfd6002e2011-01-06 11:36:22 +0000426int
427pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
428 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400429{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000430 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400431
432 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000433 spin_lock(&lo->plh_inode->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000434 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000435 status = -EAGAIN;
436 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000437 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400438
Fred Isamanfd6002e2011-01-06 11:36:22 +0000439 do {
440 seq = read_seqbegin(&open_state->seqlock);
441 memcpy(dst->data, open_state->stateid.data,
442 sizeof(open_state->stateid.data));
443 } while (read_seqretry(&open_state->seqlock, seq));
444 } else
445 memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data));
446 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400447 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000448 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400449}
450
451/*
452* Get layout from server.
453* for now, assume that whole file layouts are requested.
454* arg->offset: 0
455* arg->length: all ones
456*/
Benny Halevye5e94012010-10-20 00:18:01 -0400457static struct pnfs_layout_segment *
458send_layoutget(struct pnfs_layout_hdr *lo,
459 struct nfs_open_context *ctx,
460 u32 iomode)
461{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000462 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400463 struct nfs_server *server = NFS_SERVER(ino);
464 struct nfs4_layoutget *lgp;
465 struct pnfs_layout_segment *lseg = NULL;
Benny Halevye5e94012010-10-20 00:18:01 -0400466
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400467 dprintk("--> %s\n", __func__);
468
469 BUG_ON(ctx == NULL);
470 lgp = kzalloc(sizeof(*lgp), GFP_KERNEL);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000471 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400472 return NULL;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400473 lgp->args.minlength = NFS4_MAX_UINT64;
474 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
475 lgp->args.range.iomode = iomode;
476 lgp->args.range.offset = 0;
477 lgp->args.range.length = NFS4_MAX_UINT64;
478 lgp->args.type = server->pnfs_curr_ld->id;
479 lgp->args.inode = ino;
480 lgp->args.ctx = get_nfs_open_context(ctx);
481 lgp->lsegpp = &lseg;
482
483 /* Synchronously retrieve layout information from server and
484 * store in lseg.
485 */
486 nfs4_proc_layoutget(lgp);
487 if (!lseg) {
488 /* remember that LAYOUTGET failed and suspend trying */
Fred Isaman566052c2011-01-06 11:36:20 +0000489 set_bit(lo_fail_bit(iomode), &lo->plh_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400490 }
Andy Adamson974cec82010-10-20 00:18:02 -0400491 return lseg;
492}
493
Fred Isamanf7e89172011-01-06 11:36:32 +0000494bool pnfs_roc(struct inode *ino)
495{
496 struct pnfs_layout_hdr *lo;
497 struct pnfs_layout_segment *lseg, *tmp;
498 LIST_HEAD(tmp_list);
499 bool found = false;
500
501 spin_lock(&ino->i_lock);
502 lo = NFS_I(ino)->layout;
503 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
504 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
505 goto out_nolayout;
506 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
507 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
508 mark_lseg_invalid(lseg, &tmp_list);
509 found = true;
510 }
511 if (!found)
512 goto out_nolayout;
513 lo->plh_block_lgets++;
514 get_layout_hdr(lo); /* matched in pnfs_roc_release */
515 spin_unlock(&ino->i_lock);
516 pnfs_free_lseg_list(&tmp_list);
517 return true;
518
519out_nolayout:
520 spin_unlock(&ino->i_lock);
521 return false;
522}
523
524void pnfs_roc_release(struct inode *ino)
525{
526 struct pnfs_layout_hdr *lo;
527
528 spin_lock(&ino->i_lock);
529 lo = NFS_I(ino)->layout;
530 lo->plh_block_lgets--;
531 put_layout_hdr_locked(lo);
532 spin_unlock(&ino->i_lock);
533}
534
535void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
536{
537 struct pnfs_layout_hdr *lo;
538
539 spin_lock(&ino->i_lock);
540 lo = NFS_I(ino)->layout;
541 if ((int)(barrier - lo->plh_barrier) > 0)
542 lo->plh_barrier = barrier;
543 spin_unlock(&ino->i_lock);
544}
545
546bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
547{
548 struct nfs_inode *nfsi = NFS_I(ino);
549 struct pnfs_layout_segment *lseg;
550 bool found = false;
551
552 spin_lock(&ino->i_lock);
553 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
554 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
555 found = true;
556 break;
557 }
558 if (!found) {
559 struct pnfs_layout_hdr *lo = nfsi->layout;
560 u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid);
561
562 /* Since close does not return a layout stateid for use as
563 * a barrier, we choose the worst-case barrier.
564 */
565 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
566 }
567 spin_unlock(&ino->i_lock);
568 return found;
569}
570
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400571/*
572 * Compare two layout segments for sorting into layout cache.
573 * We want to preferentially return RW over RO layouts, so ensure those
574 * are seen first.
575 */
576static s64
577cmp_layout(u32 iomode1, u32 iomode2)
578{
579 /* read > read/write */
580 return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ);
581}
582
Andy Adamson974cec82010-10-20 00:18:02 -0400583static void
584pnfs_insert_layout(struct pnfs_layout_hdr *lo,
585 struct pnfs_layout_segment *lseg)
586{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400587 struct pnfs_layout_segment *lp;
588 int found = 0;
589
Andy Adamson974cec82010-10-20 00:18:02 -0400590 dprintk("%s:Begin\n", __func__);
591
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000592 assert_spin_locked(&lo->plh_inode->i_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000593 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000594 if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400595 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000596 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400597 dprintk("%s: inserted lseg %p "
598 "iomode %d offset %llu length %llu before "
599 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000600 __func__, lseg, lseg->pls_range.iomode,
601 lseg->pls_range.offset, lseg->pls_range.length,
602 lp, lp->pls_range.iomode, lp->pls_range.offset,
603 lp->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400604 found = 1;
605 break;
Andy Adamson974cec82010-10-20 00:18:02 -0400606 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400607 if (!found) {
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000608 list_add_tail(&lseg->pls_list, &lo->plh_segs);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400609 dprintk("%s: inserted lseg %p "
610 "iomode %d offset %llu length %llu at tail\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000611 __func__, lseg, lseg->pls_range.iomode,
612 lseg->pls_range.offset, lseg->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400613 }
Fred Isamancc6e5342011-01-06 11:36:28 +0000614 get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400615
616 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400617}
618
619static struct pnfs_layout_hdr *
620alloc_init_layout_hdr(struct inode *ino)
621{
622 struct pnfs_layout_hdr *lo;
623
624 lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL);
625 if (!lo)
626 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +0000627 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000628 INIT_LIST_HEAD(&lo->plh_layouts);
629 INIT_LIST_HEAD(&lo->plh_segs);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000630 INIT_LIST_HEAD(&lo->plh_bulk_recall);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000631 lo->plh_inode = ino;
Benny Halevye5e94012010-10-20 00:18:01 -0400632 return lo;
633}
634
635static struct pnfs_layout_hdr *
636pnfs_find_alloc_layout(struct inode *ino)
637{
638 struct nfs_inode *nfsi = NFS_I(ino);
639 struct pnfs_layout_hdr *new = NULL;
640
641 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
642
643 assert_spin_locked(&ino->i_lock);
Fred Isaman4541d162011-01-06 11:36:23 +0000644 if (nfsi->layout) {
645 if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
646 return NULL;
647 else
648 return nfsi->layout;
649 }
Benny Halevye5e94012010-10-20 00:18:01 -0400650 spin_unlock(&ino->i_lock);
651 new = alloc_init_layout_hdr(ino);
652 spin_lock(&ino->i_lock);
653
654 if (likely(nfsi->layout == NULL)) /* Won the race? */
655 nfsi->layout = new;
656 else
657 kfree(new);
658 return nfsi->layout;
659}
660
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400661/*
662 * iomode matching rules:
663 * iomode lseg match
664 * ----- ----- -----
665 * ANY READ true
666 * ANY RW true
667 * RW READ false
668 * RW RW true
669 * READ READ true
670 * READ RW true
671 */
672static int
673is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode)
674{
Fred Isaman566052c2011-01-06 11:36:20 +0000675 return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400676}
677
678/*
679 * lookup range in layout
680 */
Benny Halevye5e94012010-10-20 00:18:01 -0400681static struct pnfs_layout_segment *
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000682pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode)
Benny Halevye5e94012010-10-20 00:18:01 -0400683{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400684 struct pnfs_layout_segment *lseg, *ret = NULL;
685
686 dprintk("%s:Begin\n", __func__);
687
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000688 assert_spin_locked(&lo->plh_inode->i_lock);
689 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +0000690 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
691 is_matching_lseg(lseg, iomode)) {
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400692 ret = lseg;
693 break;
694 }
Fred Isaman566052c2011-01-06 11:36:20 +0000695 if (cmp_layout(iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400696 break;
697 }
698
699 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +0000700 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400701 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400702}
703
704/*
705 * Layout segment is retreived from the server if not cached.
706 * The appropriate layout segment is referenced and returned to the caller.
707 */
708struct pnfs_layout_segment *
709pnfs_update_layout(struct inode *ino,
710 struct nfs_open_context *ctx,
711 enum pnfs_iomode iomode)
712{
713 struct nfs_inode *nfsi = NFS_I(ino);
Fred Isaman2130ff62011-01-06 11:36:26 +0000714 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -0400715 struct pnfs_layout_hdr *lo;
716 struct pnfs_layout_segment *lseg = NULL;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000717 bool first = false;
Benny Halevye5e94012010-10-20 00:18:01 -0400718
719 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
720 return NULL;
721 spin_lock(&ino->i_lock);
722 lo = pnfs_find_alloc_layout(ino);
723 if (lo == NULL) {
724 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
725 goto out_unlock;
726 }
727
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000728 /* Do we even need to bother with this? */
729 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
730 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
731 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400732 goto out_unlock;
733 }
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000734 /* Check to see if the layout for the given range already exists */
735 lseg = pnfs_find_lseg(lo, iomode);
736 if (lseg)
737 goto out_unlock;
Benny Halevye5e94012010-10-20 00:18:01 -0400738
739 /* if LAYOUTGET already failed once we don't try again */
Fred Isaman566052c2011-01-06 11:36:20 +0000740 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
Benny Halevye5e94012010-10-20 00:18:01 -0400741 goto out_unlock;
742
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000743 if (pnfs_layoutgets_blocked(lo, NULL, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +0000744 goto out_unlock;
745 atomic_inc(&lo->plh_outstanding);
746
Fred Isamancc6e5342011-01-06 11:36:28 +0000747 get_layout_hdr(lo);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000748 if (list_empty(&lo->plh_segs))
749 first = true;
750 spin_unlock(&ino->i_lock);
751 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +0000752 /* The lo must be on the clp list if there is any
753 * chance of a CB_LAYOUTRECALL(FILE) coming in.
754 */
755 spin_lock(&clp->cl_lock);
756 BUG_ON(!list_empty(&lo->plh_layouts));
757 list_add_tail(&lo->plh_layouts, &clp->cl_layouts);
758 spin_unlock(&clp->cl_lock);
759 }
Benny Halevye5e94012010-10-20 00:18:01 -0400760
761 lseg = send_layoutget(lo, ctx, iomode);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000762 if (!lseg && first) {
763 spin_lock(&clp->cl_lock);
764 list_del_init(&lo->plh_layouts);
765 spin_unlock(&clp->cl_lock);
Fred Isaman2130ff62011-01-06 11:36:26 +0000766 }
Fred Isamancf7d63f2011-01-06 11:36:25 +0000767 atomic_dec(&lo->plh_outstanding);
Fred Isamancc6e5342011-01-06 11:36:28 +0000768 put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400769out:
770 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
Fred Isaman566052c2011-01-06 11:36:20 +0000771 nfsi->layout->plh_flags, lseg);
Benny Halevye5e94012010-10-20 00:18:01 -0400772 return lseg;
773out_unlock:
774 spin_unlock(&ino->i_lock);
775 goto out;
776}
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400777
778int
779pnfs_layout_process(struct nfs4_layoutget *lgp)
780{
781 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
782 struct nfs4_layoutget_res *res = &lgp->res;
783 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000784 struct inode *ino = lo->plh_inode;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000785 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400786 int status = 0;
787
Fred Isamanfc1794c2011-01-06 11:36:27 +0000788 /* Verify we got what we asked for.
789 * Note that because the xdr parsing only accepts a single
790 * element array, this can fail even if the server is behaving
791 * correctly.
792 */
793 if (lgp->args.range.iomode > res->range.iomode ||
794 res->range.offset != 0 ||
795 res->range.length != NFS4_MAX_UINT64) {
796 status = -EINVAL;
797 goto out;
798 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400799 /* Inject layout blob into I/O device driver */
800 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res);
801 if (!lseg || IS_ERR(lseg)) {
802 if (!lseg)
803 status = -ENOMEM;
804 else
805 status = PTR_ERR(lseg);
806 dprintk("%s: Could not allocate layout: error %d\n",
807 __func__, status);
808 goto out;
809 }
810
811 spin_lock(&ino->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000812 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
813 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
814 dprintk("%s forget reply due to recall\n", __func__);
815 goto out_forget_reply;
816 }
817
818 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
819 dprintk("%s forget reply due to state\n", __func__);
820 goto out_forget_reply;
821 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400822 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +0000823 lseg->pls_range = res->range;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400824 *lgp->lsegpp = lseg;
825 pnfs_insert_layout(lo, lseg);
826
Fred Isamanf7e89172011-01-06 11:36:32 +0000827 if (res->return_on_close) {
828 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
829 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
830 }
831
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400832 /* Done processing layoutget. Set the layout stateid */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000833 pnfs_set_layout_stateid(lo, &res->stateid, false);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400834 spin_unlock(&ino->i_lock);
835out:
836 return status;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000837
838out_forget_reply:
839 spin_unlock(&ino->i_lock);
840 lseg->pls_layout = lo;
841 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
842 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400843}
844
845/*
846 * Device ID cache. Currently supports one layout type per struct nfs_client.
847 * Add layout type to the lookup key to expand to support multiple types.
848 */
849int
850pnfs_alloc_init_deviceid_cache(struct nfs_client *clp,
851 void (*free_callback)(struct pnfs_deviceid_node *))
852{
853 struct pnfs_deviceid_cache *c;
854
855 c = kzalloc(sizeof(struct pnfs_deviceid_cache), GFP_KERNEL);
856 if (!c)
857 return -ENOMEM;
858 spin_lock(&clp->cl_lock);
859 if (clp->cl_devid_cache != NULL) {
860 atomic_inc(&clp->cl_devid_cache->dc_ref);
861 dprintk("%s [kref [%d]]\n", __func__,
862 atomic_read(&clp->cl_devid_cache->dc_ref));
863 kfree(c);
864 } else {
865 /* kzalloc initializes hlists */
866 spin_lock_init(&c->dc_lock);
867 atomic_set(&c->dc_ref, 1);
868 c->dc_free_callback = free_callback;
869 clp->cl_devid_cache = c;
870 dprintk("%s [new]\n", __func__);
871 }
872 spin_unlock(&clp->cl_lock);
873 return 0;
874}
875EXPORT_SYMBOL_GPL(pnfs_alloc_init_deviceid_cache);
876
877/*
878 * Called from pnfs_layoutdriver_type->free_lseg
879 * last layout segment reference frees deviceid
880 */
881void
882pnfs_put_deviceid(struct pnfs_deviceid_cache *c,
883 struct pnfs_deviceid_node *devid)
884{
885 struct nfs4_deviceid *id = &devid->de_id;
886 struct pnfs_deviceid_node *d;
887 struct hlist_node *n;
888 long h = nfs4_deviceid_hash(id);
889
890 dprintk("%s [%d]\n", __func__, atomic_read(&devid->de_ref));
891 if (!atomic_dec_and_lock(&devid->de_ref, &c->dc_lock))
892 return;
893
894 hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[h], de_node)
895 if (!memcmp(&d->de_id, id, sizeof(*id))) {
896 hlist_del_rcu(&d->de_node);
897 spin_unlock(&c->dc_lock);
898 synchronize_rcu();
899 c->dc_free_callback(devid);
900 return;
901 }
902 spin_unlock(&c->dc_lock);
903 /* Why wasn't it found in the list? */
904 BUG();
905}
906EXPORT_SYMBOL_GPL(pnfs_put_deviceid);
907
908/* Find and reference a deviceid */
909struct pnfs_deviceid_node *
910pnfs_find_get_deviceid(struct pnfs_deviceid_cache *c, struct nfs4_deviceid *id)
911{
912 struct pnfs_deviceid_node *d;
913 struct hlist_node *n;
914 long hash = nfs4_deviceid_hash(id);
915
916 dprintk("--> %s hash %ld\n", __func__, hash);
917 rcu_read_lock();
918 hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) {
919 if (!memcmp(&d->de_id, id, sizeof(*id))) {
920 if (!atomic_inc_not_zero(&d->de_ref)) {
921 goto fail;
922 } else {
923 rcu_read_unlock();
924 return d;
925 }
926 }
927 }
928fail:
929 rcu_read_unlock();
930 return NULL;
931}
932EXPORT_SYMBOL_GPL(pnfs_find_get_deviceid);
933
934/*
935 * Add a deviceid to the cache.
936 * GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new
937 */
938struct pnfs_deviceid_node *
939pnfs_add_deviceid(struct pnfs_deviceid_cache *c, struct pnfs_deviceid_node *new)
940{
941 struct pnfs_deviceid_node *d;
942 long hash = nfs4_deviceid_hash(&new->de_id);
943
944 dprintk("--> %s hash %ld\n", __func__, hash);
945 spin_lock(&c->dc_lock);
946 d = pnfs_find_get_deviceid(c, &new->de_id);
947 if (d) {
948 spin_unlock(&c->dc_lock);
949 dprintk("%s [discard]\n", __func__);
950 c->dc_free_callback(new);
951 return d;
952 }
953 INIT_HLIST_NODE(&new->de_node);
954 atomic_set(&new->de_ref, 1);
955 hlist_add_head_rcu(&new->de_node, &c->dc_deviceids[hash]);
956 spin_unlock(&c->dc_lock);
957 dprintk("%s [new]\n", __func__);
958 return new;
959}
960EXPORT_SYMBOL_GPL(pnfs_add_deviceid);
961
962void
963pnfs_put_deviceid_cache(struct nfs_client *clp)
964{
965 struct pnfs_deviceid_cache *local = clp->cl_devid_cache;
966
Andy Adamsonb2a28972011-01-25 15:38:03 +0000967 dprintk("--> %s ({%d})\n", __func__, atomic_read(&local->dc_ref));
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400968 if (atomic_dec_and_lock(&local->dc_ref, &clp->cl_lock)) {
969 int i;
970 /* Verify cache is empty */
971 for (i = 0; i < NFS4_DEVICE_ID_HASH_SIZE; i++)
972 BUG_ON(!hlist_empty(&local->dc_deviceids[i]));
973 clp->cl_devid_cache = NULL;
974 spin_unlock(&clp->cl_lock);
975 kfree(local);
976 }
977}
978EXPORT_SYMBOL_GPL(pnfs_put_deviceid_cache);