blob: 330cee115de0a3db6be57750df7e2a1f847e9162 [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 Isamand684d2a2011-03-01 01:34:13 +0000233static void
234put_lseg_common(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400235{
Fred Isamand684d2a2011-03-01 01:34:13 +0000236 struct inode *inode = lseg->pls_layout->plh_inode;
237
238 BUG_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
239 list_del_init(&lseg->pls_list);
240 if (list_empty(&lseg->pls_layout->plh_segs)) {
241 set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
242 /* Matched by initial refcount set in alloc_init_layout_hdr */
243 put_layout_hdr_locked(lseg->pls_layout);
244 }
245 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
246}
247
248static void
249put_lseg(struct pnfs_layout_segment *lseg)
250{
251 struct inode *inode;
252
253 if (!lseg)
254 return;
255
Fred Isaman4541d162011-01-06 11:36:23 +0000256 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
257 atomic_read(&lseg->pls_refcount),
258 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000259 inode = lseg->pls_layout->plh_inode;
260 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
261 LIST_HEAD(free_me);
Andy Adamson974cec82010-10-20 00:18:02 -0400262
Fred Isamand684d2a2011-03-01 01:34:13 +0000263 put_lseg_common(lseg);
264 list_add(&lseg->pls_list, &free_me);
265 spin_unlock(&inode->i_lock);
266 pnfs_free_lseg_list(&free_me);
Fred Isaman4541d162011-01-06 11:36:23 +0000267 }
Andy Adamson974cec82010-10-20 00:18:02 -0400268}
269
Fred Isaman4541d162011-01-06 11:36:23 +0000270static bool
271should_free_lseg(u32 lseg_iomode, u32 recall_iomode)
272{
273 return (recall_iomode == IOMODE_ANY ||
274 lseg_iomode == recall_iomode);
275}
276
277/* Returns 1 if lseg is removed from list, 0 otherwise */
278static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
279 struct list_head *tmp_list)
280{
281 int rv = 0;
282
283 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
284 /* Remove the reference keeping the lseg in the
285 * list. It will now be removed when all
286 * outstanding io is finished.
287 */
Fred Isamand684d2a2011-03-01 01:34:13 +0000288 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
289 atomic_read(&lseg->pls_refcount));
290 if (atomic_dec_and_test(&lseg->pls_refcount)) {
291 put_lseg_common(lseg);
292 list_add(&lseg->pls_list, tmp_list);
293 rv = 1;
294 }
Fred Isaman4541d162011-01-06 11:36:23 +0000295 }
296 return rv;
297}
298
299/* Returns count of number of matching invalid lsegs remaining in list
300 * after call.
301 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000302int
Fred Isaman4541d162011-01-06 11:36:23 +0000303mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
304 struct list_head *tmp_list,
305 u32 iomode)
Andy Adamson974cec82010-10-20 00:18:02 -0400306{
307 struct pnfs_layout_segment *lseg, *next;
Fred Isaman4541d162011-01-06 11:36:23 +0000308 int invalid = 0, removed = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400309
310 dprintk("%s:Begin lo %p\n", __func__, lo);
311
Fred Isaman38511722011-02-03 18:28:50 +0000312 if (list_empty(&lo->plh_segs)) {
313 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
314 put_layout_hdr_locked(lo);
315 return 0;
316 }
Fred Isaman4541d162011-01-06 11:36:23 +0000317 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
318 if (should_free_lseg(lseg->pls_range.iomode, iomode)) {
319 dprintk("%s: freeing lseg %p iomode %d "
320 "offset %llu length %llu\n", __func__,
321 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
322 lseg->pls_range.length);
323 invalid++;
324 removed += mark_lseg_invalid(lseg, tmp_list);
325 }
326 dprintk("%s:Return %i\n", __func__, invalid - removed);
327 return invalid - removed;
Andy Adamson974cec82010-10-20 00:18:02 -0400328}
329
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000330/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000331void
Fred Isaman4541d162011-01-06 11:36:23 +0000332pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400333{
Fred Isaman4541d162011-01-06 11:36:23 +0000334 struct pnfs_layout_segment *lseg, *tmp;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000335 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400336
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000337 if (list_empty(free_me))
338 return;
339
340 lo = list_first_entry(free_me, struct pnfs_layout_segment,
341 pls_list)->pls_layout;
342
343 if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
344 struct nfs_client *clp;
345
346 clp = NFS_SERVER(lo->plh_inode)->nfs_client;
347 spin_lock(&clp->cl_lock);
348 list_del_init(&lo->plh_layouts);
349 spin_unlock(&clp->cl_lock);
350 }
Fred Isaman4541d162011-01-06 11:36:23 +0000351 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000352 list_del(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000353 free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400354 }
355}
356
Benny Halevye5e94012010-10-20 00:18:01 -0400357void
358pnfs_destroy_layout(struct nfs_inode *nfsi)
359{
360 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400361 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400362
363 spin_lock(&nfsi->vfs_inode.i_lock);
364 lo = nfsi->layout;
365 if (lo) {
Fred Isaman38511722011-02-03 18:28:50 +0000366 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
Fred Isaman4541d162011-01-06 11:36:23 +0000367 mark_matching_lsegs_invalid(lo, &tmp_list, IOMODE_ANY);
Benny Halevye5e94012010-10-20 00:18:01 -0400368 }
369 spin_unlock(&nfsi->vfs_inode.i_lock);
Andy Adamson974cec82010-10-20 00:18:02 -0400370 pnfs_free_lseg_list(&tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400371}
372
Andy Adamson974cec82010-10-20 00:18:02 -0400373/*
374 * Called by the state manger to remove all layouts established under an
375 * expired lease.
376 */
377void
378pnfs_destroy_all_layouts(struct nfs_client *clp)
379{
380 struct pnfs_layout_hdr *lo;
381 LIST_HEAD(tmp_list);
382
383 spin_lock(&clp->cl_lock);
384 list_splice_init(&clp->cl_layouts, &tmp_list);
385 spin_unlock(&clp->cl_lock);
386
387 while (!list_empty(&tmp_list)) {
388 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000389 plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400390 dprintk("%s freeing layout for inode %lu\n", __func__,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000391 lo->plh_inode->i_ino);
392 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400393 }
394}
395
Fred Isamanfd6002e2011-01-06 11:36:22 +0000396/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000397void
398pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
399 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400400{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000401 u32 oldseq, newseq;
Andy Adamson974cec82010-10-20 00:18:02 -0400402
Fred Isamanfd6002e2011-01-06 11:36:22 +0000403 oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid);
404 newseq = be32_to_cpu(new->stateid.seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000405 if ((int)(newseq - oldseq) > 0) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000406 memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid));
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000407 if (update_barrier) {
408 u32 new_barrier = be32_to_cpu(new->stateid.seqid);
409
410 if ((int)(new_barrier - lo->plh_barrier))
411 lo->plh_barrier = new_barrier;
412 } else {
413 /* Because of wraparound, we want to keep the barrier
414 * "close" to the current seqids. It needs to be
415 * within 2**31 to count as "behind", so if it
416 * gets too near that limit, give us a litle leeway
417 * and bring it to within 2**30.
418 * NOTE - and yes, this is all unsigned arithmetic.
419 */
420 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
421 lo->plh_barrier = newseq - (1 << 30);
422 }
423 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400424}
425
Fred Isamancf7d63f2011-01-06 11:36:25 +0000426/* lget is set to 1 if called from inside send_layoutget call chain */
427static bool
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000428pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
429 int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000430{
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000431 if ((stateid) &&
432 (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0)
433 return true;
Fred Isamanf7e89172011-01-06 11:36:32 +0000434 return lo->plh_block_lgets ||
Fred Isaman38511722011-02-03 18:28:50 +0000435 test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
Fred Isamanf7e89172011-01-06 11:36:32 +0000436 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000437 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000438 (atomic_read(&lo->plh_outstanding) > lget));
439}
440
Fred Isamanfd6002e2011-01-06 11:36:22 +0000441int
442pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
443 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400444{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000445 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400446
447 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000448 spin_lock(&lo->plh_inode->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000449 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000450 status = -EAGAIN;
451 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000452 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400453
Fred Isamanfd6002e2011-01-06 11:36:22 +0000454 do {
455 seq = read_seqbegin(&open_state->seqlock);
456 memcpy(dst->data, open_state->stateid.data,
457 sizeof(open_state->stateid.data));
458 } while (read_seqretry(&open_state->seqlock, seq));
459 } else
460 memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data));
461 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400462 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000463 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400464}
465
466/*
467* Get layout from server.
468* for now, assume that whole file layouts are requested.
469* arg->offset: 0
470* arg->length: all ones
471*/
Benny Halevye5e94012010-10-20 00:18:01 -0400472static struct pnfs_layout_segment *
473send_layoutget(struct pnfs_layout_hdr *lo,
474 struct nfs_open_context *ctx,
475 u32 iomode)
476{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000477 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400478 struct nfs_server *server = NFS_SERVER(ino);
479 struct nfs4_layoutget *lgp;
480 struct pnfs_layout_segment *lseg = NULL;
Benny Halevye5e94012010-10-20 00:18:01 -0400481
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400482 dprintk("--> %s\n", __func__);
483
484 BUG_ON(ctx == NULL);
485 lgp = kzalloc(sizeof(*lgp), GFP_KERNEL);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000486 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400487 return NULL;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400488 lgp->args.minlength = NFS4_MAX_UINT64;
489 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
490 lgp->args.range.iomode = iomode;
491 lgp->args.range.offset = 0;
492 lgp->args.range.length = NFS4_MAX_UINT64;
493 lgp->args.type = server->pnfs_curr_ld->id;
494 lgp->args.inode = ino;
495 lgp->args.ctx = get_nfs_open_context(ctx);
496 lgp->lsegpp = &lseg;
497
498 /* Synchronously retrieve layout information from server and
499 * store in lseg.
500 */
501 nfs4_proc_layoutget(lgp);
502 if (!lseg) {
503 /* remember that LAYOUTGET failed and suspend trying */
Fred Isaman566052c2011-01-06 11:36:20 +0000504 set_bit(lo_fail_bit(iomode), &lo->plh_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400505 }
Andy Adamson974cec82010-10-20 00:18:02 -0400506 return lseg;
507}
508
Fred Isamanf7e89172011-01-06 11:36:32 +0000509bool pnfs_roc(struct inode *ino)
510{
511 struct pnfs_layout_hdr *lo;
512 struct pnfs_layout_segment *lseg, *tmp;
513 LIST_HEAD(tmp_list);
514 bool found = false;
515
516 spin_lock(&ino->i_lock);
517 lo = NFS_I(ino)->layout;
518 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
519 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
520 goto out_nolayout;
521 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
522 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
523 mark_lseg_invalid(lseg, &tmp_list);
524 found = true;
525 }
526 if (!found)
527 goto out_nolayout;
528 lo->plh_block_lgets++;
529 get_layout_hdr(lo); /* matched in pnfs_roc_release */
530 spin_unlock(&ino->i_lock);
531 pnfs_free_lseg_list(&tmp_list);
532 return true;
533
534out_nolayout:
535 spin_unlock(&ino->i_lock);
536 return false;
537}
538
539void pnfs_roc_release(struct inode *ino)
540{
541 struct pnfs_layout_hdr *lo;
542
543 spin_lock(&ino->i_lock);
544 lo = NFS_I(ino)->layout;
545 lo->plh_block_lgets--;
546 put_layout_hdr_locked(lo);
547 spin_unlock(&ino->i_lock);
548}
549
550void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
551{
552 struct pnfs_layout_hdr *lo;
553
554 spin_lock(&ino->i_lock);
555 lo = NFS_I(ino)->layout;
556 if ((int)(barrier - lo->plh_barrier) > 0)
557 lo->plh_barrier = barrier;
558 spin_unlock(&ino->i_lock);
559}
560
561bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
562{
563 struct nfs_inode *nfsi = NFS_I(ino);
564 struct pnfs_layout_segment *lseg;
565 bool found = false;
566
567 spin_lock(&ino->i_lock);
568 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
569 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
570 found = true;
571 break;
572 }
573 if (!found) {
574 struct pnfs_layout_hdr *lo = nfsi->layout;
575 u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid);
576
577 /* Since close does not return a layout stateid for use as
578 * a barrier, we choose the worst-case barrier.
579 */
580 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
581 }
582 spin_unlock(&ino->i_lock);
583 return found;
584}
585
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400586/*
587 * Compare two layout segments for sorting into layout cache.
588 * We want to preferentially return RW over RO layouts, so ensure those
589 * are seen first.
590 */
591static s64
592cmp_layout(u32 iomode1, u32 iomode2)
593{
594 /* read > read/write */
595 return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ);
596}
597
Andy Adamson974cec82010-10-20 00:18:02 -0400598static void
599pnfs_insert_layout(struct pnfs_layout_hdr *lo,
600 struct pnfs_layout_segment *lseg)
601{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400602 struct pnfs_layout_segment *lp;
603 int found = 0;
604
Andy Adamson974cec82010-10-20 00:18:02 -0400605 dprintk("%s:Begin\n", __func__);
606
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000607 assert_spin_locked(&lo->plh_inode->i_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000608 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000609 if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400610 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000611 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400612 dprintk("%s: inserted lseg %p "
613 "iomode %d offset %llu length %llu before "
614 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000615 __func__, lseg, lseg->pls_range.iomode,
616 lseg->pls_range.offset, lseg->pls_range.length,
617 lp, lp->pls_range.iomode, lp->pls_range.offset,
618 lp->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400619 found = 1;
620 break;
Andy Adamson974cec82010-10-20 00:18:02 -0400621 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400622 if (!found) {
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000623 list_add_tail(&lseg->pls_list, &lo->plh_segs);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400624 dprintk("%s: inserted lseg %p "
625 "iomode %d offset %llu length %llu at tail\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000626 __func__, lseg, lseg->pls_range.iomode,
627 lseg->pls_range.offset, lseg->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400628 }
Fred Isamancc6e5342011-01-06 11:36:28 +0000629 get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400630
631 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400632}
633
634static struct pnfs_layout_hdr *
635alloc_init_layout_hdr(struct inode *ino)
636{
637 struct pnfs_layout_hdr *lo;
638
639 lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL);
640 if (!lo)
641 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +0000642 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000643 INIT_LIST_HEAD(&lo->plh_layouts);
644 INIT_LIST_HEAD(&lo->plh_segs);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000645 INIT_LIST_HEAD(&lo->plh_bulk_recall);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000646 lo->plh_inode = ino;
Benny Halevye5e94012010-10-20 00:18:01 -0400647 return lo;
648}
649
650static struct pnfs_layout_hdr *
651pnfs_find_alloc_layout(struct inode *ino)
652{
653 struct nfs_inode *nfsi = NFS_I(ino);
654 struct pnfs_layout_hdr *new = NULL;
655
656 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
657
658 assert_spin_locked(&ino->i_lock);
Fred Isaman4541d162011-01-06 11:36:23 +0000659 if (nfsi->layout) {
660 if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
661 return NULL;
662 else
663 return nfsi->layout;
664 }
Benny Halevye5e94012010-10-20 00:18:01 -0400665 spin_unlock(&ino->i_lock);
666 new = alloc_init_layout_hdr(ino);
667 spin_lock(&ino->i_lock);
668
669 if (likely(nfsi->layout == NULL)) /* Won the race? */
670 nfsi->layout = new;
671 else
672 kfree(new);
673 return nfsi->layout;
674}
675
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400676/*
677 * iomode matching rules:
678 * iomode lseg match
679 * ----- ----- -----
680 * ANY READ true
681 * ANY RW true
682 * RW READ false
683 * RW RW true
684 * READ READ true
685 * READ RW true
686 */
687static int
688is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode)
689{
Fred Isaman566052c2011-01-06 11:36:20 +0000690 return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400691}
692
693/*
694 * lookup range in layout
695 */
Benny Halevye5e94012010-10-20 00:18:01 -0400696static struct pnfs_layout_segment *
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000697pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode)
Benny Halevye5e94012010-10-20 00:18:01 -0400698{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400699 struct pnfs_layout_segment *lseg, *ret = NULL;
700
701 dprintk("%s:Begin\n", __func__);
702
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000703 assert_spin_locked(&lo->plh_inode->i_lock);
704 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +0000705 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
706 is_matching_lseg(lseg, iomode)) {
Fred Isamand684d2a2011-03-01 01:34:13 +0000707 ret = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400708 break;
709 }
Fred Isaman566052c2011-01-06 11:36:20 +0000710 if (cmp_layout(iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400711 break;
712 }
713
714 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +0000715 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400716 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400717}
718
719/*
720 * Layout segment is retreived from the server if not cached.
721 * The appropriate layout segment is referenced and returned to the caller.
722 */
723struct pnfs_layout_segment *
724pnfs_update_layout(struct inode *ino,
725 struct nfs_open_context *ctx,
726 enum pnfs_iomode iomode)
727{
728 struct nfs_inode *nfsi = NFS_I(ino);
Fred Isaman2130ff62011-01-06 11:36:26 +0000729 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -0400730 struct pnfs_layout_hdr *lo;
731 struct pnfs_layout_segment *lseg = NULL;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000732 bool first = false;
Benny Halevye5e94012010-10-20 00:18:01 -0400733
734 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
735 return NULL;
736 spin_lock(&ino->i_lock);
737 lo = pnfs_find_alloc_layout(ino);
738 if (lo == NULL) {
739 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
740 goto out_unlock;
741 }
742
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000743 /* Do we even need to bother with this? */
744 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
745 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
746 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400747 goto out_unlock;
748 }
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000749 /* Check to see if the layout for the given range already exists */
750 lseg = pnfs_find_lseg(lo, iomode);
751 if (lseg)
752 goto out_unlock;
Benny Halevye5e94012010-10-20 00:18:01 -0400753
754 /* if LAYOUTGET already failed once we don't try again */
Fred Isaman566052c2011-01-06 11:36:20 +0000755 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
Benny Halevye5e94012010-10-20 00:18:01 -0400756 goto out_unlock;
757
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000758 if (pnfs_layoutgets_blocked(lo, NULL, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +0000759 goto out_unlock;
760 atomic_inc(&lo->plh_outstanding);
761
Fred Isamancc6e5342011-01-06 11:36:28 +0000762 get_layout_hdr(lo);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000763 if (list_empty(&lo->plh_segs))
764 first = true;
765 spin_unlock(&ino->i_lock);
766 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +0000767 /* The lo must be on the clp list if there is any
768 * chance of a CB_LAYOUTRECALL(FILE) coming in.
769 */
770 spin_lock(&clp->cl_lock);
771 BUG_ON(!list_empty(&lo->plh_layouts));
772 list_add_tail(&lo->plh_layouts, &clp->cl_layouts);
773 spin_unlock(&clp->cl_lock);
774 }
Benny Halevye5e94012010-10-20 00:18:01 -0400775
776 lseg = send_layoutget(lo, ctx, iomode);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000777 if (!lseg && first) {
778 spin_lock(&clp->cl_lock);
779 list_del_init(&lo->plh_layouts);
780 spin_unlock(&clp->cl_lock);
Fred Isaman2130ff62011-01-06 11:36:26 +0000781 }
Fred Isamancf7d63f2011-01-06 11:36:25 +0000782 atomic_dec(&lo->plh_outstanding);
Fred Isamancc6e5342011-01-06 11:36:28 +0000783 put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400784out:
785 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
Andy Adamsonbf9c1382011-03-01 01:34:07 +0000786 nfsi->layout ? nfsi->layout->plh_flags : -1, lseg);
Fred Isamand684d2a2011-03-01 01:34:13 +0000787 put_lseg(lseg); /* STUB - callers currently ignore return value */
Benny Halevye5e94012010-10-20 00:18:01 -0400788 return lseg;
789out_unlock:
790 spin_unlock(&ino->i_lock);
791 goto out;
792}
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400793
794int
795pnfs_layout_process(struct nfs4_layoutget *lgp)
796{
797 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
798 struct nfs4_layoutget_res *res = &lgp->res;
799 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000800 struct inode *ino = lo->plh_inode;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000801 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400802 int status = 0;
803
Fred Isamanfc1794c2011-01-06 11:36:27 +0000804 /* Verify we got what we asked for.
805 * Note that because the xdr parsing only accepts a single
806 * element array, this can fail even if the server is behaving
807 * correctly.
808 */
809 if (lgp->args.range.iomode > res->range.iomode ||
810 res->range.offset != 0 ||
811 res->range.length != NFS4_MAX_UINT64) {
812 status = -EINVAL;
813 goto out;
814 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400815 /* Inject layout blob into I/O device driver */
816 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res);
817 if (!lseg || IS_ERR(lseg)) {
818 if (!lseg)
819 status = -ENOMEM;
820 else
821 status = PTR_ERR(lseg);
822 dprintk("%s: Could not allocate layout: error %d\n",
823 __func__, status);
824 goto out;
825 }
826
827 spin_lock(&ino->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000828 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
829 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
830 dprintk("%s forget reply due to recall\n", __func__);
831 goto out_forget_reply;
832 }
833
834 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
835 dprintk("%s forget reply due to state\n", __func__);
836 goto out_forget_reply;
837 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400838 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +0000839 lseg->pls_range = res->range;
Fred Isamand684d2a2011-03-01 01:34:13 +0000840 *lgp->lsegpp = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400841 pnfs_insert_layout(lo, lseg);
842
Fred Isamanf7e89172011-01-06 11:36:32 +0000843 if (res->return_on_close) {
844 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
845 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
846 }
847
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400848 /* Done processing layoutget. Set the layout stateid */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000849 pnfs_set_layout_stateid(lo, &res->stateid, false);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400850 spin_unlock(&ino->i_lock);
851out:
852 return status;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000853
854out_forget_reply:
855 spin_unlock(&ino->i_lock);
856 lseg->pls_layout = lo;
857 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
858 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400859}
860
Fred Isaman94ad1c82011-03-01 01:34:14 +0000861static void
862pnfs_set_pg_test(struct inode *inode, struct nfs_pageio_descriptor *pgio)
863{
864 struct pnfs_layoutdriver_type *ld;
865
866 ld = NFS_SERVER(inode)->pnfs_curr_ld;
867 pgio->pg_test = (ld ? ld->pg_test : NULL);
868}
869
870void
871pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio,
872 struct inode *inode)
873{
874 pnfs_set_pg_test(inode, pgio);
875}
876
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400877/*
878 * Device ID cache. Currently supports one layout type per struct nfs_client.
879 * Add layout type to the lookup key to expand to support multiple types.
880 */
881int
882pnfs_alloc_init_deviceid_cache(struct nfs_client *clp,
883 void (*free_callback)(struct pnfs_deviceid_node *))
884{
885 struct pnfs_deviceid_cache *c;
886
887 c = kzalloc(sizeof(struct pnfs_deviceid_cache), GFP_KERNEL);
888 if (!c)
889 return -ENOMEM;
890 spin_lock(&clp->cl_lock);
891 if (clp->cl_devid_cache != NULL) {
892 atomic_inc(&clp->cl_devid_cache->dc_ref);
893 dprintk("%s [kref [%d]]\n", __func__,
894 atomic_read(&clp->cl_devid_cache->dc_ref));
895 kfree(c);
896 } else {
897 /* kzalloc initializes hlists */
898 spin_lock_init(&c->dc_lock);
899 atomic_set(&c->dc_ref, 1);
900 c->dc_free_callback = free_callback;
901 clp->cl_devid_cache = c;
902 dprintk("%s [new]\n", __func__);
903 }
904 spin_unlock(&clp->cl_lock);
905 return 0;
906}
907EXPORT_SYMBOL_GPL(pnfs_alloc_init_deviceid_cache);
908
909/*
910 * Called from pnfs_layoutdriver_type->free_lseg
911 * last layout segment reference frees deviceid
912 */
913void
914pnfs_put_deviceid(struct pnfs_deviceid_cache *c,
915 struct pnfs_deviceid_node *devid)
916{
917 struct nfs4_deviceid *id = &devid->de_id;
918 struct pnfs_deviceid_node *d;
919 struct hlist_node *n;
920 long h = nfs4_deviceid_hash(id);
921
922 dprintk("%s [%d]\n", __func__, atomic_read(&devid->de_ref));
923 if (!atomic_dec_and_lock(&devid->de_ref, &c->dc_lock))
924 return;
925
926 hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[h], de_node)
927 if (!memcmp(&d->de_id, id, sizeof(*id))) {
928 hlist_del_rcu(&d->de_node);
929 spin_unlock(&c->dc_lock);
930 synchronize_rcu();
931 c->dc_free_callback(devid);
932 return;
933 }
934 spin_unlock(&c->dc_lock);
935 /* Why wasn't it found in the list? */
936 BUG();
937}
938EXPORT_SYMBOL_GPL(pnfs_put_deviceid);
939
940/* Find and reference a deviceid */
941struct pnfs_deviceid_node *
942pnfs_find_get_deviceid(struct pnfs_deviceid_cache *c, struct nfs4_deviceid *id)
943{
944 struct pnfs_deviceid_node *d;
945 struct hlist_node *n;
946 long hash = nfs4_deviceid_hash(id);
947
948 dprintk("--> %s hash %ld\n", __func__, hash);
949 rcu_read_lock();
950 hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) {
951 if (!memcmp(&d->de_id, id, sizeof(*id))) {
952 if (!atomic_inc_not_zero(&d->de_ref)) {
953 goto fail;
954 } else {
955 rcu_read_unlock();
956 return d;
957 }
958 }
959 }
960fail:
961 rcu_read_unlock();
962 return NULL;
963}
964EXPORT_SYMBOL_GPL(pnfs_find_get_deviceid);
965
966/*
967 * Add a deviceid to the cache.
968 * GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new
969 */
970struct pnfs_deviceid_node *
971pnfs_add_deviceid(struct pnfs_deviceid_cache *c, struct pnfs_deviceid_node *new)
972{
973 struct pnfs_deviceid_node *d;
974 long hash = nfs4_deviceid_hash(&new->de_id);
975
976 dprintk("--> %s hash %ld\n", __func__, hash);
977 spin_lock(&c->dc_lock);
978 d = pnfs_find_get_deviceid(c, &new->de_id);
979 if (d) {
980 spin_unlock(&c->dc_lock);
981 dprintk("%s [discard]\n", __func__);
982 c->dc_free_callback(new);
983 return d;
984 }
985 INIT_HLIST_NODE(&new->de_node);
986 atomic_set(&new->de_ref, 1);
987 hlist_add_head_rcu(&new->de_node, &c->dc_deviceids[hash]);
988 spin_unlock(&c->dc_lock);
989 dprintk("%s [new]\n", __func__);
990 return new;
991}
992EXPORT_SYMBOL_GPL(pnfs_add_deviceid);
993
994void
995pnfs_put_deviceid_cache(struct nfs_client *clp)
996{
997 struct pnfs_deviceid_cache *local = clp->cl_devid_cache;
998
Andy Adamsonb2a28972011-01-25 15:38:03 +0000999 dprintk("--> %s ({%d})\n", __func__, atomic_read(&local->dc_ref));
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001000 if (atomic_dec_and_lock(&local->dc_ref, &clp->cl_lock)) {
1001 int i;
1002 /* Verify cache is empty */
1003 for (i = 0; i < NFS4_DEVICE_ID_HASH_SIZE; i++)
1004 BUG_ON(!hlist_empty(&local->dc_deviceids[i]));
1005 clp->cl_devid_cache = NULL;
1006 spin_unlock(&clp->cl_lock);
1007 kfree(local);
1008 }
1009}
1010EXPORT_SYMBOL_GPL(pnfs_put_deviceid_cache);