blob: 6736f9e4f2e114a715f64476906e68dcb1f634b3 [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
Benny Halevye5e94012010-10-20 00:18:01 -0400180static void
181get_layout_hdr_locked(struct pnfs_layout_hdr *lo)
182{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000183 assert_spin_locked(&lo->plh_inode->i_lock);
184 lo->plh_refcount++;
Benny Halevye5e94012010-10-20 00:18:01 -0400185}
186
187static void
188put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
189{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000190 assert_spin_locked(&lo->plh_inode->i_lock);
191 BUG_ON(lo->plh_refcount == 0);
Benny Halevye5e94012010-10-20 00:18:01 -0400192
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000193 lo->plh_refcount--;
194 if (!lo->plh_refcount) {
Benny Halevye5e94012010-10-20 00:18:01 -0400195 dprintk("%s: freeing layout cache %p\n", __func__, lo);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000196 BUG_ON(!list_empty(&lo->plh_layouts));
197 NFS_I(lo->plh_inode)->layout = NULL;
Benny Halevye5e94012010-10-20 00:18:01 -0400198 kfree(lo);
199 }
200}
201
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400202void
Andy Adamson974cec82010-10-20 00:18:02 -0400203put_layout_hdr(struct inode *inode)
204{
205 spin_lock(&inode->i_lock);
206 put_layout_hdr_locked(NFS_I(inode)->layout);
207 spin_unlock(&inode->i_lock);
208}
209
210static void
211init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
212{
Fred Isaman566052c2011-01-06 11:36:20 +0000213 INIT_LIST_HEAD(&lseg->pls_list);
214 kref_init(&lseg->pls_refcount);
215 lseg->pls_layout = lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400216}
217
218/* Called without i_lock held, as the free_lseg call may sleep */
219static void
220destroy_lseg(struct kref *kref)
221{
222 struct pnfs_layout_segment *lseg =
Fred Isaman566052c2011-01-06 11:36:20 +0000223 container_of(kref, struct pnfs_layout_segment, pls_refcount);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000224 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400225
226 dprintk("--> %s\n", __func__);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400227 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Fred Isaman52fabd72011-01-06 11:36:18 +0000228 /* Matched by get_layout_hdr in pnfs_insert_layout */
Andy Adamson974cec82010-10-20 00:18:02 -0400229 put_layout_hdr(ino);
230}
231
232static void
233put_lseg(struct pnfs_layout_segment *lseg)
234{
235 if (!lseg)
236 return;
237
238 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
Fred Isaman566052c2011-01-06 11:36:20 +0000239 atomic_read(&lseg->pls_refcount.refcount));
240 kref_put(&lseg->pls_refcount, destroy_lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400241}
242
243static void
244pnfs_clear_lseg_list(struct pnfs_layout_hdr *lo, struct list_head *tmp_list)
245{
246 struct pnfs_layout_segment *lseg, *next;
247 struct nfs_client *clp;
248
249 dprintk("%s:Begin lo %p\n", __func__, lo);
250
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000251 assert_spin_locked(&lo->plh_inode->i_lock);
252 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
Andy Adamson974cec82010-10-20 00:18:02 -0400253 dprintk("%s: freeing lseg %p\n", __func__, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +0000254 list_move(&lseg->pls_list, tmp_list);
Andy Adamson974cec82010-10-20 00:18:02 -0400255 }
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000256 clp = NFS_SERVER(lo->plh_inode)->nfs_client;
Andy Adamson974cec82010-10-20 00:18:02 -0400257 spin_lock(&clp->cl_lock);
258 /* List does not take a reference, so no need for put here */
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000259 list_del_init(&lo->plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400260 spin_unlock(&clp->cl_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000261 write_seqlock(&lo->plh_seqlock);
Fred Isaman566052c2011-01-06 11:36:20 +0000262 clear_bit(NFS_LAYOUT_STATEID_SET, &lo->plh_flags);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000263 write_sequnlock(&lo->plh_seqlock);
Andy Adamson974cec82010-10-20 00:18:02 -0400264
265 dprintk("%s:Return\n", __func__);
266}
267
268static void
269pnfs_free_lseg_list(struct list_head *tmp_list)
270{
271 struct pnfs_layout_segment *lseg;
272
273 while (!list_empty(tmp_list)) {
274 lseg = list_entry(tmp_list->next, struct pnfs_layout_segment,
Fred Isaman566052c2011-01-06 11:36:20 +0000275 pls_list);
Andy Adamson974cec82010-10-20 00:18:02 -0400276 dprintk("%s calling put_lseg on %p\n", __func__, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +0000277 list_del(&lseg->pls_list);
Andy Adamson974cec82010-10-20 00:18:02 -0400278 put_lseg(lseg);
279 }
280}
281
Benny Halevye5e94012010-10-20 00:18:01 -0400282void
283pnfs_destroy_layout(struct nfs_inode *nfsi)
284{
285 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400286 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400287
288 spin_lock(&nfsi->vfs_inode.i_lock);
289 lo = nfsi->layout;
290 if (lo) {
Andy Adamson974cec82010-10-20 00:18:02 -0400291 pnfs_clear_lseg_list(lo, &tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400292 /* Matched by refcount set to 1 in alloc_init_layout_hdr */
293 put_layout_hdr_locked(lo);
294 }
295 spin_unlock(&nfsi->vfs_inode.i_lock);
Andy Adamson974cec82010-10-20 00:18:02 -0400296 pnfs_free_lseg_list(&tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400297}
298
Andy Adamson974cec82010-10-20 00:18:02 -0400299/*
300 * Called by the state manger to remove all layouts established under an
301 * expired lease.
302 */
303void
304pnfs_destroy_all_layouts(struct nfs_client *clp)
305{
306 struct pnfs_layout_hdr *lo;
307 LIST_HEAD(tmp_list);
308
309 spin_lock(&clp->cl_lock);
310 list_splice_init(&clp->cl_layouts, &tmp_list);
311 spin_unlock(&clp->cl_lock);
312
313 while (!list_empty(&tmp_list)) {
314 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000315 plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400316 dprintk("%s freeing layout for inode %lu\n", __func__,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000317 lo->plh_inode->i_ino);
318 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400319 }
320}
321
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000322/* update lo->plh_stateid with new if is more recent
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400323 *
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000324 * lo->plh_stateid could be the open stateid, in which case we just use what given.
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400325 */
326static void
327pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo,
328 const nfs4_stateid *new)
329{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000330 nfs4_stateid *old = &lo->plh_stateid;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400331 bool overwrite = false;
Andy Adamson974cec82010-10-20 00:18:02 -0400332
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000333 write_seqlock(&lo->plh_seqlock);
Fred Isaman566052c2011-01-06 11:36:20 +0000334 if (!test_bit(NFS_LAYOUT_STATEID_SET, &lo->plh_flags) ||
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400335 memcmp(old->stateid.other, new->stateid.other, sizeof(new->stateid.other)))
336 overwrite = true;
337 else {
338 u32 oldseq, newseq;
339
340 oldseq = be32_to_cpu(old->stateid.seqid);
341 newseq = be32_to_cpu(new->stateid.seqid);
342 if ((int)(newseq - oldseq) > 0)
343 overwrite = true;
344 }
345 if (overwrite)
346 memcpy(&old->stateid, &new->stateid, sizeof(new->stateid));
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000347 write_sequnlock(&lo->plh_seqlock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400348}
349
350static void
351pnfs_layout_from_open_stateid(struct pnfs_layout_hdr *lo,
352 struct nfs4_state *state)
353{
354 int seq;
355
356 dprintk("--> %s\n", __func__);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000357 write_seqlock(&lo->plh_seqlock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400358 do {
359 seq = read_seqbegin(&state->seqlock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000360 memcpy(lo->plh_stateid.data, state->stateid.data,
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400361 sizeof(state->stateid.data));
362 } while (read_seqretry(&state->seqlock, seq));
Fred Isaman566052c2011-01-06 11:36:20 +0000363 set_bit(NFS_LAYOUT_STATEID_SET, &lo->plh_flags);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000364 write_sequnlock(&lo->plh_seqlock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400365 dprintk("<-- %s\n", __func__);
366}
367
368void
369pnfs_get_layout_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
370 struct nfs4_state *open_state)
371{
372 int seq;
373
374 dprintk("--> %s\n", __func__);
375 do {
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000376 seq = read_seqbegin(&lo->plh_seqlock);
Fred Isaman566052c2011-01-06 11:36:20 +0000377 if (!test_bit(NFS_LAYOUT_STATEID_SET, &lo->plh_flags)) {
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400378 /* This will trigger retry of the read */
379 pnfs_layout_from_open_stateid(lo, open_state);
380 } else
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000381 memcpy(dst->data, lo->plh_stateid.data,
382 sizeof(lo->plh_stateid.data));
383 } while (read_seqretry(&lo->plh_seqlock, seq));
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400384 dprintk("<-- %s\n", __func__);
385}
386
387/*
388* Get layout from server.
389* for now, assume that whole file layouts are requested.
390* arg->offset: 0
391* arg->length: all ones
392*/
Benny Halevye5e94012010-10-20 00:18:01 -0400393static struct pnfs_layout_segment *
394send_layoutget(struct pnfs_layout_hdr *lo,
395 struct nfs_open_context *ctx,
396 u32 iomode)
397{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000398 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400399 struct nfs_server *server = NFS_SERVER(ino);
400 struct nfs4_layoutget *lgp;
401 struct pnfs_layout_segment *lseg = NULL;
Benny Halevye5e94012010-10-20 00:18:01 -0400402
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400403 dprintk("--> %s\n", __func__);
404
405 BUG_ON(ctx == NULL);
406 lgp = kzalloc(sizeof(*lgp), GFP_KERNEL);
407 if (lgp == NULL) {
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000408 put_layout_hdr(lo->plh_inode);
Andy Adamson974cec82010-10-20 00:18:02 -0400409 return NULL;
410 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400411 lgp->args.minlength = NFS4_MAX_UINT64;
412 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
413 lgp->args.range.iomode = iomode;
414 lgp->args.range.offset = 0;
415 lgp->args.range.length = NFS4_MAX_UINT64;
416 lgp->args.type = server->pnfs_curr_ld->id;
417 lgp->args.inode = ino;
418 lgp->args.ctx = get_nfs_open_context(ctx);
419 lgp->lsegpp = &lseg;
420
421 /* Synchronously retrieve layout information from server and
422 * store in lseg.
423 */
424 nfs4_proc_layoutget(lgp);
425 if (!lseg) {
426 /* remember that LAYOUTGET failed and suspend trying */
Fred Isaman566052c2011-01-06 11:36:20 +0000427 set_bit(lo_fail_bit(iomode), &lo->plh_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400428 }
Andy Adamson974cec82010-10-20 00:18:02 -0400429 return lseg;
430}
431
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400432/*
433 * Compare two layout segments for sorting into layout cache.
434 * We want to preferentially return RW over RO layouts, so ensure those
435 * are seen first.
436 */
437static s64
438cmp_layout(u32 iomode1, u32 iomode2)
439{
440 /* read > read/write */
441 return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ);
442}
443
Andy Adamson974cec82010-10-20 00:18:02 -0400444static void
445pnfs_insert_layout(struct pnfs_layout_hdr *lo,
446 struct pnfs_layout_segment *lseg)
447{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400448 struct pnfs_layout_segment *lp;
449 int found = 0;
450
Andy Adamson974cec82010-10-20 00:18:02 -0400451 dprintk("%s:Begin\n", __func__);
452
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000453 assert_spin_locked(&lo->plh_inode->i_lock);
454 if (list_empty(&lo->plh_segs)) {
455 struct nfs_client *clp = NFS_SERVER(lo->plh_inode)->nfs_client;
Andy Adamson974cec82010-10-20 00:18:02 -0400456
457 spin_lock(&clp->cl_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000458 BUG_ON(!list_empty(&lo->plh_layouts));
459 list_add_tail(&lo->plh_layouts, &clp->cl_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400460 spin_unlock(&clp->cl_lock);
461 }
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000462 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000463 if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400464 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000465 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400466 dprintk("%s: inserted lseg %p "
467 "iomode %d offset %llu length %llu before "
468 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000469 __func__, lseg, lseg->pls_range.iomode,
470 lseg->pls_range.offset, lseg->pls_range.length,
471 lp, lp->pls_range.iomode, lp->pls_range.offset,
472 lp->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400473 found = 1;
474 break;
Andy Adamson974cec82010-10-20 00:18:02 -0400475 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400476 if (!found) {
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000477 list_add_tail(&lseg->pls_list, &lo->plh_segs);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400478 dprintk("%s: inserted lseg %p "
479 "iomode %d offset %llu length %llu at tail\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000480 __func__, lseg, lseg->pls_range.iomode,
481 lseg->pls_range.offset, lseg->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400482 }
483 get_layout_hdr_locked(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400484
485 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400486}
487
488static struct pnfs_layout_hdr *
489alloc_init_layout_hdr(struct inode *ino)
490{
491 struct pnfs_layout_hdr *lo;
492
493 lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL);
494 if (!lo)
495 return NULL;
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000496 lo->plh_refcount = 1;
497 INIT_LIST_HEAD(&lo->plh_layouts);
498 INIT_LIST_HEAD(&lo->plh_segs);
499 seqlock_init(&lo->plh_seqlock);
500 lo->plh_inode = ino;
Benny Halevye5e94012010-10-20 00:18:01 -0400501 return lo;
502}
503
504static struct pnfs_layout_hdr *
505pnfs_find_alloc_layout(struct inode *ino)
506{
507 struct nfs_inode *nfsi = NFS_I(ino);
508 struct pnfs_layout_hdr *new = NULL;
509
510 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
511
512 assert_spin_locked(&ino->i_lock);
513 if (nfsi->layout)
514 return nfsi->layout;
515
516 spin_unlock(&ino->i_lock);
517 new = alloc_init_layout_hdr(ino);
518 spin_lock(&ino->i_lock);
519
520 if (likely(nfsi->layout == NULL)) /* Won the race? */
521 nfsi->layout = new;
522 else
523 kfree(new);
524 return nfsi->layout;
525}
526
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400527/*
528 * iomode matching rules:
529 * iomode lseg match
530 * ----- ----- -----
531 * ANY READ true
532 * ANY RW true
533 * RW READ false
534 * RW RW true
535 * READ READ true
536 * READ RW true
537 */
538static int
539is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode)
540{
Fred Isaman566052c2011-01-06 11:36:20 +0000541 return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400542}
543
544/*
545 * lookup range in layout
546 */
Benny Halevye5e94012010-10-20 00:18:01 -0400547static struct pnfs_layout_segment *
548pnfs_has_layout(struct pnfs_layout_hdr *lo, u32 iomode)
549{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400550 struct pnfs_layout_segment *lseg, *ret = NULL;
551
552 dprintk("%s:Begin\n", __func__);
553
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000554 assert_spin_locked(&lo->plh_inode->i_lock);
555 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400556 if (is_matching_lseg(lseg, iomode)) {
557 ret = lseg;
558 break;
559 }
Fred Isaman566052c2011-01-06 11:36:20 +0000560 if (cmp_layout(iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400561 break;
562 }
563
564 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000565 __func__, ret, ret ? atomic_read(&ret->pls_refcount.refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400566 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400567}
568
569/*
570 * Layout segment is retreived from the server if not cached.
571 * The appropriate layout segment is referenced and returned to the caller.
572 */
573struct pnfs_layout_segment *
574pnfs_update_layout(struct inode *ino,
575 struct nfs_open_context *ctx,
576 enum pnfs_iomode iomode)
577{
578 struct nfs_inode *nfsi = NFS_I(ino);
579 struct pnfs_layout_hdr *lo;
580 struct pnfs_layout_segment *lseg = NULL;
581
582 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
583 return NULL;
584 spin_lock(&ino->i_lock);
585 lo = pnfs_find_alloc_layout(ino);
586 if (lo == NULL) {
587 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
588 goto out_unlock;
589 }
590
591 /* Check to see if the layout for the given range already exists */
592 lseg = pnfs_has_layout(lo, iomode);
593 if (lseg) {
594 dprintk("%s: Using cached lseg %p for iomode %d)\n",
595 __func__, lseg, iomode);
596 goto out_unlock;
597 }
598
599 /* if LAYOUTGET already failed once we don't try again */
Fred Isaman566052c2011-01-06 11:36:20 +0000600 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
Benny Halevye5e94012010-10-20 00:18:01 -0400601 goto out_unlock;
602
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400603 get_layout_hdr_locked(lo); /* Matched in nfs4_layoutget_release */
Benny Halevye5e94012010-10-20 00:18:01 -0400604 spin_unlock(&ino->i_lock);
605
606 lseg = send_layoutget(lo, ctx, iomode);
607out:
608 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
Fred Isaman566052c2011-01-06 11:36:20 +0000609 nfsi->layout->plh_flags, lseg);
Benny Halevye5e94012010-10-20 00:18:01 -0400610 return lseg;
611out_unlock:
612 spin_unlock(&ino->i_lock);
613 goto out;
614}
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400615
616int
617pnfs_layout_process(struct nfs4_layoutget *lgp)
618{
619 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
620 struct nfs4_layoutget_res *res = &lgp->res;
621 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000622 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400623 int status = 0;
624
625 /* Inject layout blob into I/O device driver */
626 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res);
627 if (!lseg || IS_ERR(lseg)) {
628 if (!lseg)
629 status = -ENOMEM;
630 else
631 status = PTR_ERR(lseg);
632 dprintk("%s: Could not allocate layout: error %d\n",
633 __func__, status);
634 goto out;
635 }
636
637 spin_lock(&ino->i_lock);
638 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +0000639 lseg->pls_range = res->range;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400640 *lgp->lsegpp = lseg;
641 pnfs_insert_layout(lo, lseg);
642
643 /* Done processing layoutget. Set the layout stateid */
644 pnfs_set_layout_stateid(lo, &res->stateid);
645 spin_unlock(&ino->i_lock);
646out:
647 return status;
648}
649
650/*
651 * Device ID cache. Currently supports one layout type per struct nfs_client.
652 * Add layout type to the lookup key to expand to support multiple types.
653 */
654int
655pnfs_alloc_init_deviceid_cache(struct nfs_client *clp,
656 void (*free_callback)(struct pnfs_deviceid_node *))
657{
658 struct pnfs_deviceid_cache *c;
659
660 c = kzalloc(sizeof(struct pnfs_deviceid_cache), GFP_KERNEL);
661 if (!c)
662 return -ENOMEM;
663 spin_lock(&clp->cl_lock);
664 if (clp->cl_devid_cache != NULL) {
665 atomic_inc(&clp->cl_devid_cache->dc_ref);
666 dprintk("%s [kref [%d]]\n", __func__,
667 atomic_read(&clp->cl_devid_cache->dc_ref));
668 kfree(c);
669 } else {
670 /* kzalloc initializes hlists */
671 spin_lock_init(&c->dc_lock);
672 atomic_set(&c->dc_ref, 1);
673 c->dc_free_callback = free_callback;
674 clp->cl_devid_cache = c;
675 dprintk("%s [new]\n", __func__);
676 }
677 spin_unlock(&clp->cl_lock);
678 return 0;
679}
680EXPORT_SYMBOL_GPL(pnfs_alloc_init_deviceid_cache);
681
682/*
683 * Called from pnfs_layoutdriver_type->free_lseg
684 * last layout segment reference frees deviceid
685 */
686void
687pnfs_put_deviceid(struct pnfs_deviceid_cache *c,
688 struct pnfs_deviceid_node *devid)
689{
690 struct nfs4_deviceid *id = &devid->de_id;
691 struct pnfs_deviceid_node *d;
692 struct hlist_node *n;
693 long h = nfs4_deviceid_hash(id);
694
695 dprintk("%s [%d]\n", __func__, atomic_read(&devid->de_ref));
696 if (!atomic_dec_and_lock(&devid->de_ref, &c->dc_lock))
697 return;
698
699 hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[h], de_node)
700 if (!memcmp(&d->de_id, id, sizeof(*id))) {
701 hlist_del_rcu(&d->de_node);
702 spin_unlock(&c->dc_lock);
703 synchronize_rcu();
704 c->dc_free_callback(devid);
705 return;
706 }
707 spin_unlock(&c->dc_lock);
708 /* Why wasn't it found in the list? */
709 BUG();
710}
711EXPORT_SYMBOL_GPL(pnfs_put_deviceid);
712
713/* Find and reference a deviceid */
714struct pnfs_deviceid_node *
715pnfs_find_get_deviceid(struct pnfs_deviceid_cache *c, struct nfs4_deviceid *id)
716{
717 struct pnfs_deviceid_node *d;
718 struct hlist_node *n;
719 long hash = nfs4_deviceid_hash(id);
720
721 dprintk("--> %s hash %ld\n", __func__, hash);
722 rcu_read_lock();
723 hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) {
724 if (!memcmp(&d->de_id, id, sizeof(*id))) {
725 if (!atomic_inc_not_zero(&d->de_ref)) {
726 goto fail;
727 } else {
728 rcu_read_unlock();
729 return d;
730 }
731 }
732 }
733fail:
734 rcu_read_unlock();
735 return NULL;
736}
737EXPORT_SYMBOL_GPL(pnfs_find_get_deviceid);
738
739/*
740 * Add a deviceid to the cache.
741 * GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new
742 */
743struct pnfs_deviceid_node *
744pnfs_add_deviceid(struct pnfs_deviceid_cache *c, struct pnfs_deviceid_node *new)
745{
746 struct pnfs_deviceid_node *d;
747 long hash = nfs4_deviceid_hash(&new->de_id);
748
749 dprintk("--> %s hash %ld\n", __func__, hash);
750 spin_lock(&c->dc_lock);
751 d = pnfs_find_get_deviceid(c, &new->de_id);
752 if (d) {
753 spin_unlock(&c->dc_lock);
754 dprintk("%s [discard]\n", __func__);
755 c->dc_free_callback(new);
756 return d;
757 }
758 INIT_HLIST_NODE(&new->de_node);
759 atomic_set(&new->de_ref, 1);
760 hlist_add_head_rcu(&new->de_node, &c->dc_deviceids[hash]);
761 spin_unlock(&c->dc_lock);
762 dprintk("%s [new]\n", __func__);
763 return new;
764}
765EXPORT_SYMBOL_GPL(pnfs_add_deviceid);
766
767void
768pnfs_put_deviceid_cache(struct nfs_client *clp)
769{
770 struct pnfs_deviceid_cache *local = clp->cl_devid_cache;
771
772 dprintk("--> %s cl_devid_cache %p\n", __func__, clp->cl_devid_cache);
773 if (atomic_dec_and_lock(&local->dc_ref, &clp->cl_lock)) {
774 int i;
775 /* Verify cache is empty */
776 for (i = 0; i < NFS4_DEVICE_ID_HASH_SIZE; i++)
777 BUG_ON(!hlist_empty(&local->dc_deviceids[i]));
778 clp->cl_devid_cache = NULL;
779 spin_unlock(&clp->cl_lock);
780 kfree(local);
781 }
782}
783EXPORT_SYMBOL_GPL(pnfs_put_deviceid_cache);