blob: ff681ab65d31f54654521e560b9cd5780db8318f [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"
Andy Adamson64419a92011-03-01 01:34:16 +000033#include "iostat.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040034
35#define NFSDBG_FACILITY NFSDBG_PNFS
36
Fred Isaman02c35fc2010-10-20 00:17:59 -040037/* Locking:
38 *
39 * pnfs_spinlock:
40 * protects pnfs_modules_tbl.
41 */
42static DEFINE_SPINLOCK(pnfs_spinlock);
43
44/*
45 * pnfs_modules_tbl holds all pnfs modules
46 */
47static LIST_HEAD(pnfs_modules_tbl);
48
49/* Return the registered pnfs layout driver module matching given id */
50static struct pnfs_layoutdriver_type *
51find_pnfs_driver_locked(u32 id)
52{
53 struct pnfs_layoutdriver_type *local;
54
55 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
56 if (local->id == id)
57 goto out;
58 local = NULL;
59out:
60 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
61 return local;
62}
63
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040064static struct pnfs_layoutdriver_type *
65find_pnfs_driver(u32 id)
66{
Fred Isaman02c35fc2010-10-20 00:17:59 -040067 struct pnfs_layoutdriver_type *local;
68
69 spin_lock(&pnfs_spinlock);
70 local = find_pnfs_driver_locked(id);
71 spin_unlock(&pnfs_spinlock);
72 return local;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040073}
74
75void
76unset_pnfs_layoutdriver(struct nfs_server *nfss)
77{
Christoph Hellwigea8eecd2011-03-01 01:34:21 +000078 if (nfss->pnfs_curr_ld)
Fred Isaman02c35fc2010-10-20 00:17:59 -040079 module_put(nfss->pnfs_curr_ld->owner);
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040080 nfss->pnfs_curr_ld = NULL;
81}
82
83/*
84 * Try to set the server's pnfs module to the pnfs layout type specified by id.
85 * Currently only one pNFS layout driver per filesystem is supported.
86 *
87 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
88 */
89void
90set_pnfs_layoutdriver(struct nfs_server *server, u32 id)
91{
92 struct pnfs_layoutdriver_type *ld_type = NULL;
93
94 if (id == 0)
95 goto out_no_driver;
96 if (!(server->nfs_client->cl_exchange_flags &
97 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
98 printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__,
99 id, server->nfs_client->cl_exchange_flags);
100 goto out_no_driver;
101 }
102 ld_type = find_pnfs_driver(id);
103 if (!ld_type) {
104 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
105 ld_type = find_pnfs_driver(id);
106 if (!ld_type) {
107 dprintk("%s: No pNFS module found for %u.\n",
108 __func__, id);
109 goto out_no_driver;
110 }
111 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400112 if (!try_module_get(ld_type->owner)) {
113 dprintk("%s: Could not grab reference on module\n", __func__);
114 goto out_no_driver;
115 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400116 server->pnfs_curr_ld = ld_type;
Christoph Hellwigea8eecd2011-03-01 01:34:21 +0000117
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400118 dprintk("%s: pNFS module for %u set\n", __func__, id);
119 return;
120
121out_no_driver:
122 dprintk("%s: Using NFSv4 I/O\n", __func__);
123 server->pnfs_curr_ld = NULL;
124}
Fred Isaman02c35fc2010-10-20 00:17:59 -0400125
126int
127pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
128{
129 int status = -EINVAL;
130 struct pnfs_layoutdriver_type *tmp;
131
132 if (ld_type->id == 0) {
133 printk(KERN_ERR "%s id 0 is reserved\n", __func__);
134 return status;
135 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400136 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
137 printk(KERN_ERR "%s Layout driver must provide "
138 "alloc_lseg and free_lseg.\n", __func__);
139 return status;
140 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400141
142 spin_lock(&pnfs_spinlock);
143 tmp = find_pnfs_driver_locked(ld_type->id);
144 if (!tmp) {
145 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
146 status = 0;
147 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
148 ld_type->name);
149 } else {
150 printk(KERN_ERR "%s Module with id %d already loaded!\n",
151 __func__, ld_type->id);
152 }
153 spin_unlock(&pnfs_spinlock);
154
155 return status;
156}
157EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
158
159void
160pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
161{
162 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
163 spin_lock(&pnfs_spinlock);
164 list_del(&ld_type->pnfs_tblid);
165 spin_unlock(&pnfs_spinlock);
166}
167EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
Benny Halevye5e94012010-10-20 00:18:01 -0400168
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400169/*
170 * pNFS client layout cache
171 */
172
Fred Isamancc6e5342011-01-06 11:36:28 +0000173/* Need to hold i_lock if caller does not already hold reference */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000174void
Fred Isamancc6e5342011-01-06 11:36:28 +0000175get_layout_hdr(struct pnfs_layout_hdr *lo)
Benny Halevye5e94012010-10-20 00:18:01 -0400176{
Fred Isamancc6e5342011-01-06 11:36:28 +0000177 atomic_inc(&lo->plh_refcount);
178}
179
180static void
181destroy_layout_hdr(struct pnfs_layout_hdr *lo)
182{
183 dprintk("%s: freeing layout cache %p\n", __func__, lo);
184 BUG_ON(!list_empty(&lo->plh_layouts));
185 NFS_I(lo->plh_inode)->layout = NULL;
186 kfree(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400187}
188
189static void
190put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
191{
Fred Isamancc6e5342011-01-06 11:36:28 +0000192 if (atomic_dec_and_test(&lo->plh_refcount))
193 destroy_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400194}
195
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400196void
Fred Isamancc6e5342011-01-06 11:36:28 +0000197put_layout_hdr(struct pnfs_layout_hdr *lo)
Andy Adamson974cec82010-10-20 00:18:02 -0400198{
Fred Isamancc6e5342011-01-06 11:36:28 +0000199 struct inode *inode = lo->plh_inode;
200
201 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
202 destroy_layout_hdr(lo);
203 spin_unlock(&inode->i_lock);
204 }
Andy Adamson974cec82010-10-20 00:18:02 -0400205}
206
207static void
208init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
209{
Fred Isaman566052c2011-01-06 11:36:20 +0000210 INIT_LIST_HEAD(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000211 atomic_set(&lseg->pls_refcount, 1);
212 smp_mb();
213 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
Fred Isaman566052c2011-01-06 11:36:20 +0000214 lseg->pls_layout = lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400215}
216
Fred Isaman4541d162011-01-06 11:36:23 +0000217static void free_lseg(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400218{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000219 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400220
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400221 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Fred Isaman52fabd72011-01-06 11:36:18 +0000222 /* Matched by get_layout_hdr in pnfs_insert_layout */
Fred Isamancc6e5342011-01-06 11:36:28 +0000223 put_layout_hdr(NFS_I(ino)->layout);
Andy Adamson974cec82010-10-20 00:18:02 -0400224}
225
Fred Isamand684d2a2011-03-01 01:34:13 +0000226static void
227put_lseg_common(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400228{
Fred Isamand684d2a2011-03-01 01:34:13 +0000229 struct inode *inode = lseg->pls_layout->plh_inode;
230
231 BUG_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
232 list_del_init(&lseg->pls_list);
233 if (list_empty(&lseg->pls_layout->plh_segs)) {
234 set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
235 /* Matched by initial refcount set in alloc_init_layout_hdr */
236 put_layout_hdr_locked(lseg->pls_layout);
237 }
238 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
239}
240
Fred Isamanbae724e2011-03-01 01:34:15 +0000241void
Fred Isamand684d2a2011-03-01 01:34:13 +0000242put_lseg(struct pnfs_layout_segment *lseg)
243{
244 struct inode *inode;
245
246 if (!lseg)
247 return;
248
Fred Isaman4541d162011-01-06 11:36:23 +0000249 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
250 atomic_read(&lseg->pls_refcount),
251 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000252 inode = lseg->pls_layout->plh_inode;
253 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
254 LIST_HEAD(free_me);
Andy Adamson974cec82010-10-20 00:18:02 -0400255
Fred Isamand684d2a2011-03-01 01:34:13 +0000256 put_lseg_common(lseg);
257 list_add(&lseg->pls_list, &free_me);
258 spin_unlock(&inode->i_lock);
259 pnfs_free_lseg_list(&free_me);
Fred Isaman4541d162011-01-06 11:36:23 +0000260 }
Andy Adamson974cec82010-10-20 00:18:02 -0400261}
Fred Isamane0c2b382011-03-23 13:27:53 +0000262EXPORT_SYMBOL_GPL(put_lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400263
Fred Isaman4541d162011-01-06 11:36:23 +0000264static bool
265should_free_lseg(u32 lseg_iomode, u32 recall_iomode)
266{
267 return (recall_iomode == IOMODE_ANY ||
268 lseg_iomode == recall_iomode);
269}
270
271/* Returns 1 if lseg is removed from list, 0 otherwise */
272static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
273 struct list_head *tmp_list)
274{
275 int rv = 0;
276
277 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
278 /* Remove the reference keeping the lseg in the
279 * list. It will now be removed when all
280 * outstanding io is finished.
281 */
Fred Isamand684d2a2011-03-01 01:34:13 +0000282 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
283 atomic_read(&lseg->pls_refcount));
284 if (atomic_dec_and_test(&lseg->pls_refcount)) {
285 put_lseg_common(lseg);
286 list_add(&lseg->pls_list, tmp_list);
287 rv = 1;
288 }
Fred Isaman4541d162011-01-06 11:36:23 +0000289 }
290 return rv;
291}
292
293/* Returns count of number of matching invalid lsegs remaining in list
294 * after call.
295 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000296int
Fred Isaman4541d162011-01-06 11:36:23 +0000297mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
298 struct list_head *tmp_list,
299 u32 iomode)
Andy Adamson974cec82010-10-20 00:18:02 -0400300{
301 struct pnfs_layout_segment *lseg, *next;
Fred Isaman4541d162011-01-06 11:36:23 +0000302 int invalid = 0, removed = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400303
304 dprintk("%s:Begin lo %p\n", __func__, lo);
305
Fred Isaman38511722011-02-03 18:28:50 +0000306 if (list_empty(&lo->plh_segs)) {
307 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
308 put_layout_hdr_locked(lo);
309 return 0;
310 }
Fred Isaman4541d162011-01-06 11:36:23 +0000311 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
312 if (should_free_lseg(lseg->pls_range.iomode, iomode)) {
313 dprintk("%s: freeing lseg %p iomode %d "
314 "offset %llu length %llu\n", __func__,
315 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
316 lseg->pls_range.length);
317 invalid++;
318 removed += mark_lseg_invalid(lseg, tmp_list);
319 }
320 dprintk("%s:Return %i\n", __func__, invalid - removed);
321 return invalid - removed;
Andy Adamson974cec82010-10-20 00:18:02 -0400322}
323
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000324/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000325void
Fred Isaman4541d162011-01-06 11:36:23 +0000326pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400327{
Fred Isaman4541d162011-01-06 11:36:23 +0000328 struct pnfs_layout_segment *lseg, *tmp;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000329 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400330
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000331 if (list_empty(free_me))
332 return;
333
334 lo = list_first_entry(free_me, struct pnfs_layout_segment,
335 pls_list)->pls_layout;
336
337 if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
338 struct nfs_client *clp;
339
340 clp = NFS_SERVER(lo->plh_inode)->nfs_client;
341 spin_lock(&clp->cl_lock);
342 list_del_init(&lo->plh_layouts);
343 spin_unlock(&clp->cl_lock);
344 }
Fred Isaman4541d162011-01-06 11:36:23 +0000345 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000346 list_del(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000347 free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400348 }
349}
350
Benny Halevye5e94012010-10-20 00:18:01 -0400351void
352pnfs_destroy_layout(struct nfs_inode *nfsi)
353{
354 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400355 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400356
357 spin_lock(&nfsi->vfs_inode.i_lock);
358 lo = nfsi->layout;
359 if (lo) {
Fred Isaman38511722011-02-03 18:28:50 +0000360 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
Fred Isaman4541d162011-01-06 11:36:23 +0000361 mark_matching_lsegs_invalid(lo, &tmp_list, IOMODE_ANY);
Benny Halevye5e94012010-10-20 00:18:01 -0400362 }
363 spin_unlock(&nfsi->vfs_inode.i_lock);
Andy Adamson974cec82010-10-20 00:18:02 -0400364 pnfs_free_lseg_list(&tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400365}
366
Andy Adamson974cec82010-10-20 00:18:02 -0400367/*
368 * Called by the state manger to remove all layouts established under an
369 * expired lease.
370 */
371void
372pnfs_destroy_all_layouts(struct nfs_client *clp)
373{
374 struct pnfs_layout_hdr *lo;
375 LIST_HEAD(tmp_list);
376
377 spin_lock(&clp->cl_lock);
378 list_splice_init(&clp->cl_layouts, &tmp_list);
379 spin_unlock(&clp->cl_lock);
380
381 while (!list_empty(&tmp_list)) {
382 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000383 plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400384 dprintk("%s freeing layout for inode %lu\n", __func__,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000385 lo->plh_inode->i_ino);
386 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400387 }
388}
389
Fred Isamanfd6002e2011-01-06 11:36:22 +0000390/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000391void
392pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
393 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400394{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000395 u32 oldseq, newseq;
Andy Adamson974cec82010-10-20 00:18:02 -0400396
Fred Isamanfd6002e2011-01-06 11:36:22 +0000397 oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid);
398 newseq = be32_to_cpu(new->stateid.seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000399 if ((int)(newseq - oldseq) > 0) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000400 memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid));
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000401 if (update_barrier) {
402 u32 new_barrier = be32_to_cpu(new->stateid.seqid);
403
404 if ((int)(new_barrier - lo->plh_barrier))
405 lo->plh_barrier = new_barrier;
406 } else {
407 /* Because of wraparound, we want to keep the barrier
408 * "close" to the current seqids. It needs to be
409 * within 2**31 to count as "behind", so if it
410 * gets too near that limit, give us a litle leeway
411 * and bring it to within 2**30.
412 * NOTE - and yes, this is all unsigned arithmetic.
413 */
414 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
415 lo->plh_barrier = newseq - (1 << 30);
416 }
417 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400418}
419
Fred Isamancf7d63f2011-01-06 11:36:25 +0000420/* lget is set to 1 if called from inside send_layoutget call chain */
421static bool
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000422pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
423 int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000424{
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000425 if ((stateid) &&
426 (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0)
427 return true;
Fred Isamanf7e89172011-01-06 11:36:32 +0000428 return lo->plh_block_lgets ||
Fred Isaman38511722011-02-03 18:28:50 +0000429 test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
Fred Isamanf7e89172011-01-06 11:36:32 +0000430 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000431 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000432 (atomic_read(&lo->plh_outstanding) > lget));
433}
434
Fred Isamanfd6002e2011-01-06 11:36:22 +0000435int
436pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
437 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400438{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000439 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400440
441 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000442 spin_lock(&lo->plh_inode->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000443 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000444 status = -EAGAIN;
445 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000446 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400447
Fred Isamanfd6002e2011-01-06 11:36:22 +0000448 do {
449 seq = read_seqbegin(&open_state->seqlock);
450 memcpy(dst->data, open_state->stateid.data,
451 sizeof(open_state->stateid.data));
452 } while (read_seqretry(&open_state->seqlock, seq));
453 } else
454 memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data));
455 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400456 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000457 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400458}
459
460/*
461* Get layout from server.
462* for now, assume that whole file layouts are requested.
463* arg->offset: 0
464* arg->length: all ones
465*/
Benny Halevye5e94012010-10-20 00:18:01 -0400466static struct pnfs_layout_segment *
467send_layoutget(struct pnfs_layout_hdr *lo,
468 struct nfs_open_context *ctx,
469 u32 iomode)
470{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000471 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400472 struct nfs_server *server = NFS_SERVER(ino);
473 struct nfs4_layoutget *lgp;
474 struct pnfs_layout_segment *lseg = NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400475 struct page **pages = NULL;
476 int i;
477 u32 max_resp_sz, max_pages;
Benny Halevye5e94012010-10-20 00:18:01 -0400478
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400479 dprintk("--> %s\n", __func__);
480
481 BUG_ON(ctx == NULL);
482 lgp = kzalloc(sizeof(*lgp), GFP_KERNEL);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000483 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400484 return NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400485
486 /* allocate pages for xdr post processing */
487 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
488 max_pages = max_resp_sz >> PAGE_SHIFT;
489
490 pages = kzalloc(max_pages * sizeof(struct page *), GFP_KERNEL);
491 if (!pages)
492 goto out_err_free;
493
494 for (i = 0; i < max_pages; i++) {
495 pages[i] = alloc_page(GFP_KERNEL);
496 if (!pages[i])
497 goto out_err_free;
498 }
499
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400500 lgp->args.minlength = NFS4_MAX_UINT64;
501 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
502 lgp->args.range.iomode = iomode;
503 lgp->args.range.offset = 0;
504 lgp->args.range.length = NFS4_MAX_UINT64;
505 lgp->args.type = server->pnfs_curr_ld->id;
506 lgp->args.inode = ino;
507 lgp->args.ctx = get_nfs_open_context(ctx);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400508 lgp->args.layout.pages = pages;
509 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400510 lgp->lsegpp = &lseg;
511
512 /* Synchronously retrieve layout information from server and
513 * store in lseg.
514 */
515 nfs4_proc_layoutget(lgp);
516 if (!lseg) {
517 /* remember that LAYOUTGET failed and suspend trying */
Fred Isaman566052c2011-01-06 11:36:20 +0000518 set_bit(lo_fail_bit(iomode), &lo->plh_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400519 }
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400520
521 /* free xdr pages */
522 for (i = 0; i < max_pages; i++)
523 __free_page(pages[i]);
524 kfree(pages);
525
Andy Adamson974cec82010-10-20 00:18:02 -0400526 return lseg;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400527
528out_err_free:
529 /* free any allocated xdr pages, lgp as it's not used */
530 if (pages) {
531 for (i = 0; i < max_pages; i++) {
532 if (!pages[i])
533 break;
534 __free_page(pages[i]);
535 }
536 kfree(pages);
537 }
538 kfree(lgp);
539 return NULL;
Andy Adamson974cec82010-10-20 00:18:02 -0400540}
541
Fred Isamanf7e89172011-01-06 11:36:32 +0000542bool pnfs_roc(struct inode *ino)
543{
544 struct pnfs_layout_hdr *lo;
545 struct pnfs_layout_segment *lseg, *tmp;
546 LIST_HEAD(tmp_list);
547 bool found = false;
548
549 spin_lock(&ino->i_lock);
550 lo = NFS_I(ino)->layout;
551 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
552 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
553 goto out_nolayout;
554 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
555 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
556 mark_lseg_invalid(lseg, &tmp_list);
557 found = true;
558 }
559 if (!found)
560 goto out_nolayout;
561 lo->plh_block_lgets++;
562 get_layout_hdr(lo); /* matched in pnfs_roc_release */
563 spin_unlock(&ino->i_lock);
564 pnfs_free_lseg_list(&tmp_list);
565 return true;
566
567out_nolayout:
568 spin_unlock(&ino->i_lock);
569 return false;
570}
571
572void pnfs_roc_release(struct inode *ino)
573{
574 struct pnfs_layout_hdr *lo;
575
576 spin_lock(&ino->i_lock);
577 lo = NFS_I(ino)->layout;
578 lo->plh_block_lgets--;
579 put_layout_hdr_locked(lo);
580 spin_unlock(&ino->i_lock);
581}
582
583void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
584{
585 struct pnfs_layout_hdr *lo;
586
587 spin_lock(&ino->i_lock);
588 lo = NFS_I(ino)->layout;
589 if ((int)(barrier - lo->plh_barrier) > 0)
590 lo->plh_barrier = barrier;
591 spin_unlock(&ino->i_lock);
592}
593
594bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
595{
596 struct nfs_inode *nfsi = NFS_I(ino);
597 struct pnfs_layout_segment *lseg;
598 bool found = false;
599
600 spin_lock(&ino->i_lock);
601 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
602 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
603 found = true;
604 break;
605 }
606 if (!found) {
607 struct pnfs_layout_hdr *lo = nfsi->layout;
608 u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid);
609
610 /* Since close does not return a layout stateid for use as
611 * a barrier, we choose the worst-case barrier.
612 */
613 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
614 }
615 spin_unlock(&ino->i_lock);
616 return found;
617}
618
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400619/*
620 * Compare two layout segments for sorting into layout cache.
621 * We want to preferentially return RW over RO layouts, so ensure those
622 * are seen first.
623 */
624static s64
625cmp_layout(u32 iomode1, u32 iomode2)
626{
627 /* read > read/write */
628 return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ);
629}
630
Andy Adamson974cec82010-10-20 00:18:02 -0400631static void
632pnfs_insert_layout(struct pnfs_layout_hdr *lo,
633 struct pnfs_layout_segment *lseg)
634{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400635 struct pnfs_layout_segment *lp;
636 int found = 0;
637
Andy Adamson974cec82010-10-20 00:18:02 -0400638 dprintk("%s:Begin\n", __func__);
639
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000640 assert_spin_locked(&lo->plh_inode->i_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000641 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000642 if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400643 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000644 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400645 dprintk("%s: inserted lseg %p "
646 "iomode %d offset %llu length %llu before "
647 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000648 __func__, lseg, lseg->pls_range.iomode,
649 lseg->pls_range.offset, lseg->pls_range.length,
650 lp, lp->pls_range.iomode, lp->pls_range.offset,
651 lp->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400652 found = 1;
653 break;
Andy Adamson974cec82010-10-20 00:18:02 -0400654 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400655 if (!found) {
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000656 list_add_tail(&lseg->pls_list, &lo->plh_segs);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400657 dprintk("%s: inserted lseg %p "
658 "iomode %d offset %llu length %llu at tail\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000659 __func__, lseg, lseg->pls_range.iomode,
660 lseg->pls_range.offset, lseg->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400661 }
Fred Isamancc6e5342011-01-06 11:36:28 +0000662 get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400663
664 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400665}
666
667static struct pnfs_layout_hdr *
668alloc_init_layout_hdr(struct inode *ino)
669{
670 struct pnfs_layout_hdr *lo;
671
672 lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL);
673 if (!lo)
674 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +0000675 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000676 INIT_LIST_HEAD(&lo->plh_layouts);
677 INIT_LIST_HEAD(&lo->plh_segs);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000678 INIT_LIST_HEAD(&lo->plh_bulk_recall);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000679 lo->plh_inode = ino;
Benny Halevye5e94012010-10-20 00:18:01 -0400680 return lo;
681}
682
683static struct pnfs_layout_hdr *
684pnfs_find_alloc_layout(struct inode *ino)
685{
686 struct nfs_inode *nfsi = NFS_I(ino);
687 struct pnfs_layout_hdr *new = NULL;
688
689 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
690
691 assert_spin_locked(&ino->i_lock);
Fred Isaman4541d162011-01-06 11:36:23 +0000692 if (nfsi->layout) {
693 if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
694 return NULL;
695 else
696 return nfsi->layout;
697 }
Benny Halevye5e94012010-10-20 00:18:01 -0400698 spin_unlock(&ino->i_lock);
699 new = alloc_init_layout_hdr(ino);
700 spin_lock(&ino->i_lock);
701
702 if (likely(nfsi->layout == NULL)) /* Won the race? */
703 nfsi->layout = new;
704 else
705 kfree(new);
706 return nfsi->layout;
707}
708
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400709/*
710 * iomode matching rules:
711 * iomode lseg match
712 * ----- ----- -----
713 * ANY READ true
714 * ANY RW true
715 * RW READ false
716 * RW RW true
717 * READ READ true
718 * READ RW true
719 */
720static int
721is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode)
722{
Fred Isaman566052c2011-01-06 11:36:20 +0000723 return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400724}
725
726/*
727 * lookup range in layout
728 */
Benny Halevye5e94012010-10-20 00:18:01 -0400729static struct pnfs_layout_segment *
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000730pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode)
Benny Halevye5e94012010-10-20 00:18:01 -0400731{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400732 struct pnfs_layout_segment *lseg, *ret = NULL;
733
734 dprintk("%s:Begin\n", __func__);
735
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000736 assert_spin_locked(&lo->plh_inode->i_lock);
737 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +0000738 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
739 is_matching_lseg(lseg, iomode)) {
Fred Isamand684d2a2011-03-01 01:34:13 +0000740 ret = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400741 break;
742 }
Fred Isaman566052c2011-01-06 11:36:20 +0000743 if (cmp_layout(iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400744 break;
745 }
746
747 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +0000748 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400749 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400750}
751
752/*
753 * Layout segment is retreived from the server if not cached.
754 * The appropriate layout segment is referenced and returned to the caller.
755 */
756struct pnfs_layout_segment *
757pnfs_update_layout(struct inode *ino,
758 struct nfs_open_context *ctx,
759 enum pnfs_iomode iomode)
760{
761 struct nfs_inode *nfsi = NFS_I(ino);
Fred Isaman2130ff62011-01-06 11:36:26 +0000762 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -0400763 struct pnfs_layout_hdr *lo;
764 struct pnfs_layout_segment *lseg = NULL;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000765 bool first = false;
Benny Halevye5e94012010-10-20 00:18:01 -0400766
767 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
768 return NULL;
769 spin_lock(&ino->i_lock);
770 lo = pnfs_find_alloc_layout(ino);
771 if (lo == NULL) {
772 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
773 goto out_unlock;
774 }
775
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000776 /* Do we even need to bother with this? */
777 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
778 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
779 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400780 goto out_unlock;
781 }
782
783 /* if LAYOUTGET already failed once we don't try again */
Fred Isaman566052c2011-01-06 11:36:20 +0000784 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
Benny Halevye5e94012010-10-20 00:18:01 -0400785 goto out_unlock;
786
Andy Adamson568e8c42011-03-01 01:34:22 +0000787 /* Check to see if the layout for the given range already exists */
788 lseg = pnfs_find_lseg(lo, iomode);
789 if (lseg)
790 goto out_unlock;
791
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000792 if (pnfs_layoutgets_blocked(lo, NULL, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +0000793 goto out_unlock;
794 atomic_inc(&lo->plh_outstanding);
795
Fred Isamancc6e5342011-01-06 11:36:28 +0000796 get_layout_hdr(lo);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000797 if (list_empty(&lo->plh_segs))
798 first = true;
799 spin_unlock(&ino->i_lock);
800 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +0000801 /* The lo must be on the clp list if there is any
802 * chance of a CB_LAYOUTRECALL(FILE) coming in.
803 */
804 spin_lock(&clp->cl_lock);
805 BUG_ON(!list_empty(&lo->plh_layouts));
806 list_add_tail(&lo->plh_layouts, &clp->cl_layouts);
807 spin_unlock(&clp->cl_lock);
808 }
Benny Halevye5e94012010-10-20 00:18:01 -0400809
810 lseg = send_layoutget(lo, ctx, iomode);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000811 if (!lseg && first) {
812 spin_lock(&clp->cl_lock);
813 list_del_init(&lo->plh_layouts);
814 spin_unlock(&clp->cl_lock);
Fred Isaman2130ff62011-01-06 11:36:26 +0000815 }
Fred Isamancf7d63f2011-01-06 11:36:25 +0000816 atomic_dec(&lo->plh_outstanding);
Fred Isamancc6e5342011-01-06 11:36:28 +0000817 put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400818out:
819 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
Andy Adamsonbf9c1382011-03-01 01:34:07 +0000820 nfsi->layout ? nfsi->layout->plh_flags : -1, lseg);
Benny Halevye5e94012010-10-20 00:18:01 -0400821 return lseg;
822out_unlock:
823 spin_unlock(&ino->i_lock);
824 goto out;
825}
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400826
827int
828pnfs_layout_process(struct nfs4_layoutget *lgp)
829{
830 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
831 struct nfs4_layoutget_res *res = &lgp->res;
832 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000833 struct inode *ino = lo->plh_inode;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000834 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400835 int status = 0;
836
Fred Isamanfc1794c2011-01-06 11:36:27 +0000837 /* Verify we got what we asked for.
838 * Note that because the xdr parsing only accepts a single
839 * element array, this can fail even if the server is behaving
840 * correctly.
841 */
842 if (lgp->args.range.iomode > res->range.iomode ||
843 res->range.offset != 0 ||
844 res->range.length != NFS4_MAX_UINT64) {
845 status = -EINVAL;
846 goto out;
847 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400848 /* Inject layout blob into I/O device driver */
849 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res);
850 if (!lseg || IS_ERR(lseg)) {
851 if (!lseg)
852 status = -ENOMEM;
853 else
854 status = PTR_ERR(lseg);
855 dprintk("%s: Could not allocate layout: error %d\n",
856 __func__, status);
857 goto out;
858 }
859
860 spin_lock(&ino->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000861 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
862 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
863 dprintk("%s forget reply due to recall\n", __func__);
864 goto out_forget_reply;
865 }
866
867 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
868 dprintk("%s forget reply due to state\n", __func__);
869 goto out_forget_reply;
870 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400871 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +0000872 lseg->pls_range = res->range;
Fred Isamand684d2a2011-03-01 01:34:13 +0000873 *lgp->lsegpp = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400874 pnfs_insert_layout(lo, lseg);
875
Fred Isamanf7e89172011-01-06 11:36:32 +0000876 if (res->return_on_close) {
877 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
878 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
879 }
880
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400881 /* Done processing layoutget. Set the layout stateid */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000882 pnfs_set_layout_stateid(lo, &res->stateid, false);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400883 spin_unlock(&ino->i_lock);
884out:
885 return status;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000886
887out_forget_reply:
888 spin_unlock(&ino->i_lock);
889 lseg->pls_layout = lo;
890 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
891 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400892}
893
Fred Isamanbae724e2011-03-01 01:34:15 +0000894static int pnfs_read_pg_test(struct nfs_pageio_descriptor *pgio,
895 struct nfs_page *prev,
896 struct nfs_page *req)
897{
898 if (pgio->pg_count == prev->wb_bytes) {
899 /* This is first coelesce call for a series of nfs_pages */
900 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
901 prev->wb_context,
902 IOMODE_READ);
903 }
904 return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req);
905}
906
907void
908pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode)
Fred Isaman94ad1c82011-03-01 01:34:14 +0000909{
910 struct pnfs_layoutdriver_type *ld;
911
912 ld = NFS_SERVER(inode)->pnfs_curr_ld;
Fred Isamanbae724e2011-03-01 01:34:15 +0000913 pgio->pg_test = (ld && ld->pg_test) ? pnfs_read_pg_test : NULL;
Fred Isaman94ad1c82011-03-01 01:34:14 +0000914}
915
Fred Isaman44b83792011-03-03 15:13:44 +0000916static int pnfs_write_pg_test(struct nfs_pageio_descriptor *pgio,
917 struct nfs_page *prev,
918 struct nfs_page *req)
919{
920 if (pgio->pg_count == prev->wb_bytes) {
921 /* This is first coelesce call for a series of nfs_pages */
922 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
923 prev->wb_context,
924 IOMODE_RW);
925 }
926 return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req);
927}
928
929void
930pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode)
931{
932 struct pnfs_layoutdriver_type *ld;
933
934 ld = NFS_SERVER(inode)->pnfs_curr_ld;
935 pgio->pg_test = (ld && ld->pg_test) ? pnfs_write_pg_test : NULL;
936}
937
Andy Adamson0382b742011-03-03 15:13:45 +0000938enum pnfs_try_status
939pnfs_try_to_write_data(struct nfs_write_data *wdata,
940 const struct rpc_call_ops *call_ops, int how)
941{
942 struct inode *inode = wdata->inode;
943 enum pnfs_try_status trypnfs;
944 struct nfs_server *nfss = NFS_SERVER(inode);
945
946 wdata->mds_ops = call_ops;
947
948 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
949 inode->i_ino, wdata->args.count, wdata->args.offset, how);
950
951 trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
952 if (trypnfs == PNFS_NOT_ATTEMPTED) {
953 put_lseg(wdata->lseg);
954 wdata->lseg = NULL;
955 } else
956 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
957
958 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
959 return trypnfs;
960}
961
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400962/*
Andy Adamson64419a92011-03-01 01:34:16 +0000963 * Call the appropriate parallel I/O subsystem read function.
964 */
965enum pnfs_try_status
966pnfs_try_to_read_data(struct nfs_read_data *rdata,
967 const struct rpc_call_ops *call_ops)
968{
969 struct inode *inode = rdata->inode;
970 struct nfs_server *nfss = NFS_SERVER(inode);
971 enum pnfs_try_status trypnfs;
972
973 rdata->mds_ops = call_ops;
974
975 dprintk("%s: Reading ino:%lu %u@%llu\n",
976 __func__, inode->i_ino, rdata->args.count, rdata->args.offset);
977
978 trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
979 if (trypnfs == PNFS_NOT_ATTEMPTED) {
980 put_lseg(rdata->lseg);
981 rdata->lseg = NULL;
982 } else {
983 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
984 }
985 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
986 return trypnfs;
987}
Andy Adamson863a3c62011-03-23 13:27:54 +0000988
989/*
990 * Currently there is only one (whole file) write lseg.
991 */
992static struct pnfs_layout_segment *pnfs_list_write_lseg(struct inode *inode)
993{
994 struct pnfs_layout_segment *lseg, *rv = NULL;
995
996 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
997 if (lseg->pls_range.iomode == IOMODE_RW)
998 rv = lseg;
999 return rv;
1000}
1001
1002void
1003pnfs_set_layoutcommit(struct nfs_write_data *wdata)
1004{
1005 struct nfs_inode *nfsi = NFS_I(wdata->inode);
1006 loff_t end_pos = wdata->args.offset + wdata->res.count;
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001007 bool mark_as_dirty = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001008
1009 spin_lock(&nfsi->vfs_inode.i_lock);
1010 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1011 /* references matched in nfs4_layoutcommit_release */
1012 get_lseg(wdata->lseg);
1013 wdata->lseg->pls_lc_cred =
1014 get_rpccred(wdata->args.context->state->owner->so_cred);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001015 mark_as_dirty = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001016 dprintk("%s: Set layoutcommit for inode %lu ",
1017 __func__, wdata->inode->i_ino);
1018 }
1019 if (end_pos > wdata->lseg->pls_end_pos)
1020 wdata->lseg->pls_end_pos = end_pos;
1021 spin_unlock(&nfsi->vfs_inode.i_lock);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001022
1023 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1024 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1025 if (mark_as_dirty)
1026 mark_inode_dirty_sync(wdata->inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001027}
1028EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
1029
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001030/*
1031 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1032 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1033 * data to disk to allow the server to recover the data if it crashes.
1034 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1035 * is off, and a COMMIT is sent to a data server, or
1036 * if WRITEs to a data server return NFS_DATA_SYNC.
1037 */
Andy Adamson863a3c62011-03-23 13:27:54 +00001038int
Andy Adamsonef311532011-03-12 02:58:10 -05001039pnfs_layoutcommit_inode(struct inode *inode, bool sync)
Andy Adamson863a3c62011-03-23 13:27:54 +00001040{
1041 struct nfs4_layoutcommit_data *data;
1042 struct nfs_inode *nfsi = NFS_I(inode);
1043 struct pnfs_layout_segment *lseg;
1044 struct rpc_cred *cred;
1045 loff_t end_pos;
1046 int status = 0;
1047
1048 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
1049
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001050 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1051 return 0;
1052
Andy Adamson863a3c62011-03-23 13:27:54 +00001053 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1054 data = kzalloc(sizeof(*data), GFP_NOFS);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001055 if (!data) {
1056 mark_inode_dirty_sync(inode);
1057 status = -ENOMEM;
1058 goto out;
1059 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001060
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001061 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00001062 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1063 spin_unlock(&inode->i_lock);
1064 kfree(data);
1065 goto out;
1066 }
1067 /*
1068 * Currently only one (whole file) write lseg which is referenced
1069 * in pnfs_set_layoutcommit and will be found.
1070 */
1071 lseg = pnfs_list_write_lseg(inode);
1072
1073 end_pos = lseg->pls_end_pos;
1074 cred = lseg->pls_lc_cred;
1075 lseg->pls_end_pos = 0;
1076 lseg->pls_lc_cred = NULL;
1077
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001078 memcpy(&data->args.stateid.data, nfsi->layout->plh_stateid.data,
1079 sizeof(nfsi->layout->plh_stateid.data));
Andy Adamson863a3c62011-03-23 13:27:54 +00001080 spin_unlock(&inode->i_lock);
1081
1082 data->args.inode = inode;
1083 data->lseg = lseg;
1084 data->cred = cred;
1085 nfs_fattr_init(&data->fattr);
1086 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
1087 data->res.fattr = &data->fattr;
1088 data->args.lastbytewritten = end_pos - 1;
1089 data->res.server = NFS_SERVER(inode);
1090
1091 status = nfs4_proc_layoutcommit(data, sync);
1092out:
1093 dprintk("<-- %s status %d\n", __func__, status);
1094 return status;
1095}