blob: f57f5281a52090ac8445a3ebb362515708225661 [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);
Andy Adamson2887fe42011-05-11 01:19:58 -0400386 list_del_init(&lo->plh_layouts);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000387 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400388 }
389}
390
Fred Isamanfd6002e2011-01-06 11:36:22 +0000391/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000392void
393pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
394 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400395{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000396 u32 oldseq, newseq;
Andy Adamson974cec82010-10-20 00:18:02 -0400397
Fred Isamanfd6002e2011-01-06 11:36:22 +0000398 oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid);
399 newseq = be32_to_cpu(new->stateid.seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000400 if ((int)(newseq - oldseq) > 0) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000401 memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid));
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000402 if (update_barrier) {
403 u32 new_barrier = be32_to_cpu(new->stateid.seqid);
404
405 if ((int)(new_barrier - lo->plh_barrier))
406 lo->plh_barrier = new_barrier;
407 } else {
408 /* Because of wraparound, we want to keep the barrier
409 * "close" to the current seqids. It needs to be
410 * within 2**31 to count as "behind", so if it
411 * gets too near that limit, give us a litle leeway
412 * and bring it to within 2**30.
413 * NOTE - and yes, this is all unsigned arithmetic.
414 */
415 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
416 lo->plh_barrier = newseq - (1 << 30);
417 }
418 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400419}
420
Fred Isamancf7d63f2011-01-06 11:36:25 +0000421/* lget is set to 1 if called from inside send_layoutget call chain */
422static bool
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000423pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
424 int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000425{
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000426 if ((stateid) &&
427 (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0)
428 return true;
Fred Isamanf7e89172011-01-06 11:36:32 +0000429 return lo->plh_block_lgets ||
Fred Isaman38511722011-02-03 18:28:50 +0000430 test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
Fred Isamanf7e89172011-01-06 11:36:32 +0000431 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000432 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000433 (atomic_read(&lo->plh_outstanding) > lget));
434}
435
Fred Isamanfd6002e2011-01-06 11:36:22 +0000436int
437pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
438 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400439{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000440 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400441
442 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000443 spin_lock(&lo->plh_inode->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000444 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000445 status = -EAGAIN;
446 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000447 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400448
Fred Isamanfd6002e2011-01-06 11:36:22 +0000449 do {
450 seq = read_seqbegin(&open_state->seqlock);
451 memcpy(dst->data, open_state->stateid.data,
452 sizeof(open_state->stateid.data));
453 } while (read_seqretry(&open_state->seqlock, seq));
454 } else
455 memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data));
456 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400457 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000458 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400459}
460
461/*
462* Get layout from server.
463* for now, assume that whole file layouts are requested.
464* arg->offset: 0
465* arg->length: all ones
466*/
Benny Halevye5e94012010-10-20 00:18:01 -0400467static struct pnfs_layout_segment *
468send_layoutget(struct pnfs_layout_hdr *lo,
469 struct nfs_open_context *ctx,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400470 u32 iomode,
471 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400472{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000473 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400474 struct nfs_server *server = NFS_SERVER(ino);
475 struct nfs4_layoutget *lgp;
476 struct pnfs_layout_segment *lseg = NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400477 struct page **pages = NULL;
478 int i;
479 u32 max_resp_sz, max_pages;
Benny Halevye5e94012010-10-20 00:18:01 -0400480
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400481 dprintk("--> %s\n", __func__);
482
483 BUG_ON(ctx == NULL);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400484 lgp = kzalloc(sizeof(*lgp), gfp_flags);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000485 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400486 return NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400487
488 /* allocate pages for xdr post processing */
489 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
490 max_pages = max_resp_sz >> PAGE_SHIFT;
491
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400492 pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400493 if (!pages)
494 goto out_err_free;
495
496 for (i = 0; i < max_pages; i++) {
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400497 pages[i] = alloc_page(gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400498 if (!pages[i])
499 goto out_err_free;
500 }
501
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400502 lgp->args.minlength = NFS4_MAX_UINT64;
503 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
504 lgp->args.range.iomode = iomode;
505 lgp->args.range.offset = 0;
506 lgp->args.range.length = NFS4_MAX_UINT64;
507 lgp->args.type = server->pnfs_curr_ld->id;
508 lgp->args.inode = ino;
509 lgp->args.ctx = get_nfs_open_context(ctx);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400510 lgp->args.layout.pages = pages;
511 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400512 lgp->lsegpp = &lseg;
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400513 lgp->gfp_flags = gfp_flags;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400514
515 /* Synchronously retrieve layout information from server and
516 * store in lseg.
517 */
518 nfs4_proc_layoutget(lgp);
519 if (!lseg) {
520 /* remember that LAYOUTGET failed and suspend trying */
Fred Isaman566052c2011-01-06 11:36:20 +0000521 set_bit(lo_fail_bit(iomode), &lo->plh_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400522 }
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400523
524 /* free xdr pages */
525 for (i = 0; i < max_pages; i++)
526 __free_page(pages[i]);
527 kfree(pages);
528
Andy Adamson974cec82010-10-20 00:18:02 -0400529 return lseg;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400530
531out_err_free:
532 /* free any allocated xdr pages, lgp as it's not used */
533 if (pages) {
534 for (i = 0; i < max_pages; i++) {
535 if (!pages[i])
536 break;
537 __free_page(pages[i]);
538 }
539 kfree(pages);
540 }
541 kfree(lgp);
542 return NULL;
Andy Adamson974cec82010-10-20 00:18:02 -0400543}
544
Fred Isamanf7e89172011-01-06 11:36:32 +0000545bool pnfs_roc(struct inode *ino)
546{
547 struct pnfs_layout_hdr *lo;
548 struct pnfs_layout_segment *lseg, *tmp;
549 LIST_HEAD(tmp_list);
550 bool found = false;
551
552 spin_lock(&ino->i_lock);
553 lo = NFS_I(ino)->layout;
554 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
555 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
556 goto out_nolayout;
557 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
558 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
559 mark_lseg_invalid(lseg, &tmp_list);
560 found = true;
561 }
562 if (!found)
563 goto out_nolayout;
564 lo->plh_block_lgets++;
565 get_layout_hdr(lo); /* matched in pnfs_roc_release */
566 spin_unlock(&ino->i_lock);
567 pnfs_free_lseg_list(&tmp_list);
568 return true;
569
570out_nolayout:
571 spin_unlock(&ino->i_lock);
572 return false;
573}
574
575void pnfs_roc_release(struct inode *ino)
576{
577 struct pnfs_layout_hdr *lo;
578
579 spin_lock(&ino->i_lock);
580 lo = NFS_I(ino)->layout;
581 lo->plh_block_lgets--;
582 put_layout_hdr_locked(lo);
583 spin_unlock(&ino->i_lock);
584}
585
586void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
587{
588 struct pnfs_layout_hdr *lo;
589
590 spin_lock(&ino->i_lock);
591 lo = NFS_I(ino)->layout;
592 if ((int)(barrier - lo->plh_barrier) > 0)
593 lo->plh_barrier = barrier;
594 spin_unlock(&ino->i_lock);
595}
596
597bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
598{
599 struct nfs_inode *nfsi = NFS_I(ino);
600 struct pnfs_layout_segment *lseg;
601 bool found = false;
602
603 spin_lock(&ino->i_lock);
604 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
605 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
606 found = true;
607 break;
608 }
609 if (!found) {
610 struct pnfs_layout_hdr *lo = nfsi->layout;
611 u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid);
612
613 /* Since close does not return a layout stateid for use as
614 * a barrier, we choose the worst-case barrier.
615 */
616 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
617 }
618 spin_unlock(&ino->i_lock);
619 return found;
620}
621
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400622/*
623 * Compare two layout segments for sorting into layout cache.
624 * We want to preferentially return RW over RO layouts, so ensure those
625 * are seen first.
626 */
627static s64
628cmp_layout(u32 iomode1, u32 iomode2)
629{
630 /* read > read/write */
631 return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ);
632}
633
Andy Adamson974cec82010-10-20 00:18:02 -0400634static void
635pnfs_insert_layout(struct pnfs_layout_hdr *lo,
636 struct pnfs_layout_segment *lseg)
637{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400638 struct pnfs_layout_segment *lp;
639 int found = 0;
640
Andy Adamson974cec82010-10-20 00:18:02 -0400641 dprintk("%s:Begin\n", __func__);
642
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000643 assert_spin_locked(&lo->plh_inode->i_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000644 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000645 if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400646 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000647 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400648 dprintk("%s: inserted lseg %p "
649 "iomode %d offset %llu length %llu before "
650 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000651 __func__, lseg, lseg->pls_range.iomode,
652 lseg->pls_range.offset, lseg->pls_range.length,
653 lp, lp->pls_range.iomode, lp->pls_range.offset,
654 lp->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400655 found = 1;
656 break;
Andy Adamson974cec82010-10-20 00:18:02 -0400657 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400658 if (!found) {
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000659 list_add_tail(&lseg->pls_list, &lo->plh_segs);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400660 dprintk("%s: inserted lseg %p "
661 "iomode %d offset %llu length %llu at tail\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000662 __func__, lseg, lseg->pls_range.iomode,
663 lseg->pls_range.offset, lseg->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400664 }
Fred Isamancc6e5342011-01-06 11:36:28 +0000665 get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400666
667 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400668}
669
670static struct pnfs_layout_hdr *
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400671alloc_init_layout_hdr(struct inode *ino, gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400672{
673 struct pnfs_layout_hdr *lo;
674
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400675 lo = kzalloc(sizeof(struct pnfs_layout_hdr), gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400676 if (!lo)
677 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +0000678 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000679 INIT_LIST_HEAD(&lo->plh_layouts);
680 INIT_LIST_HEAD(&lo->plh_segs);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000681 INIT_LIST_HEAD(&lo->plh_bulk_recall);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000682 lo->plh_inode = ino;
Benny Halevye5e94012010-10-20 00:18:01 -0400683 return lo;
684}
685
686static struct pnfs_layout_hdr *
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400687pnfs_find_alloc_layout(struct inode *ino, gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400688{
689 struct nfs_inode *nfsi = NFS_I(ino);
690 struct pnfs_layout_hdr *new = NULL;
691
692 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
693
694 assert_spin_locked(&ino->i_lock);
Fred Isaman4541d162011-01-06 11:36:23 +0000695 if (nfsi->layout) {
696 if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
697 return NULL;
698 else
699 return nfsi->layout;
700 }
Benny Halevye5e94012010-10-20 00:18:01 -0400701 spin_unlock(&ino->i_lock);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400702 new = alloc_init_layout_hdr(ino, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400703 spin_lock(&ino->i_lock);
704
705 if (likely(nfsi->layout == NULL)) /* Won the race? */
706 nfsi->layout = new;
707 else
708 kfree(new);
709 return nfsi->layout;
710}
711
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400712/*
713 * iomode matching rules:
714 * iomode lseg match
715 * ----- ----- -----
716 * ANY READ true
717 * ANY RW true
718 * RW READ false
719 * RW RW true
720 * READ READ true
721 * READ RW true
722 */
723static int
724is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode)
725{
Fred Isaman566052c2011-01-06 11:36:20 +0000726 return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400727}
728
729/*
730 * lookup range in layout
731 */
Benny Halevye5e94012010-10-20 00:18:01 -0400732static struct pnfs_layout_segment *
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000733pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode)
Benny Halevye5e94012010-10-20 00:18:01 -0400734{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400735 struct pnfs_layout_segment *lseg, *ret = NULL;
736
737 dprintk("%s:Begin\n", __func__);
738
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000739 assert_spin_locked(&lo->plh_inode->i_lock);
740 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +0000741 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
742 is_matching_lseg(lseg, iomode)) {
Fred Isamand684d2a2011-03-01 01:34:13 +0000743 ret = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400744 break;
745 }
Fred Isaman566052c2011-01-06 11:36:20 +0000746 if (cmp_layout(iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400747 break;
748 }
749
750 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +0000751 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400752 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400753}
754
755/*
756 * Layout segment is retreived from the server if not cached.
757 * The appropriate layout segment is referenced and returned to the caller.
758 */
759struct pnfs_layout_segment *
760pnfs_update_layout(struct inode *ino,
761 struct nfs_open_context *ctx,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400762 enum pnfs_iomode iomode,
763 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400764{
765 struct nfs_inode *nfsi = NFS_I(ino);
Fred Isaman2130ff62011-01-06 11:36:26 +0000766 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -0400767 struct pnfs_layout_hdr *lo;
768 struct pnfs_layout_segment *lseg = NULL;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000769 bool first = false;
Benny Halevye5e94012010-10-20 00:18:01 -0400770
771 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
772 return NULL;
773 spin_lock(&ino->i_lock);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400774 lo = pnfs_find_alloc_layout(ino, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400775 if (lo == NULL) {
776 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
777 goto out_unlock;
778 }
779
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000780 /* Do we even need to bother with this? */
781 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
782 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
783 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400784 goto out_unlock;
785 }
786
787 /* if LAYOUTGET already failed once we don't try again */
Fred Isaman566052c2011-01-06 11:36:20 +0000788 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
Benny Halevye5e94012010-10-20 00:18:01 -0400789 goto out_unlock;
790
Andy Adamson568e8c42011-03-01 01:34:22 +0000791 /* Check to see if the layout for the given range already exists */
792 lseg = pnfs_find_lseg(lo, iomode);
793 if (lseg)
794 goto out_unlock;
795
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000796 if (pnfs_layoutgets_blocked(lo, NULL, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +0000797 goto out_unlock;
798 atomic_inc(&lo->plh_outstanding);
799
Fred Isamancc6e5342011-01-06 11:36:28 +0000800 get_layout_hdr(lo);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000801 if (list_empty(&lo->plh_segs))
802 first = true;
803 spin_unlock(&ino->i_lock);
804 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +0000805 /* The lo must be on the clp list if there is any
806 * chance of a CB_LAYOUTRECALL(FILE) coming in.
807 */
808 spin_lock(&clp->cl_lock);
809 BUG_ON(!list_empty(&lo->plh_layouts));
810 list_add_tail(&lo->plh_layouts, &clp->cl_layouts);
811 spin_unlock(&clp->cl_lock);
812 }
Benny Halevye5e94012010-10-20 00:18:01 -0400813
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400814 lseg = send_layoutget(lo, ctx, iomode, gfp_flags);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000815 if (!lseg && first) {
816 spin_lock(&clp->cl_lock);
817 list_del_init(&lo->plh_layouts);
818 spin_unlock(&clp->cl_lock);
Fred Isaman2130ff62011-01-06 11:36:26 +0000819 }
Fred Isamancf7d63f2011-01-06 11:36:25 +0000820 atomic_dec(&lo->plh_outstanding);
Fred Isamancc6e5342011-01-06 11:36:28 +0000821 put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400822out:
823 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
Andy Adamsonbf9c1382011-03-01 01:34:07 +0000824 nfsi->layout ? nfsi->layout->plh_flags : -1, lseg);
Benny Halevye5e94012010-10-20 00:18:01 -0400825 return lseg;
826out_unlock:
827 spin_unlock(&ino->i_lock);
828 goto out;
829}
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400830
831int
832pnfs_layout_process(struct nfs4_layoutget *lgp)
833{
834 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
835 struct nfs4_layoutget_res *res = &lgp->res;
836 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000837 struct inode *ino = lo->plh_inode;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000838 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400839 int status = 0;
840
Fred Isamanfc1794c2011-01-06 11:36:27 +0000841 /* Verify we got what we asked for.
842 * Note that because the xdr parsing only accepts a single
843 * element array, this can fail even if the server is behaving
844 * correctly.
845 */
846 if (lgp->args.range.iomode > res->range.iomode ||
847 res->range.offset != 0 ||
848 res->range.length != NFS4_MAX_UINT64) {
849 status = -EINVAL;
850 goto out;
851 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400852 /* Inject layout blob into I/O device driver */
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400853 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400854 if (!lseg || IS_ERR(lseg)) {
855 if (!lseg)
856 status = -ENOMEM;
857 else
858 status = PTR_ERR(lseg);
859 dprintk("%s: Could not allocate layout: error %d\n",
860 __func__, status);
861 goto out;
862 }
863
864 spin_lock(&ino->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000865 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
866 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
867 dprintk("%s forget reply due to recall\n", __func__);
868 goto out_forget_reply;
869 }
870
871 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
872 dprintk("%s forget reply due to state\n", __func__);
873 goto out_forget_reply;
874 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400875 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +0000876 lseg->pls_range = res->range;
Fred Isamand684d2a2011-03-01 01:34:13 +0000877 *lgp->lsegpp = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400878 pnfs_insert_layout(lo, lseg);
879
Fred Isamanf7e89172011-01-06 11:36:32 +0000880 if (res->return_on_close) {
881 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
882 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
883 }
884
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400885 /* Done processing layoutget. Set the layout stateid */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000886 pnfs_set_layout_stateid(lo, &res->stateid, false);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400887 spin_unlock(&ino->i_lock);
888out:
889 return status;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000890
891out_forget_reply:
892 spin_unlock(&ino->i_lock);
893 lseg->pls_layout = lo;
894 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
895 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400896}
897
Fred Isamanbae724e2011-03-01 01:34:15 +0000898static int pnfs_read_pg_test(struct nfs_pageio_descriptor *pgio,
899 struct nfs_page *prev,
900 struct nfs_page *req)
901{
902 if (pgio->pg_count == prev->wb_bytes) {
903 /* This is first coelesce call for a series of nfs_pages */
904 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
905 prev->wb_context,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400906 IOMODE_READ,
907 GFP_KERNEL);
Fred Isamanbae724e2011-03-01 01:34:15 +0000908 }
909 return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req);
910}
911
912void
913pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode)
Fred Isaman94ad1c82011-03-01 01:34:14 +0000914{
915 struct pnfs_layoutdriver_type *ld;
916
917 ld = NFS_SERVER(inode)->pnfs_curr_ld;
Fred Isamanbae724e2011-03-01 01:34:15 +0000918 pgio->pg_test = (ld && ld->pg_test) ? pnfs_read_pg_test : NULL;
Fred Isaman94ad1c82011-03-01 01:34:14 +0000919}
920
Fred Isaman44b83792011-03-03 15:13:44 +0000921static int pnfs_write_pg_test(struct nfs_pageio_descriptor *pgio,
922 struct nfs_page *prev,
923 struct nfs_page *req)
924{
925 if (pgio->pg_count == prev->wb_bytes) {
926 /* This is first coelesce call for a series of nfs_pages */
927 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
928 prev->wb_context,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400929 IOMODE_RW,
930 GFP_NOFS);
Fred Isaman44b83792011-03-03 15:13:44 +0000931 }
932 return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req);
933}
934
935void
936pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode)
937{
938 struct pnfs_layoutdriver_type *ld;
939
940 ld = NFS_SERVER(inode)->pnfs_curr_ld;
941 pgio->pg_test = (ld && ld->pg_test) ? pnfs_write_pg_test : NULL;
942}
943
Andy Adamson0382b742011-03-03 15:13:45 +0000944enum pnfs_try_status
945pnfs_try_to_write_data(struct nfs_write_data *wdata,
946 const struct rpc_call_ops *call_ops, int how)
947{
948 struct inode *inode = wdata->inode;
949 enum pnfs_try_status trypnfs;
950 struct nfs_server *nfss = NFS_SERVER(inode);
951
952 wdata->mds_ops = call_ops;
953
954 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
955 inode->i_ino, wdata->args.count, wdata->args.offset, how);
956
957 trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
958 if (trypnfs == PNFS_NOT_ATTEMPTED) {
959 put_lseg(wdata->lseg);
960 wdata->lseg = NULL;
961 } else
962 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
963
964 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
965 return trypnfs;
966}
967
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400968/*
Andy Adamson64419a92011-03-01 01:34:16 +0000969 * Call the appropriate parallel I/O subsystem read function.
970 */
971enum pnfs_try_status
972pnfs_try_to_read_data(struct nfs_read_data *rdata,
973 const struct rpc_call_ops *call_ops)
974{
975 struct inode *inode = rdata->inode;
976 struct nfs_server *nfss = NFS_SERVER(inode);
977 enum pnfs_try_status trypnfs;
978
979 rdata->mds_ops = call_ops;
980
981 dprintk("%s: Reading ino:%lu %u@%llu\n",
982 __func__, inode->i_ino, rdata->args.count, rdata->args.offset);
983
984 trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
985 if (trypnfs == PNFS_NOT_ATTEMPTED) {
986 put_lseg(rdata->lseg);
987 rdata->lseg = NULL;
988 } else {
989 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
990 }
991 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
992 return trypnfs;
993}
Andy Adamson863a3c62011-03-23 13:27:54 +0000994
995/*
996 * Currently there is only one (whole file) write lseg.
997 */
998static struct pnfs_layout_segment *pnfs_list_write_lseg(struct inode *inode)
999{
1000 struct pnfs_layout_segment *lseg, *rv = NULL;
1001
1002 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
1003 if (lseg->pls_range.iomode == IOMODE_RW)
1004 rv = lseg;
1005 return rv;
1006}
1007
1008void
1009pnfs_set_layoutcommit(struct nfs_write_data *wdata)
1010{
1011 struct nfs_inode *nfsi = NFS_I(wdata->inode);
1012 loff_t end_pos = wdata->args.offset + wdata->res.count;
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001013 bool mark_as_dirty = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001014
1015 spin_lock(&nfsi->vfs_inode.i_lock);
1016 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1017 /* references matched in nfs4_layoutcommit_release */
1018 get_lseg(wdata->lseg);
1019 wdata->lseg->pls_lc_cred =
1020 get_rpccred(wdata->args.context->state->owner->so_cred);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001021 mark_as_dirty = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001022 dprintk("%s: Set layoutcommit for inode %lu ",
1023 __func__, wdata->inode->i_ino);
1024 }
1025 if (end_pos > wdata->lseg->pls_end_pos)
1026 wdata->lseg->pls_end_pos = end_pos;
1027 spin_unlock(&nfsi->vfs_inode.i_lock);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001028
1029 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1030 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1031 if (mark_as_dirty)
1032 mark_inode_dirty_sync(wdata->inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001033}
1034EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
1035
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001036/*
1037 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1038 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1039 * data to disk to allow the server to recover the data if it crashes.
1040 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1041 * is off, and a COMMIT is sent to a data server, or
1042 * if WRITEs to a data server return NFS_DATA_SYNC.
1043 */
Andy Adamson863a3c62011-03-23 13:27:54 +00001044int
Andy Adamsonef311532011-03-12 02:58:10 -05001045pnfs_layoutcommit_inode(struct inode *inode, bool sync)
Andy Adamson863a3c62011-03-23 13:27:54 +00001046{
1047 struct nfs4_layoutcommit_data *data;
1048 struct nfs_inode *nfsi = NFS_I(inode);
1049 struct pnfs_layout_segment *lseg;
1050 struct rpc_cred *cred;
1051 loff_t end_pos;
1052 int status = 0;
1053
1054 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
1055
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001056 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1057 return 0;
1058
Andy Adamson863a3c62011-03-23 13:27:54 +00001059 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1060 data = kzalloc(sizeof(*data), GFP_NOFS);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001061 if (!data) {
1062 mark_inode_dirty_sync(inode);
1063 status = -ENOMEM;
1064 goto out;
1065 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001066
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001067 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00001068 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1069 spin_unlock(&inode->i_lock);
1070 kfree(data);
1071 goto out;
1072 }
1073 /*
1074 * Currently only one (whole file) write lseg which is referenced
1075 * in pnfs_set_layoutcommit and will be found.
1076 */
1077 lseg = pnfs_list_write_lseg(inode);
1078
1079 end_pos = lseg->pls_end_pos;
1080 cred = lseg->pls_lc_cred;
1081 lseg->pls_end_pos = 0;
1082 lseg->pls_lc_cred = NULL;
1083
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001084 memcpy(&data->args.stateid.data, nfsi->layout->plh_stateid.data,
1085 sizeof(nfsi->layout->plh_stateid.data));
Andy Adamson863a3c62011-03-23 13:27:54 +00001086 spin_unlock(&inode->i_lock);
1087
1088 data->args.inode = inode;
1089 data->lseg = lseg;
1090 data->cred = cred;
1091 nfs_fattr_init(&data->fattr);
1092 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
1093 data->res.fattr = &data->fattr;
1094 data->args.lastbytewritten = end_pos - 1;
1095 data->res.server = NFS_SERVER(inode);
1096
1097 status = nfs4_proc_layoutcommit(data, sync);
1098out:
1099 dprintk("<-- %s status %d\n", __func__, status);
1100 return status;
1101}