blob: 4755858e37a0b93f64643811f683b0de7f0d6def [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>
Trond Myklebust493292d2011-07-13 15:58:28 -040031#include <linux/nfs_page.h>
Paul Gortmaker143cb492011-07-01 14:23:34 -040032#include <linux/module.h>
Andy Adamson974cec82010-10-20 00:18:02 -040033#include "internal.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040034#include "pnfs.h"
Andy Adamson64419a92011-03-01 01:34:16 +000035#include "iostat.h"
Trond Myklebustcc668ab2013-08-14 15:31:28 -040036#include "nfs4trace.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040037
38#define NFSDBG_FACILITY NFSDBG_PNFS
Trond Myklebust25c75332012-09-18 17:01:12 -040039#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040040
Fred Isaman02c35fc2010-10-20 00:17:59 -040041/* Locking:
42 *
43 * pnfs_spinlock:
44 * protects pnfs_modules_tbl.
45 */
46static DEFINE_SPINLOCK(pnfs_spinlock);
47
48/*
49 * pnfs_modules_tbl holds all pnfs modules
50 */
51static LIST_HEAD(pnfs_modules_tbl);
52
53/* Return the registered pnfs layout driver module matching given id */
54static struct pnfs_layoutdriver_type *
55find_pnfs_driver_locked(u32 id)
56{
57 struct pnfs_layoutdriver_type *local;
58
59 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
60 if (local->id == id)
61 goto out;
62 local = NULL;
63out:
64 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
65 return local;
66}
67
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040068static struct pnfs_layoutdriver_type *
69find_pnfs_driver(u32 id)
70{
Fred Isaman02c35fc2010-10-20 00:17:59 -040071 struct pnfs_layoutdriver_type *local;
72
73 spin_lock(&pnfs_spinlock);
74 local = find_pnfs_driver_locked(id);
Trond Myklebust0a9c63f2012-06-15 13:02:58 -040075 if (local != NULL && !try_module_get(local->owner)) {
76 dprintk("%s: Could not grab reference on module\n", __func__);
77 local = NULL;
78 }
Fred Isaman02c35fc2010-10-20 00:17:59 -040079 spin_unlock(&pnfs_spinlock);
80 return local;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040081}
82
83void
84unset_pnfs_layoutdriver(struct nfs_server *nfss)
85{
Benny Halevy738fd0f32011-07-30 20:52:36 -040086 if (nfss->pnfs_curr_ld) {
87 if (nfss->pnfs_curr_ld->clear_layoutdriver)
88 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
Trond Myklebust2a4c8992012-06-14 13:08:38 -040089 /* Decrement the MDS count. Purge the deviceid cache if zero */
90 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
91 nfs4_deviceid_purge_client(nfss->nfs_client);
Fred Isaman02c35fc2010-10-20 00:17:59 -040092 module_put(nfss->pnfs_curr_ld->owner);
Benny Halevy738fd0f32011-07-30 20:52:36 -040093 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040094 nfss->pnfs_curr_ld = NULL;
95}
96
97/*
98 * Try to set the server's pnfs module to the pnfs layout type specified by id.
99 * Currently only one pNFS layout driver per filesystem is supported.
100 *
101 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
102 */
103void
Benny Halevy738fd0f32011-07-30 20:52:36 -0400104set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
105 u32 id)
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400106{
107 struct pnfs_layoutdriver_type *ld_type = NULL;
108
109 if (id == 0)
110 goto out_no_driver;
111 if (!(server->nfs_client->cl_exchange_flags &
112 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500113 printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n",
114 __func__, id, server->nfs_client->cl_exchange_flags);
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400115 goto out_no_driver;
116 }
117 ld_type = find_pnfs_driver(id);
118 if (!ld_type) {
119 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
120 ld_type = find_pnfs_driver(id);
121 if (!ld_type) {
122 dprintk("%s: No pNFS module found for %u.\n",
123 __func__, id);
124 goto out_no_driver;
125 }
126 }
127 server->pnfs_curr_ld = ld_type;
Benny Halevy738fd0f32011-07-30 20:52:36 -0400128 if (ld_type->set_layoutdriver
129 && ld_type->set_layoutdriver(server, mntfh)) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500130 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
131 "driver %u.\n", __func__, id);
Benny Halevy738fd0f32011-07-30 20:52:36 -0400132 module_put(ld_type->owner);
133 goto out_no_driver;
134 }
Trond Myklebust2a4c8992012-06-14 13:08:38 -0400135 /* Bump the MDS count */
136 atomic_inc(&server->nfs_client->cl_mds_count);
Christoph Hellwigea8eecd2011-03-01 01:34:21 +0000137
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400138 dprintk("%s: pNFS module for %u set\n", __func__, id);
139 return;
140
141out_no_driver:
142 dprintk("%s: Using NFSv4 I/O\n", __func__);
143 server->pnfs_curr_ld = NULL;
144}
Fred Isaman02c35fc2010-10-20 00:17:59 -0400145
146int
147pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
148{
149 int status = -EINVAL;
150 struct pnfs_layoutdriver_type *tmp;
151
152 if (ld_type->id == 0) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500153 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
Fred Isaman02c35fc2010-10-20 00:17:59 -0400154 return status;
155 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400156 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500157 printk(KERN_ERR "NFS: %s Layout driver must provide "
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400158 "alloc_lseg and free_lseg.\n", __func__);
159 return status;
160 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400161
162 spin_lock(&pnfs_spinlock);
163 tmp = find_pnfs_driver_locked(ld_type->id);
164 if (!tmp) {
165 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
166 status = 0;
167 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
168 ld_type->name);
169 } else {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500170 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
Fred Isaman02c35fc2010-10-20 00:17:59 -0400171 __func__, ld_type->id);
172 }
173 spin_unlock(&pnfs_spinlock);
174
175 return status;
176}
177EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
178
179void
180pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
181{
182 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
183 spin_lock(&pnfs_spinlock);
184 list_del(&ld_type->pnfs_tblid);
185 spin_unlock(&pnfs_spinlock);
186}
187EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
Benny Halevye5e94012010-10-20 00:18:01 -0400188
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400189/*
190 * pNFS client layout cache
191 */
192
Fred Isamancc6e5342011-01-06 11:36:28 +0000193/* Need to hold i_lock if caller does not already hold reference */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000194void
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400195pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
Benny Halevye5e94012010-10-20 00:18:01 -0400196{
Fred Isamancc6e5342011-01-06 11:36:28 +0000197 atomic_inc(&lo->plh_refcount);
198}
199
Benny Halevy636fb9c2011-05-22 19:51:33 +0300200static struct pnfs_layout_hdr *
201pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
202{
203 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
Trond Myklebust57934272012-09-20 20:37:23 -0400204 return ld->alloc_layout_hdr(ino, gfp_flags);
Benny Halevy636fb9c2011-05-22 19:51:33 +0300205}
206
207static void
208pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
209{
Trond Myklebust9c626382012-09-20 17:31:43 -0400210 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
211 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
212
213 if (!list_empty(&lo->plh_layouts)) {
214 struct nfs_client *clp = server->nfs_client;
215
216 spin_lock(&clp->cl_lock);
217 list_del_init(&lo->plh_layouts);
218 spin_unlock(&clp->cl_lock);
219 }
Peng Tao9fa40752011-07-30 20:52:32 -0400220 put_rpccred(lo->plh_lc_cred);
Trond Myklebust57934272012-09-20 20:37:23 -0400221 return ld->free_layout_hdr(lo);
Benny Halevy636fb9c2011-05-22 19:51:33 +0300222}
223
Fred Isamancc6e5342011-01-06 11:36:28 +0000224static void
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400225pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
Fred Isamancc6e5342011-01-06 11:36:28 +0000226{
Trond Myklebustbb346f62012-09-20 15:52:13 -0400227 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
Fred Isamancc6e5342011-01-06 11:36:28 +0000228 dprintk("%s: freeing layout cache %p\n", __func__, lo);
Trond Myklebustbb346f62012-09-20 15:52:13 -0400229 nfsi->layout = NULL;
230 /* Reset MDS Threshold I/O counters */
231 nfsi->write_io = 0;
232 nfsi->read_io = 0;
Benny Halevye5e94012010-10-20 00:18:01 -0400233}
234
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400235void
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400236pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
Andy Adamson974cec82010-10-20 00:18:02 -0400237{
Fred Isamancc6e5342011-01-06 11:36:28 +0000238 struct inode *inode = lo->plh_inode;
239
240 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400241 pnfs_detach_layout_hdr(lo);
Fred Isamancc6e5342011-01-06 11:36:28 +0000242 spin_unlock(&inode->i_lock);
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400243 pnfs_free_layout_hdr(lo);
Fred Isamancc6e5342011-01-06 11:36:28 +0000244 }
Andy Adamson974cec82010-10-20 00:18:02 -0400245}
246
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400247static int
248pnfs_iomode_to_fail_bit(u32 iomode)
249{
250 return iomode == IOMODE_RW ?
251 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
252}
253
254static void
Trond Myklebust3e621212012-09-24 13:07:16 -0400255pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400256{
Trond Myklebust25c75332012-09-18 17:01:12 -0400257 lo->plh_retry_timestamp = jiffies;
Yanchuan Nian39e88fc2013-01-04 20:19:49 +0800258 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
Trond Myklebust3e621212012-09-24 13:07:16 -0400259 atomic_inc(&lo->plh_refcount);
260}
261
262static void
263pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
264{
265 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
266 atomic_dec(&lo->plh_refcount);
267}
268
269static void
270pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
271{
272 struct inode *inode = lo->plh_inode;
Trond Myklebust115ce572012-09-20 21:19:43 -0400273 struct pnfs_layout_range range = {
274 .iomode = iomode,
275 .offset = 0,
276 .length = NFS4_MAX_UINT64,
277 };
278 LIST_HEAD(head);
Trond Myklebust3e621212012-09-24 13:07:16 -0400279
280 spin_lock(&inode->i_lock);
281 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
Trond Myklebust115ce572012-09-20 21:19:43 -0400282 pnfs_mark_matching_lsegs_invalid(lo, &head, &range);
Trond Myklebust3e621212012-09-24 13:07:16 -0400283 spin_unlock(&inode->i_lock);
Trond Myklebust115ce572012-09-20 21:19:43 -0400284 pnfs_free_lseg_list(&head);
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400285 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
286 iomode == IOMODE_RW ? "RW" : "READ");
287}
288
289static bool
290pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
291{
Trond Myklebust25c75332012-09-18 17:01:12 -0400292 unsigned long start, end;
Trond Myklebust3e621212012-09-24 13:07:16 -0400293 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
294
295 if (test_bit(fail_bit, &lo->plh_flags) == 0)
Trond Myklebust25c75332012-09-18 17:01:12 -0400296 return false;
297 end = jiffies;
298 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
299 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
300 /* It is time to retry the failed layoutgets */
Trond Myklebust3e621212012-09-24 13:07:16 -0400301 pnfs_layout_clear_fail_bit(lo, fail_bit);
Trond Myklebust25c75332012-09-18 17:01:12 -0400302 return false;
303 }
304 return true;
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400305}
306
Andy Adamson974cec82010-10-20 00:18:02 -0400307static void
308init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
309{
Fred Isaman566052c2011-01-06 11:36:20 +0000310 INIT_LIST_HEAD(&lseg->pls_list);
Peng Taoa9bae562011-07-30 20:52:33 -0400311 INIT_LIST_HEAD(&lseg->pls_lc_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000312 atomic_set(&lseg->pls_refcount, 1);
313 smp_mb();
314 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
Fred Isaman566052c2011-01-06 11:36:20 +0000315 lseg->pls_layout = lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400316}
317
Trond Myklebust905ca192012-09-20 20:46:49 -0400318static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400319{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000320 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400321
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400322 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400323}
324
Fred Isamand684d2a2011-03-01 01:34:13 +0000325static void
Trond Myklebust57036a32012-09-20 16:33:30 -0400326pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
327 struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400328{
Trond Myklebust57036a32012-09-20 16:33:30 -0400329 struct inode *inode = lo->plh_inode;
Fred Isamand684d2a2011-03-01 01:34:13 +0000330
Benny Halevyd20581a2011-05-22 19:52:03 +0300331 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000332 list_del_init(&lseg->pls_list);
Trond Myklebust8f0d27d2012-09-20 20:57:11 -0400333 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
334 atomic_dec(&lo->plh_refcount);
Trond Myklebust173f77e2012-09-21 15:49:42 -0400335 if (list_empty(&lo->plh_segs))
336 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
Fred Isamand684d2a2011-03-01 01:34:13 +0000337 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
338}
339
Fred Isamanbae724e2011-03-01 01:34:15 +0000340void
Trond Myklebust9369a432012-09-18 20:57:08 -0400341pnfs_put_lseg(struct pnfs_layout_segment *lseg)
Fred Isamand684d2a2011-03-01 01:34:13 +0000342{
Trond Myklebust57036a32012-09-20 16:33:30 -0400343 struct pnfs_layout_hdr *lo;
Fred Isamand684d2a2011-03-01 01:34:13 +0000344 struct inode *inode;
345
346 if (!lseg)
347 return;
348
Fred Isaman4541d162011-01-06 11:36:23 +0000349 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
350 atomic_read(&lseg->pls_refcount),
351 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Trond Myklebust57036a32012-09-20 16:33:30 -0400352 lo = lseg->pls_layout;
353 inode = lo->plh_inode;
Fred Isamand684d2a2011-03-01 01:34:13 +0000354 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
Trond Myklebust8f0d27d2012-09-20 20:57:11 -0400355 pnfs_get_layout_hdr(lo);
Trond Myklebust57036a32012-09-20 16:33:30 -0400356 pnfs_layout_remove_lseg(lo, lseg);
Fred Isamand684d2a2011-03-01 01:34:13 +0000357 spin_unlock(&inode->i_lock);
Trond Myklebust905ca192012-09-20 20:46:49 -0400358 pnfs_free_lseg(lseg);
Trond Myklebust8f0d27d2012-09-20 20:57:11 -0400359 pnfs_put_layout_hdr(lo);
Fred Isaman4541d162011-01-06 11:36:23 +0000360 }
Andy Adamson974cec82010-10-20 00:18:02 -0400361}
Trond Myklebust9369a432012-09-18 20:57:08 -0400362EXPORT_SYMBOL_GPL(pnfs_put_lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400363
Trond Myklebust3cb2df12013-06-03 11:24:36 -0400364static u64
Benny Halevyfb3296e2011-05-22 19:47:26 +0300365end_offset(u64 start, u64 len)
Fred Isaman4541d162011-01-06 11:36:23 +0000366{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300367 u64 end;
368
369 end = start + len;
370 return end >= start ? end : NFS4_MAX_UINT64;
371}
372
Benny Halevyfb3296e2011-05-22 19:47:26 +0300373/*
374 * is l2 fully contained in l1?
375 * start1 end1
376 * [----------------------------------)
377 * start2 end2
378 * [----------------)
379 */
Trond Myklebust3cb2df12013-06-03 11:24:36 -0400380static bool
Trond Myklebust7dc0ac72013-06-03 11:30:24 -0400381pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
Trond Myklebust3cb2df12013-06-03 11:24:36 -0400382 const struct pnfs_layout_range *l2)
Benny Halevyfb3296e2011-05-22 19:47:26 +0300383{
384 u64 start1 = l1->offset;
385 u64 end1 = end_offset(start1, l1->length);
386 u64 start2 = l2->offset;
387 u64 end2 = end_offset(start2, l2->length);
388
389 return (start1 <= start2) && (end1 >= end2);
390}
391
392/*
393 * is l1 and l2 intersecting?
394 * start1 end1
395 * [----------------------------------)
396 * start2 end2
397 * [----------------)
398 */
Trond Myklebust3cb2df12013-06-03 11:24:36 -0400399static bool
Trond Myklebust7dc0ac72013-06-03 11:30:24 -0400400pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
Trond Myklebust3cb2df12013-06-03 11:24:36 -0400401 const struct pnfs_layout_range *l2)
Benny Halevyfb3296e2011-05-22 19:47:26 +0300402{
403 u64 start1 = l1->offset;
404 u64 end1 = end_offset(start1, l1->length);
405 u64 start2 = l2->offset;
406 u64 end2 = end_offset(start2, l2->length);
407
408 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
409 (end2 == NFS4_MAX_UINT64 || end2 > start1);
410}
411
Fred Isaman4541d162011-01-06 11:36:23 +0000412static bool
Trond Myklebust3cb2df12013-06-03 11:24:36 -0400413should_free_lseg(const struct pnfs_layout_range *lseg_range,
414 const struct pnfs_layout_range *recall_range)
Fred Isaman4541d162011-01-06 11:36:23 +0000415{
Benny Halevy778b5502011-05-22 19:48:02 +0300416 return (recall_range->iomode == IOMODE_ANY ||
417 lseg_range->iomode == recall_range->iomode) &&
Trond Myklebust7dc0ac72013-06-03 11:30:24 -0400418 pnfs_lseg_range_intersecting(lseg_range, recall_range);
Fred Isaman4541d162011-01-06 11:36:23 +0000419}
420
Trond Myklebust24956802013-03-20 13:03:00 -0400421static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
422 struct list_head *tmp_list)
423{
424 if (!atomic_dec_and_test(&lseg->pls_refcount))
425 return false;
426 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
427 list_add(&lseg->pls_list, tmp_list);
428 return true;
429}
430
Fred Isaman4541d162011-01-06 11:36:23 +0000431/* Returns 1 if lseg is removed from list, 0 otherwise */
432static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
433 struct list_head *tmp_list)
434{
435 int rv = 0;
436
437 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
438 /* Remove the reference keeping the lseg in the
439 * list. It will now be removed when all
440 * outstanding io is finished.
441 */
Fred Isamand684d2a2011-03-01 01:34:13 +0000442 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
443 atomic_read(&lseg->pls_refcount));
Trond Myklebust24956802013-03-20 13:03:00 -0400444 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
Fred Isamand684d2a2011-03-01 01:34:13 +0000445 rv = 1;
Fred Isaman4541d162011-01-06 11:36:23 +0000446 }
447 return rv;
448}
449
450/* Returns count of number of matching invalid lsegs remaining in list
451 * after call.
452 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000453int
Trond Myklebust49a85062012-09-18 20:43:31 -0400454pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
Fred Isaman4541d162011-01-06 11:36:23 +0000455 struct list_head *tmp_list,
Benny Halevy778b5502011-05-22 19:48:02 +0300456 struct pnfs_layout_range *recall_range)
Andy Adamson974cec82010-10-20 00:18:02 -0400457{
458 struct pnfs_layout_segment *lseg, *next;
Fred Isaman4541d162011-01-06 11:36:23 +0000459 int invalid = 0, removed = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400460
461 dprintk("%s:Begin lo %p\n", __func__, lo);
462
Trond Myklebust8006bfb2012-09-21 14:48:04 -0400463 if (list_empty(&lo->plh_segs))
Fred Isaman38511722011-02-03 18:28:50 +0000464 return 0;
Fred Isaman4541d162011-01-06 11:36:23 +0000465 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
Benny Halevy778b5502011-05-22 19:48:02 +0300466 if (!recall_range ||
467 should_free_lseg(&lseg->pls_range, recall_range)) {
Fred Isaman4541d162011-01-06 11:36:23 +0000468 dprintk("%s: freeing lseg %p iomode %d "
469 "offset %llu length %llu\n", __func__,
470 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
471 lseg->pls_range.length);
472 invalid++;
473 removed += mark_lseg_invalid(lseg, tmp_list);
474 }
475 dprintk("%s:Return %i\n", __func__, invalid - removed);
476 return invalid - removed;
Andy Adamson974cec82010-10-20 00:18:02 -0400477}
478
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000479/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000480void
Fred Isaman4541d162011-01-06 11:36:23 +0000481pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400482{
Fred Isaman4541d162011-01-06 11:36:23 +0000483 struct pnfs_layout_segment *lseg, *tmp;
Andy Adamson974cec82010-10-20 00:18:02 -0400484
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000485 if (list_empty(free_me))
486 return;
487
Fred Isaman4541d162011-01-06 11:36:23 +0000488 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000489 list_del(&lseg->pls_list);
Trond Myklebust905ca192012-09-20 20:46:49 -0400490 pnfs_free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400491 }
492}
493
Benny Halevye5e94012010-10-20 00:18:01 -0400494void
495pnfs_destroy_layout(struct nfs_inode *nfsi)
496{
497 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400498 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400499
500 spin_lock(&nfsi->vfs_inode.i_lock);
501 lo = nfsi->layout;
502 if (lo) {
Fred Isaman38511722011-02-03 18:28:50 +0000503 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
Trond Myklebust49a85062012-09-18 20:43:31 -0400504 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
Trond Myklebust3e621212012-09-24 13:07:16 -0400505 pnfs_get_layout_hdr(lo);
506 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
507 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
508 spin_unlock(&nfsi->vfs_inode.i_lock);
509 pnfs_free_lseg_list(&tmp_list);
510 pnfs_put_layout_hdr(lo);
511 } else
512 spin_unlock(&nfsi->vfs_inode.i_lock);
Benny Halevye5e94012010-10-20 00:18:01 -0400513}
Andy Adamson041245c2012-04-27 17:53:53 -0400514EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
Benny Halevye5e94012010-10-20 00:18:01 -0400515
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500516static bool
517pnfs_layout_add_bulk_destroy_list(struct inode *inode,
518 struct list_head *layout_list)
519{
520 struct pnfs_layout_hdr *lo;
521 bool ret = false;
522
523 spin_lock(&inode->i_lock);
524 lo = NFS_I(inode)->layout;
525 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
526 pnfs_get_layout_hdr(lo);
527 list_add(&lo->plh_bulk_destroy, layout_list);
528 ret = true;
529 }
530 spin_unlock(&inode->i_lock);
531 return ret;
532}
533
534/* Caller must hold rcu_read_lock and clp->cl_lock */
535static int
536pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
537 struct nfs_server *server,
538 struct list_head *layout_list)
539{
540 struct pnfs_layout_hdr *lo, *next;
541 struct inode *inode;
542
543 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
544 inode = igrab(lo->plh_inode);
545 if (inode == NULL)
546 continue;
547 list_del_init(&lo->plh_layouts);
548 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
549 continue;
550 rcu_read_unlock();
551 spin_unlock(&clp->cl_lock);
552 iput(inode);
553 spin_lock(&clp->cl_lock);
554 rcu_read_lock();
555 return -EAGAIN;
556 }
557 return 0;
558}
559
560static int
561pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
562 bool is_bulk_recall)
563{
564 struct pnfs_layout_hdr *lo;
565 struct inode *inode;
566 struct pnfs_layout_range range = {
567 .iomode = IOMODE_ANY,
568 .offset = 0,
569 .length = NFS4_MAX_UINT64,
570 };
571 LIST_HEAD(lseg_list);
572 int ret = 0;
573
574 while (!list_empty(layout_list)) {
575 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
576 plh_bulk_destroy);
577 dprintk("%s freeing layout for inode %lu\n", __func__,
578 lo->plh_inode->i_ino);
579 inode = lo->plh_inode;
580 spin_lock(&inode->i_lock);
581 list_del_init(&lo->plh_bulk_destroy);
582 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
583 if (is_bulk_recall)
584 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
585 if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range))
586 ret = -EAGAIN;
587 spin_unlock(&inode->i_lock);
588 pnfs_free_lseg_list(&lseg_list);
589 pnfs_put_layout_hdr(lo);
590 iput(inode);
591 }
592 return ret;
593}
594
595int
596pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
597 struct nfs_fsid *fsid,
598 bool is_recall)
599{
600 struct nfs_server *server;
601 LIST_HEAD(layout_list);
602
603 spin_lock(&clp->cl_lock);
604 rcu_read_lock();
605restart:
606 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
607 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
608 continue;
609 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
610 server,
611 &layout_list) != 0)
612 goto restart;
613 }
614 rcu_read_unlock();
615 spin_unlock(&clp->cl_lock);
616
617 if (list_empty(&layout_list))
618 return 0;
619 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
620}
621
622int
623pnfs_destroy_layouts_byclid(struct nfs_client *clp,
624 bool is_recall)
625{
626 struct nfs_server *server;
627 LIST_HEAD(layout_list);
628
629 spin_lock(&clp->cl_lock);
630 rcu_read_lock();
631restart:
632 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
633 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
634 server,
635 &layout_list) != 0)
636 goto restart;
637 }
638 rcu_read_unlock();
639 spin_unlock(&clp->cl_lock);
640
641 if (list_empty(&layout_list))
642 return 0;
643 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
644}
645
Andy Adamson974cec82010-10-20 00:18:02 -0400646/*
647 * Called by the state manger to remove all layouts established under an
648 * expired lease.
649 */
650void
651pnfs_destroy_all_layouts(struct nfs_client *clp)
652{
Andy Adamsonc47abcf2011-06-15 17:52:40 -0400653 nfs4_deviceid_mark_client_invalid(clp);
654 nfs4_deviceid_purge_client(clp);
655
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500656 pnfs_destroy_layouts_byclid(clp, false);
Andy Adamson974cec82010-10-20 00:18:02 -0400657}
658
Trond Myklebust5a655032012-10-02 16:47:14 -0700659/*
660 * Compare 2 layout stateid sequence ids, to see which is newer,
661 * taking into account wraparound issues.
662 */
663static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
664{
665 return (s32)s1 - (s32)s2 > 0;
666}
667
Fred Isamanfd6002e2011-01-06 11:36:22 +0000668/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000669void
670pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
671 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400672{
Trond Myklebust22aaf712012-10-04 16:32:22 -0700673 u32 oldseq, newseq, new_barrier;
674 int empty = list_empty(&lo->plh_segs);
Andy Adamson974cec82010-10-20 00:18:02 -0400675
Trond Myklebust2d2f24a2012-03-04 18:13:57 -0500676 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
677 newseq = be32_to_cpu(new->seqid);
Trond Myklebust22aaf712012-10-04 16:32:22 -0700678 if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
Trond Myklebustf597c532012-03-04 18:13:56 -0500679 nfs4_stateid_copy(&lo->plh_stateid, new);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000680 if (update_barrier) {
Trond Myklebust22aaf712012-10-04 16:32:22 -0700681 new_barrier = be32_to_cpu(new->seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000682 } else {
683 /* Because of wraparound, we want to keep the barrier
Trond Myklebust22aaf712012-10-04 16:32:22 -0700684 * "close" to the current seqids.
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000685 */
Trond Myklebust22aaf712012-10-04 16:32:22 -0700686 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000687 }
Trond Myklebust22aaf712012-10-04 16:32:22 -0700688 if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
689 lo->plh_barrier = new_barrier;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000690 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400691}
692
Trond Myklebust19c54ab2012-10-05 16:56:58 -0700693static bool
694pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
695 const nfs4_stateid *stateid)
696{
697 u32 seqid = be32_to_cpu(stateid->seqid);
698
699 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
700}
701
Fred Isamancf7d63f2011-01-06 11:36:25 +0000702/* lget is set to 1 if called from inside send_layoutget call chain */
703static bool
Trond Myklebust19c54ab2012-10-05 16:56:58 -0700704pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000705{
Fred Isamanf7e89172011-01-06 11:36:32 +0000706 return lo->plh_block_lgets ||
707 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000708 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000709 (atomic_read(&lo->plh_outstanding) > lget));
710}
711
Fred Isamanfd6002e2011-01-06 11:36:22 +0000712int
713pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
714 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400715{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000716 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400717
718 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000719 spin_lock(&lo->plh_inode->i_lock);
Trond Myklebust19c54ab2012-10-05 16:56:58 -0700720 if (pnfs_layoutgets_blocked(lo, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000721 status = -EAGAIN;
Trond Myklebust5d422302013-03-14 16:57:48 -0400722 } else if (!nfs4_valid_open_stateid(open_state)) {
723 status = -EBADF;
Fred Isamancf7d63f2011-01-06 11:36:25 +0000724 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000725 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400726
Fred Isamanfd6002e2011-01-06 11:36:22 +0000727 do {
728 seq = read_seqbegin(&open_state->seqlock);
Trond Myklebustf597c532012-03-04 18:13:56 -0500729 nfs4_stateid_copy(dst, &open_state->stateid);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000730 } while (read_seqretry(&open_state->seqlock, seq));
731 } else
Trond Myklebustf597c532012-03-04 18:13:56 -0500732 nfs4_stateid_copy(dst, &lo->plh_stateid);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000733 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400734 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000735 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400736}
737
738/*
739* Get layout from server.
740* for now, assume that whole file layouts are requested.
741* arg->offset: 0
742* arg->length: all ones
743*/
Benny Halevye5e94012010-10-20 00:18:01 -0400744static struct pnfs_layout_segment *
745send_layoutget(struct pnfs_layout_hdr *lo,
746 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +0300747 struct pnfs_layout_range *range,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400748 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400749{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000750 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400751 struct nfs_server *server = NFS_SERVER(ino);
752 struct nfs4_layoutget *lgp;
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -0400753 struct pnfs_layout_segment *lseg;
Benny Halevye5e94012010-10-20 00:18:01 -0400754
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400755 dprintk("--> %s\n", __func__);
756
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400757 lgp = kzalloc(sizeof(*lgp), gfp_flags);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000758 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400759 return NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400760
Benny Halevyfb3296e2011-05-22 19:47:26 +0300761 lgp->args.minlength = PAGE_CACHE_SIZE;
762 if (lgp->args.minlength > range->length)
763 lgp->args.minlength = range->length;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400764 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
Benny Halevyfb3296e2011-05-22 19:47:26 +0300765 lgp->args.range = *range;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400766 lgp->args.type = server->pnfs_curr_ld->id;
767 lgp->args.inode = ino;
768 lgp->args.ctx = get_nfs_open_context(ctx);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400769 lgp->gfp_flags = gfp_flags;
Trond Myklebust6ab59342013-05-20 10:49:34 -0400770 lgp->cred = lo->plh_lc_cred;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400771
772 /* Synchronously retrieve layout information from server and
773 * store in lseg.
774 */
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -0400775 lseg = nfs4_proc_layoutget(lgp, gfp_flags);
776 if (IS_ERR(lseg)) {
777 switch (PTR_ERR(lseg)) {
778 case -ENOMEM:
779 case -ERESTARTSYS:
780 break;
781 default:
782 /* remember that LAYOUTGET failed and suspend trying */
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400783 pnfs_layout_io_set_failed(lo, range->iomode);
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -0400784 }
785 return NULL;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400786 }
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400787
Andy Adamson974cec82010-10-20 00:18:02 -0400788 return lseg;
789}
790
Trond Myklebust24956802013-03-20 13:03:00 -0400791static void pnfs_clear_layoutcommit(struct inode *inode,
792 struct list_head *head)
793{
794 struct nfs_inode *nfsi = NFS_I(inode);
795 struct pnfs_layout_segment *lseg, *tmp;
796
797 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
798 return;
799 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
800 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
801 continue;
802 pnfs_lseg_dec_and_remove_zero(lseg, head);
803 }
804}
805
Andy Adamson293b3b02012-06-20 15:03:34 -0400806/*
807 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
808 * when the layout segment list is empty.
809 *
810 * Note that a pnfs_layout_hdr can exist with an empty layout segment
811 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
812 * deviceid is marked invalid.
813 */
Benny Halevycbe82602011-05-22 19:52:37 +0300814int
815_pnfs_return_layout(struct inode *ino)
816{
817 struct pnfs_layout_hdr *lo = NULL;
818 struct nfs_inode *nfsi = NFS_I(ino);
819 LIST_HEAD(tmp_list);
820 struct nfs4_layoutreturn *lrp;
821 nfs4_stateid stateid;
Andy Adamson293b3b02012-06-20 15:03:34 -0400822 int status = 0, empty;
Benny Halevycbe82602011-05-22 19:52:37 +0300823
Andy Adamson366d5052012-06-20 15:03:33 -0400824 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
Benny Halevycbe82602011-05-22 19:52:37 +0300825
826 spin_lock(&ino->i_lock);
827 lo = nfsi->layout;
Trond Myklebuste5929f32012-09-21 16:37:02 -0400828 if (!lo) {
Benny Halevycbe82602011-05-22 19:52:37 +0300829 spin_unlock(&ino->i_lock);
Andy Adamson293b3b02012-06-20 15:03:34 -0400830 dprintk("NFS: %s no layout to return\n", __func__);
831 goto out;
Benny Halevycbe82602011-05-22 19:52:37 +0300832 }
833 stateid = nfsi->layout->plh_stateid;
834 /* Reference matched in nfs4_layoutreturn_release */
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400835 pnfs_get_layout_hdr(lo);
Andy Adamson293b3b02012-06-20 15:03:34 -0400836 empty = list_empty(&lo->plh_segs);
Trond Myklebust24956802013-03-20 13:03:00 -0400837 pnfs_clear_layoutcommit(ino, &tmp_list);
Trond Myklebust49a85062012-09-18 20:43:31 -0400838 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
Andy Adamson293b3b02012-06-20 15:03:34 -0400839 /* Don't send a LAYOUTRETURN if list was initially empty */
840 if (empty) {
841 spin_unlock(&ino->i_lock);
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400842 pnfs_put_layout_hdr(lo);
Andy Adamson293b3b02012-06-20 15:03:34 -0400843 dprintk("NFS: %s no layout segments to return\n", __func__);
844 goto out;
845 }
Fred Isamanea0ded72011-06-15 12:31:02 -0400846 lo->plh_block_lgets++;
Benny Halevycbe82602011-05-22 19:52:37 +0300847 spin_unlock(&ino->i_lock);
848 pnfs_free_lseg_list(&tmp_list);
849
Benny Halevycbe82602011-05-22 19:52:37 +0300850 lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
851 if (unlikely(lrp == NULL)) {
852 status = -ENOMEM;
Trond Myklebust65857d52012-09-24 13:49:27 -0400853 spin_lock(&ino->i_lock);
854 lo->plh_block_lgets--;
855 spin_unlock(&ino->i_lock);
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400856 pnfs_put_layout_hdr(lo);
Benny Halevycbe82602011-05-22 19:52:37 +0300857 goto out;
858 }
859
860 lrp->args.stateid = stateid;
861 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
862 lrp->args.inode = ino;
Trond Myklebusta56aaa02011-06-15 11:59:10 -0400863 lrp->args.layout = lo;
Benny Halevycbe82602011-05-22 19:52:37 +0300864 lrp->clp = NFS_SERVER(ino)->nfs_client;
Trond Myklebust95560002013-05-20 10:43:47 -0400865 lrp->cred = lo->plh_lc_cred;
Benny Halevycbe82602011-05-22 19:52:37 +0300866
867 status = nfs4_proc_layoutreturn(lrp);
868out:
869 dprintk("<-- %s status: %d\n", __func__, status);
870 return status;
871}
Andy Adamson0a57cda2012-04-27 17:53:50 -0400872EXPORT_SYMBOL_GPL(_pnfs_return_layout);
Benny Halevycbe82602011-05-22 19:52:37 +0300873
Trond Myklebust24028672013-03-20 13:23:33 -0400874int
875pnfs_commit_and_return_layout(struct inode *inode)
876{
877 struct pnfs_layout_hdr *lo;
878 int ret;
879
880 spin_lock(&inode->i_lock);
881 lo = NFS_I(inode)->layout;
882 if (lo == NULL) {
883 spin_unlock(&inode->i_lock);
884 return 0;
885 }
886 pnfs_get_layout_hdr(lo);
887 /* Block new layoutgets and read/write to ds */
888 lo->plh_block_lgets++;
889 spin_unlock(&inode->i_lock);
890 filemap_fdatawait(inode->i_mapping);
891 ret = pnfs_layoutcommit_inode(inode, true);
892 if (ret == 0)
893 ret = _pnfs_return_layout(inode);
894 spin_lock(&inode->i_lock);
895 lo->plh_block_lgets--;
896 spin_unlock(&inode->i_lock);
897 pnfs_put_layout_hdr(lo);
898 return ret;
899}
900
Fred Isamanf7e89172011-01-06 11:36:32 +0000901bool pnfs_roc(struct inode *ino)
902{
903 struct pnfs_layout_hdr *lo;
904 struct pnfs_layout_segment *lseg, *tmp;
905 LIST_HEAD(tmp_list);
906 bool found = false;
907
908 spin_lock(&ino->i_lock);
909 lo = NFS_I(ino)->layout;
910 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
911 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
912 goto out_nolayout;
913 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
914 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
915 mark_lseg_invalid(lseg, &tmp_list);
916 found = true;
917 }
918 if (!found)
919 goto out_nolayout;
920 lo->plh_block_lgets++;
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400921 pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
Fred Isamanf7e89172011-01-06 11:36:32 +0000922 spin_unlock(&ino->i_lock);
923 pnfs_free_lseg_list(&tmp_list);
924 return true;
925
926out_nolayout:
927 spin_unlock(&ino->i_lock);
928 return false;
929}
930
931void pnfs_roc_release(struct inode *ino)
932{
933 struct pnfs_layout_hdr *lo;
934
935 spin_lock(&ino->i_lock);
936 lo = NFS_I(ino)->layout;
937 lo->plh_block_lgets--;
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400938 if (atomic_dec_and_test(&lo->plh_refcount)) {
939 pnfs_detach_layout_hdr(lo);
940 spin_unlock(&ino->i_lock);
941 pnfs_free_layout_hdr(lo);
942 } else
943 spin_unlock(&ino->i_lock);
Fred Isamanf7e89172011-01-06 11:36:32 +0000944}
945
946void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
947{
948 struct pnfs_layout_hdr *lo;
949
950 spin_lock(&ino->i_lock);
951 lo = NFS_I(ino)->layout;
Trond Myklebust0f35ad62012-10-04 16:28:17 -0700952 if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
Fred Isamanf7e89172011-01-06 11:36:32 +0000953 lo->plh_barrier = barrier;
954 spin_unlock(&ino->i_lock);
955}
956
Trond Myklebust7fdab062012-09-20 20:15:57 -0400957bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
Fred Isamanf7e89172011-01-06 11:36:32 +0000958{
959 struct nfs_inode *nfsi = NFS_I(ino);
Trond Myklebust7fdab062012-09-20 20:15:57 -0400960 struct pnfs_layout_hdr *lo;
Fred Isamanf7e89172011-01-06 11:36:32 +0000961 struct pnfs_layout_segment *lseg;
Trond Myklebust7fdab062012-09-20 20:15:57 -0400962 u32 current_seqid;
Fred Isamanf7e89172011-01-06 11:36:32 +0000963 bool found = false;
964
965 spin_lock(&ino->i_lock);
966 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
967 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
Trond Myklebust7fdab062012-09-20 20:15:57 -0400968 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
Fred Isamanf7e89172011-01-06 11:36:32 +0000969 found = true;
Trond Myklebust7fdab062012-09-20 20:15:57 -0400970 goto out;
Fred Isamanf7e89172011-01-06 11:36:32 +0000971 }
Trond Myklebust7fdab062012-09-20 20:15:57 -0400972 lo = nfsi->layout;
973 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
Fred Isamanf7e89172011-01-06 11:36:32 +0000974
Trond Myklebust7fdab062012-09-20 20:15:57 -0400975 /* Since close does not return a layout stateid for use as
976 * a barrier, we choose the worst-case barrier.
977 */
978 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
979out:
Fred Isamanf7e89172011-01-06 11:36:32 +0000980 spin_unlock(&ino->i_lock);
981 return found;
982}
983
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400984/*
985 * Compare two layout segments for sorting into layout cache.
986 * We want to preferentially return RW over RO layouts, so ensure those
987 * are seen first.
988 */
989static s64
Trond Myklebust7dc0ac72013-06-03 11:30:24 -0400990pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
Trond Myklebust3cb2df12013-06-03 11:24:36 -0400991 const struct pnfs_layout_range *l2)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400992{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300993 s64 d;
994
995 /* high offset > low offset */
996 d = l1->offset - l2->offset;
997 if (d)
998 return d;
999
1000 /* short length > long length */
1001 d = l2->length - l1->length;
1002 if (d)
1003 return d;
1004
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001005 /* read > read/write */
Benny Halevyfb3296e2011-05-22 19:47:26 +03001006 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001007}
1008
Andy Adamson974cec82010-10-20 00:18:02 -04001009static void
Trond Myklebust57036a32012-09-20 16:33:30 -04001010pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
Andy Adamson974cec82010-10-20 00:18:02 -04001011 struct pnfs_layout_segment *lseg)
1012{
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001013 struct pnfs_layout_segment *lp;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001014
Andy Adamson974cec82010-10-20 00:18:02 -04001015 dprintk("%s:Begin\n", __func__);
1016
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001017 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Trond Myklebust7dc0ac72013-06-03 11:30:24 -04001018 if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001019 continue;
Fred Isaman566052c2011-01-06 11:36:20 +00001020 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001021 dprintk("%s: inserted lseg %p "
1022 "iomode %d offset %llu length %llu before "
1023 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +00001024 __func__, lseg, lseg->pls_range.iomode,
1025 lseg->pls_range.offset, lseg->pls_range.length,
1026 lp, lp->pls_range.iomode, lp->pls_range.offset,
1027 lp->pls_range.length);
Benny Halevyfb3296e2011-05-22 19:47:26 +03001028 goto out;
Andy Adamson974cec82010-10-20 00:18:02 -04001029 }
Benny Halevyfb3296e2011-05-22 19:47:26 +03001030 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1031 dprintk("%s: inserted lseg %p "
1032 "iomode %d offset %llu length %llu at tail\n",
1033 __func__, lseg, lseg->pls_range.iomode,
1034 lseg->pls_range.offset, lseg->pls_range.length);
1035out:
Trond Myklebust70c3bd22012-09-18 20:51:13 -04001036 pnfs_get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -04001037
1038 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -04001039}
1040
1041static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -04001042alloc_init_layout_hdr(struct inode *ino,
1043 struct nfs_open_context *ctx,
1044 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -04001045{
1046 struct pnfs_layout_hdr *lo;
1047
Benny Halevy636fb9c2011-05-22 19:51:33 +03001048 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -04001049 if (!lo)
1050 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +00001051 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001052 INIT_LIST_HEAD(&lo->plh_layouts);
1053 INIT_LIST_HEAD(&lo->plh_segs);
Trond Myklebustfd9a8d72013-02-12 09:48:42 -05001054 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001055 lo->plh_inode = ino;
Trond Myklebust5cc22162013-05-21 09:26:49 -04001056 lo->plh_lc_cred = get_rpccred(ctx->cred);
Benny Halevye5e94012010-10-20 00:18:01 -04001057 return lo;
1058}
1059
1060static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -04001061pnfs_find_alloc_layout(struct inode *ino,
1062 struct nfs_open_context *ctx,
1063 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -04001064{
1065 struct nfs_inode *nfsi = NFS_I(ino);
1066 struct pnfs_layout_hdr *new = NULL;
1067
1068 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1069
Trond Myklebust251ec412012-10-02 15:41:05 -07001070 if (nfsi->layout != NULL)
1071 goto out_existing;
Benny Halevye5e94012010-10-20 00:18:01 -04001072 spin_unlock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -04001073 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -04001074 spin_lock(&ino->i_lock);
1075
Trond Myklebust251ec412012-10-02 15:41:05 -07001076 if (likely(nfsi->layout == NULL)) { /* Won the race? */
Benny Halevye5e94012010-10-20 00:18:01 -04001077 nfsi->layout = new;
Trond Myklebust251ec412012-10-02 15:41:05 -07001078 return new;
Yanchuan Nian7175fe92012-10-31 16:05:48 +08001079 } else if (new != NULL)
1080 pnfs_free_layout_hdr(new);
Trond Myklebust251ec412012-10-02 15:41:05 -07001081out_existing:
1082 pnfs_get_layout_hdr(nfsi->layout);
Benny Halevye5e94012010-10-20 00:18:01 -04001083 return nfsi->layout;
1084}
1085
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001086/*
1087 * iomode matching rules:
1088 * iomode lseg match
1089 * ----- ----- -----
1090 * ANY READ true
1091 * ANY RW true
1092 * RW READ false
1093 * RW RW true
1094 * READ READ true
1095 * READ RW true
1096 */
Trond Myklebust3cb2df12013-06-03 11:24:36 -04001097static bool
Trond Myklebust7dc0ac72013-06-03 11:30:24 -04001098pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
Trond Myklebust3cb2df12013-06-03 11:24:36 -04001099 const struct pnfs_layout_range *range)
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001100{
Benny Halevyfb3296e2011-05-22 19:47:26 +03001101 struct pnfs_layout_range range1;
1102
1103 if ((range->iomode == IOMODE_RW &&
1104 ls_range->iomode != IOMODE_RW) ||
Trond Myklebust7dc0ac72013-06-03 11:30:24 -04001105 !pnfs_lseg_range_intersecting(ls_range, range))
Benny Halevyfb3296e2011-05-22 19:47:26 +03001106 return 0;
1107
1108 /* range1 covers only the first byte in the range */
1109 range1 = *range;
1110 range1.length = 1;
Trond Myklebust7dc0ac72013-06-03 11:30:24 -04001111 return pnfs_lseg_range_contained(ls_range, &range1);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001112}
1113
1114/*
1115 * lookup range in layout
1116 */
Benny Halevye5e94012010-10-20 00:18:01 -04001117static struct pnfs_layout_segment *
Benny Halevyfb3296e2011-05-22 19:47:26 +03001118pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1119 struct pnfs_layout_range *range)
Benny Halevye5e94012010-10-20 00:18:01 -04001120{
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001121 struct pnfs_layout_segment *lseg, *ret = NULL;
1122
1123 dprintk("%s:Begin\n", __func__);
1124
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001125 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +00001126 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
Trond Myklebust7dc0ac72013-06-03 11:30:24 -04001127 pnfs_lseg_range_match(&lseg->pls_range, range)) {
Trond Myklebust9369a432012-09-18 20:57:08 -04001128 ret = pnfs_get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001129 break;
1130 }
Benny Halevyd771e3a2011-06-14 16:30:16 -04001131 if (lseg->pls_range.offset > range->offset)
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001132 break;
1133 }
1134
1135 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +00001136 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001137 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -04001138}
1139
1140/*
Andy Adamsond23d61c2012-05-23 05:02:37 -04001141 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1142 * to the MDS or over pNFS
1143 *
1144 * The nfs_inode read_io and write_io fields are cumulative counters reset
1145 * when there are no layout segments. Note that in pnfs_update_layout iomode
1146 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1147 * WRITE request.
1148 *
1149 * A return of true means use MDS I/O.
1150 *
1151 * From rfc 5661:
1152 * If a file's size is smaller than the file size threshold, data accesses
1153 * SHOULD be sent to the metadata server. If an I/O request has a length that
1154 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1155 * server. If both file size and I/O size are provided, the client SHOULD
1156 * reach or exceed both thresholds before sending its read or write
1157 * requests to the data server.
1158 */
1159static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1160 struct inode *ino, int iomode)
1161{
1162 struct nfs4_threshold *t = ctx->mdsthreshold;
1163 struct nfs_inode *nfsi = NFS_I(ino);
1164 loff_t fsize = i_size_read(ino);
1165 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1166
1167 if (t == NULL)
1168 return ret;
1169
1170 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1171 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1172
1173 switch (iomode) {
1174 case IOMODE_READ:
1175 if (t->bm & THRESHOLD_RD) {
1176 dprintk("%s fsize %llu\n", __func__, fsize);
1177 size_set = true;
1178 if (fsize < t->rd_sz)
1179 size = true;
1180 }
1181 if (t->bm & THRESHOLD_RD_IO) {
1182 dprintk("%s nfsi->read_io %llu\n", __func__,
1183 nfsi->read_io);
1184 io_set = true;
1185 if (nfsi->read_io < t->rd_io_sz)
1186 io = true;
1187 }
1188 break;
1189 case IOMODE_RW:
1190 if (t->bm & THRESHOLD_WR) {
1191 dprintk("%s fsize %llu\n", __func__, fsize);
1192 size_set = true;
1193 if (fsize < t->wr_sz)
1194 size = true;
1195 }
1196 if (t->bm & THRESHOLD_WR_IO) {
1197 dprintk("%s nfsi->write_io %llu\n", __func__,
1198 nfsi->write_io);
1199 io_set = true;
1200 if (nfsi->write_io < t->wr_io_sz)
1201 io = true;
1202 }
1203 break;
1204 }
1205 if (size_set && io_set) {
1206 if (size && io)
1207 ret = true;
1208 } else if (size || io)
1209 ret = true;
1210
1211 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1212 return ret;
1213}
1214
1215/*
Benny Halevye5e94012010-10-20 00:18:01 -04001216 * Layout segment is retreived from the server if not cached.
1217 * The appropriate layout segment is referenced and returned to the caller.
1218 */
Andy Adamson7c24d942011-06-13 18:22:38 -04001219struct pnfs_layout_segment *
Benny Halevye5e94012010-10-20 00:18:01 -04001220pnfs_update_layout(struct inode *ino,
1221 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +03001222 loff_t pos,
1223 u64 count,
Trond Myklebusta75b9df2011-05-11 18:00:51 -04001224 enum pnfs_iomode iomode,
1225 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -04001226{
Benny Halevyfb3296e2011-05-22 19:47:26 +03001227 struct pnfs_layout_range arg = {
1228 .iomode = iomode,
1229 .offset = pos,
1230 .length = count,
1231 };
Benny Halevy707ed5f2011-05-22 19:47:46 +03001232 unsigned pg_offset;
Weston Andros Adamson6382a442011-06-01 16:44:44 -04001233 struct nfs_server *server = NFS_SERVER(ino);
1234 struct nfs_client *clp = server->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -04001235 struct pnfs_layout_hdr *lo;
1236 struct pnfs_layout_segment *lseg = NULL;
Weston Andros Adamson30005122013-02-28 20:30:10 -05001237 bool first;
Benny Halevye5e94012010-10-20 00:18:01 -04001238
1239 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04001240 goto out;
Andy Adamsond23d61c2012-05-23 05:02:37 -04001241
1242 if (pnfs_within_mdsthreshold(ctx, ino, iomode))
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04001243 goto out;
Andy Adamsond23d61c2012-05-23 05:02:37 -04001244
Benny Halevye5e94012010-10-20 00:18:01 -04001245 spin_lock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -04001246 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
Trond Myklebust830ffb52012-09-20 21:25:19 -04001247 if (lo == NULL) {
1248 spin_unlock(&ino->i_lock);
1249 goto out;
1250 }
Benny Halevye5e94012010-10-20 00:18:01 -04001251
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001252 /* Do we even need to bother with this? */
Trond Myklebusta59c30a2012-03-01 11:17:47 -05001253 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001254 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -04001255 goto out_unlock;
1256 }
1257
1258 /* if LAYOUTGET already failed once we don't try again */
Trond Myklebustb9e028f2012-09-18 16:41:18 -04001259 if (pnfs_layout_io_test_failed(lo, iomode))
Benny Halevye5e94012010-10-20 00:18:01 -04001260 goto out_unlock;
1261
Andy Adamson568e8c42011-03-01 01:34:22 +00001262 /* Check to see if the layout for the given range already exists */
Benny Halevyfb3296e2011-05-22 19:47:26 +03001263 lseg = pnfs_find_lseg(lo, &arg);
Andy Adamson568e8c42011-03-01 01:34:22 +00001264 if (lseg)
1265 goto out_unlock;
1266
Trond Myklebust19c54ab2012-10-05 16:56:58 -07001267 if (pnfs_layoutgets_blocked(lo, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +00001268 goto out_unlock;
1269 atomic_inc(&lo->plh_outstanding);
1270
Weston Andros Adamson30005122013-02-28 20:30:10 -05001271 first = list_empty(&lo->plh_layouts) ? true : false;
Fred Isamanf49f9ba2011-02-03 18:28:52 +00001272 spin_unlock(&ino->i_lock);
Weston Andros Adamson30005122013-02-28 20:30:10 -05001273
Fred Isamanf49f9ba2011-02-03 18:28:52 +00001274 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +00001275 /* The lo must be on the clp list if there is any
1276 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1277 */
1278 spin_lock(&clp->cl_lock);
Weston Andros Adamson6382a442011-06-01 16:44:44 -04001279 list_add_tail(&lo->plh_layouts, &server->layouts);
Fred Isaman2130ff62011-01-06 11:36:26 +00001280 spin_unlock(&clp->cl_lock);
1281 }
Benny Halevye5e94012010-10-20 00:18:01 -04001282
Benny Halevy707ed5f2011-05-22 19:47:46 +03001283 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1284 if (pg_offset) {
1285 arg.offset -= pg_offset;
1286 arg.length += pg_offset;
1287 }
Andy Adamson7c24d942011-06-13 18:22:38 -04001288 if (arg.length != NFS4_MAX_UINT64)
1289 arg.length = PAGE_CACHE_ALIGN(arg.length);
Benny Halevy707ed5f2011-05-22 19:47:46 +03001290
Benny Halevyfb3296e2011-05-22 19:47:26 +03001291 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
Fred Isamancf7d63f2011-01-06 11:36:25 +00001292 atomic_dec(&lo->plh_outstanding);
Trond Myklebust830ffb52012-09-20 21:25:19 -04001293out_put_layout_hdr:
Trond Myklebust70c3bd22012-09-18 20:51:13 -04001294 pnfs_put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -04001295out:
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04001296 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1297 "(%s, offset: %llu, length: %llu)\n",
1298 __func__, ino->i_sb->s_id,
1299 (unsigned long long)NFS_FILEID(ino),
1300 lseg == NULL ? "not found" : "found",
1301 iomode==IOMODE_RW ? "read/write" : "read-only",
1302 (unsigned long long)pos,
1303 (unsigned long long)count);
Benny Halevye5e94012010-10-20 00:18:01 -04001304 return lseg;
1305out_unlock:
1306 spin_unlock(&ino->i_lock);
Trond Myklebust830ffb52012-09-20 21:25:19 -04001307 goto out_put_layout_hdr;
Benny Halevye5e94012010-10-20 00:18:01 -04001308}
Andy Adamson7c24d942011-06-13 18:22:38 -04001309EXPORT_SYMBOL_GPL(pnfs_update_layout);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001310
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -04001311struct pnfs_layout_segment *
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001312pnfs_layout_process(struct nfs4_layoutget *lgp)
1313{
1314 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1315 struct nfs4_layoutget_res *res = &lgp->res;
1316 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001317 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001318 int status = 0;
1319
1320 /* Inject layout blob into I/O device driver */
Trond Myklebusta75b9df2011-05-11 18:00:51 -04001321 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001322 if (!lseg || IS_ERR(lseg)) {
1323 if (!lseg)
1324 status = -ENOMEM;
1325 else
1326 status = PTR_ERR(lseg);
1327 dprintk("%s: Could not allocate layout: error %d\n",
1328 __func__, status);
1329 goto out;
1330 }
1331
1332 spin_lock(&ino->i_lock);
Trond Myklebusta59c30a2012-03-01 11:17:47 -05001333 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001334 dprintk("%s forget reply due to recall\n", __func__);
1335 goto out_forget_reply;
1336 }
1337
Trond Myklebust19c54ab2012-10-05 16:56:58 -07001338 if (pnfs_layoutgets_blocked(lo, 1) ||
1339 pnfs_layout_stateid_blocked(lo, &res->stateid)) {
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001340 dprintk("%s forget reply due to state\n", __func__);
1341 goto out_forget_reply;
1342 }
Trond Myklebust038d6492012-10-02 16:38:41 -07001343
1344 /* Done processing layoutget. Set the layout stateid */
1345 pnfs_set_layout_stateid(lo, &res->stateid, false);
1346
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001347 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +00001348 lseg->pls_range = res->range;
Trond Myklebust9369a432012-09-18 20:57:08 -04001349 pnfs_get_lseg(lseg);
Trond Myklebust57036a32012-09-20 16:33:30 -04001350 pnfs_layout_insert_lseg(lo, lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001351
Fred Isamanf7e89172011-01-06 11:36:32 +00001352 if (res->return_on_close) {
1353 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1354 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
1355 }
1356
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001357 spin_unlock(&ino->i_lock);
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -04001358 return lseg;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001359out:
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -04001360 return ERR_PTR(status);
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001361
1362out_forget_reply:
1363 spin_unlock(&ino->i_lock);
1364 lseg->pls_layout = lo;
1365 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1366 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001367}
1368
Trond Myklebustd8007d42011-06-10 13:30:23 -04001369void
1370pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1371{
Peng Tao1fd937b2012-09-25 14:55:57 +08001372 u64 rd_size = req->wb_bytes;
1373
Trond Myklebustbc5a89b2012-10-15 14:58:04 -04001374 WARN_ON_ONCE(pgio->pg_lseg != NULL);
Trond Myklebustd8007d42011-06-10 13:30:23 -04001375
Fred Isaman1825a0d2012-04-20 19:55:31 -04001376 if (req->wb_offset != req->wb_pgbase) {
1377 nfs_pageio_reset_read_mds(pgio);
1378 return;
1379 }
Peng Tao1fd937b2012-09-25 14:55:57 +08001380
1381 if (pgio->pg_dreq == NULL)
1382 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
1383 else
1384 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1385
Trond Myklebustd8007d42011-06-10 13:30:23 -04001386 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1387 req->wb_context,
1388 req_offset(req),
Peng Tao1fd937b2012-09-25 14:55:57 +08001389 rd_size,
Trond Myklebustd8007d42011-06-10 13:30:23 -04001390 IOMODE_READ,
1391 GFP_KERNEL);
Trond Myklebuste885de12011-06-10 13:30:23 -04001392 /* If no lseg, fall back to read through mds */
1393 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001394 nfs_pageio_reset_read_mds(pgio);
Trond Myklebuste885de12011-06-10 13:30:23 -04001395
Trond Myklebustd8007d42011-06-10 13:30:23 -04001396}
1397EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1398
1399void
Peng Tao62965562012-09-25 14:55:57 +08001400pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
1401 struct nfs_page *req, u64 wb_size)
Trond Myklebustd8007d42011-06-10 13:30:23 -04001402{
Trond Myklebustbc5a89b2012-10-15 14:58:04 -04001403 WARN_ON_ONCE(pgio->pg_lseg != NULL);
Trond Myklebustd8007d42011-06-10 13:30:23 -04001404
Fred Isaman1825a0d2012-04-20 19:55:31 -04001405 if (req->wb_offset != req->wb_pgbase) {
1406 nfs_pageio_reset_write_mds(pgio);
1407 return;
1408 }
Peng Tao62965562012-09-25 14:55:57 +08001409
Trond Myklebustd8007d42011-06-10 13:30:23 -04001410 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1411 req->wb_context,
1412 req_offset(req),
Peng Tao62965562012-09-25 14:55:57 +08001413 wb_size,
Trond Myklebustd8007d42011-06-10 13:30:23 -04001414 IOMODE_RW,
1415 GFP_NOFS);
Trond Myklebuste885de12011-06-10 13:30:23 -04001416 /* If no lseg, fall back to write through mds */
1417 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001418 nfs_pageio_reset_write_mds(pgio);
Trond Myklebustd8007d42011-06-10 13:30:23 -04001419}
1420EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1421
Bryan Schumaker1abb50882012-06-20 15:53:47 -04001422void
Fred Isaman061ae2e2012-04-20 14:47:48 -04001423pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode,
1424 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001425{
1426 struct nfs_server *server = NFS_SERVER(inode);
1427 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1428
1429 if (ld == NULL)
Bryan Schumaker1abb50882012-06-20 15:53:47 -04001430 nfs_pageio_init_read(pgio, inode, compl_ops);
1431 else
1432 nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops, server->rsize, 0);
Trond Myklebust1751c362011-06-10 13:30:23 -04001433}
1434
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001435void
Fred Isaman061ae2e2012-04-20 14:47:48 -04001436pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode,
1437 int ioflags,
1438 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001439{
1440 struct nfs_server *server = NFS_SERVER(inode);
1441 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1442
1443 if (ld == NULL)
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001444 nfs_pageio_init_write(pgio, inode, ioflags, compl_ops);
1445 else
1446 nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops, server->wsize, ioflags);
Trond Myklebust1751c362011-06-10 13:30:23 -04001447}
1448
1449bool
Benny Halevydfed2062011-05-25 20:25:22 +03001450pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1451 struct nfs_page *req)
Fred Isamanbae724e2011-03-01 01:34:15 +00001452{
Trond Myklebustd8007d42011-06-10 13:30:23 -04001453 if (pgio->pg_lseg == NULL)
1454 return nfs_generic_pg_test(pgio, prev, req);
Benny Halevy89a58e32011-05-25 20:54:40 +03001455
Trond Myklebust19982ba2011-06-10 13:30:23 -04001456 /*
1457 * Test if a nfs_page is fully contained in the pnfs_layout_range.
1458 * Note that this test makes several assumptions:
1459 * - that the previous nfs_page in the struct nfs_pageio_descriptor
1460 * is known to lie within the range.
1461 * - that the nfs_page being tested is known to be contiguous with the
1462 * previous nfs_page.
1463 * - Layout ranges are page aligned, so we only have to test the
1464 * start offset of the request.
1465 *
1466 * Please also note that 'end_offset' is actually the offset of the
1467 * first byte that lies outside the pnfs_layout_range. FIXME?
1468 *
1469 */
1470 return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset,
1471 pgio->pg_lseg->pls_range.length);
Fred Isamanbae724e2011-03-01 01:34:15 +00001472}
Benny Halevy89a58e32011-05-25 20:54:40 +03001473EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
Fred Isamanbae724e2011-03-01 01:34:15 +00001474
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001475int pnfs_write_done_resend_to_mds(struct inode *inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001476 struct list_head *head,
Benny Halevy78f33272013-02-24 09:55:57 -05001477 const struct nfs_pgio_completion_ops *compl_ops,
1478 struct nfs_direct_req *dreq)
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001479{
1480 struct nfs_pageio_descriptor pgio;
1481 LIST_HEAD(failed);
1482
1483 /* Resend all requests through the MDS */
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001484 nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, compl_ops);
Benny Halevy78f33272013-02-24 09:55:57 -05001485 pgio.pg_dreq = dreq;
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001486 while (!list_empty(head)) {
1487 struct nfs_page *req = nfs_list_entry(head->next);
1488
1489 nfs_list_remove_request(req);
1490 if (!nfs_pageio_add_request(&pgio, req))
1491 nfs_list_add_request(req, &failed);
1492 }
1493 nfs_pageio_complete(&pgio);
1494
1495 if (!list_empty(&failed)) {
1496 /* For some reason our attempt to resend pages. Mark the
1497 * overall send request as having failed, and let
1498 * nfs_writeback_release_full deal with the error.
1499 */
1500 list_move(&failed, head);
1501 return -EIO;
1502 }
1503 return 0;
1504}
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001505EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001506
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001507static void pnfs_ld_handle_write_error(struct nfs_write_data *data)
1508{
Fred Isamancd841602012-04-20 14:47:44 -04001509 struct nfs_pgio_header *hdr = data->header;
1510
1511 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
1512 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001513 PNFS_LAYOUTRET_ON_ERROR) {
Fred Isamancd841602012-04-20 14:47:44 -04001514 pnfs_return_layout(hdr->inode);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001515 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001516 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1517 data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001518 &hdr->pages,
Benny Halevy78f33272013-02-24 09:55:57 -05001519 hdr->completion_ops,
1520 hdr->dreq);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001521}
1522
Benny Halevyd20581a2011-05-22 19:52:03 +03001523/*
1524 * Called by non rpc-based layout drivers
1525 */
Peng Tao8ce160c2011-09-22 21:50:14 -04001526void pnfs_ld_write_done(struct nfs_write_data *data)
Fred Isaman94ad1c82011-03-01 01:34:14 +00001527{
Fred Isamancd841602012-04-20 14:47:44 -04001528 struct nfs_pgio_header *hdr = data->header;
1529
Trond Myklebustcc668ab2013-08-14 15:31:28 -04001530 trace_nfs4_pnfs_write(data, hdr->pnfs_error);
Fred Isamancd841602012-04-20 14:47:44 -04001531 if (!hdr->pnfs_error) {
Benny Halevyd20581a2011-05-22 19:52:03 +03001532 pnfs_set_layoutcommit(data);
Fred Isamancd841602012-04-20 14:47:44 -04001533 hdr->mds_ops->rpc_call_done(&data->task, data);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001534 } else
1535 pnfs_ld_handle_write_error(data);
Fred Isamancd841602012-04-20 14:47:44 -04001536 hdr->mds_ops->rpc_release(data);
Fred Isaman44b83792011-03-03 15:13:44 +00001537}
Benny Halevyd20581a2011-05-22 19:52:03 +03001538EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
Fred Isaman44b83792011-03-03 15:13:44 +00001539
Trond Myklebustdce81292011-07-13 15:59:19 -04001540static void
1541pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1542 struct nfs_write_data *data)
1543{
Fred Isamancd841602012-04-20 14:47:44 -04001544 struct nfs_pgio_header *hdr = data->header;
1545
Fred Isaman6c75dc02012-04-20 14:47:47 -04001546 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1547 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1548 nfs_pageio_reset_write_mds(desc);
1549 desc->pg_recoalesce = 1;
1550 }
Trond Myklebustdce81292011-07-13 15:59:19 -04001551 nfs_writedata_release(data);
1552}
1553
1554static enum pnfs_try_status
Andy Adamson0382b742011-03-03 15:13:45 +00001555pnfs_try_to_write_data(struct nfs_write_data *wdata,
Trond Myklebustdce81292011-07-13 15:59:19 -04001556 const struct rpc_call_ops *call_ops,
1557 struct pnfs_layout_segment *lseg,
1558 int how)
Andy Adamson0382b742011-03-03 15:13:45 +00001559{
Fred Isamancd841602012-04-20 14:47:44 -04001560 struct nfs_pgio_header *hdr = wdata->header;
1561 struct inode *inode = hdr->inode;
Andy Adamson0382b742011-03-03 15:13:45 +00001562 enum pnfs_try_status trypnfs;
1563 struct nfs_server *nfss = NFS_SERVER(inode);
1564
Fred Isamancd841602012-04-20 14:47:44 -04001565 hdr->mds_ops = call_ops;
Andy Adamson0382b742011-03-03 15:13:45 +00001566
1567 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1568 inode->i_ino, wdata->args.count, wdata->args.offset, how);
Andy Adamson0382b742011-03-03 15:13:45 +00001569 trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001570 if (trypnfs != PNFS_NOT_ATTEMPTED)
Andy Adamson0382b742011-03-03 15:13:45 +00001571 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
Andy Adamson0382b742011-03-03 15:13:45 +00001572 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1573 return trypnfs;
1574}
1575
Trond Myklebustdce81292011-07-13 15:59:19 -04001576static void
1577pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how)
1578{
1579 struct nfs_write_data *data;
1580 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1581 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1582
1583 desc->pg_lseg = NULL;
1584 while (!list_empty(head)) {
1585 enum pnfs_try_status trypnfs;
1586
Fred Isaman6c75dc02012-04-20 14:47:47 -04001587 data = list_first_entry(head, struct nfs_write_data, list);
Trond Myklebustdce81292011-07-13 15:59:19 -04001588 list_del_init(&data->list);
1589
1590 trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how);
1591 if (trypnfs == PNFS_NOT_ATTEMPTED)
1592 pnfs_write_through_mds(desc, data);
1593 }
Trond Myklebust9369a432012-09-18 20:57:08 -04001594 pnfs_put_lseg(lseg);
Trond Myklebustdce81292011-07-13 15:59:19 -04001595}
1596
Fred Isaman6c75dc02012-04-20 14:47:47 -04001597static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
1598{
Trond Myklebust9369a432012-09-18 20:57:08 -04001599 pnfs_put_lseg(hdr->lseg);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001600 nfs_writehdr_free(hdr);
1601}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001602EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001603
Trond Myklebustdce81292011-07-13 15:59:19 -04001604int
1605pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1606{
Fred Isaman6c75dc02012-04-20 14:47:47 -04001607 struct nfs_write_header *whdr;
1608 struct nfs_pgio_header *hdr;
Trond Myklebustdce81292011-07-13 15:59:19 -04001609 int ret;
1610
Fred Isaman6c75dc02012-04-20 14:47:47 -04001611 whdr = nfs_writehdr_alloc();
1612 if (!whdr) {
Trond Myklebust9b5415b2012-04-27 14:31:47 -04001613 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
Trond Myklebust9369a432012-09-18 20:57:08 -04001614 pnfs_put_lseg(desc->pg_lseg);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001615 desc->pg_lseg = NULL;
1616 return -ENOMEM;
1617 }
1618 hdr = &whdr->header;
1619 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
Trond Myklebust9369a432012-09-18 20:57:08 -04001620 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001621 atomic_inc(&hdr->refcnt);
1622 ret = nfs_generic_flush(desc, hdr);
Trond Myklebustdce81292011-07-13 15:59:19 -04001623 if (ret != 0) {
Trond Myklebust9369a432012-09-18 20:57:08 -04001624 pnfs_put_lseg(desc->pg_lseg);
Trond Myklebustdce81292011-07-13 15:59:19 -04001625 desc->pg_lseg = NULL;
Fred Isaman6c75dc02012-04-20 14:47:47 -04001626 } else
1627 pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
1628 if (atomic_dec_and_test(&hdr->refcnt))
Fred Isaman061ae2e2012-04-20 14:47:48 -04001629 hdr->completion_ops->completion(hdr);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001630 return ret;
Trond Myklebustdce81292011-07-13 15:59:19 -04001631}
1632EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
1633
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001634int pnfs_read_done_resend_to_mds(struct inode *inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001635 struct list_head *head,
Benny Halevy78f33272013-02-24 09:55:57 -05001636 const struct nfs_pgio_completion_ops *compl_ops,
1637 struct nfs_direct_req *dreq)
Trond Myklebust62e4a762011-11-10 14:30:37 -05001638{
1639 struct nfs_pageio_descriptor pgio;
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001640 LIST_HEAD(failed);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001641
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001642 /* Resend all requests through the MDS */
Bryan Schumaker1abb50882012-06-20 15:53:47 -04001643 nfs_pageio_init_read(&pgio, inode, compl_ops);
Benny Halevy78f33272013-02-24 09:55:57 -05001644 pgio.pg_dreq = dreq;
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001645 while (!list_empty(head)) {
1646 struct nfs_page *req = nfs_list_entry(head->next);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001647
1648 nfs_list_remove_request(req);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001649 if (!nfs_pageio_add_request(&pgio, req))
1650 nfs_list_add_request(req, &failed);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001651 }
1652 nfs_pageio_complete(&pgio);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001653
1654 if (!list_empty(&failed)) {
1655 list_move(&failed, head);
1656 return -EIO;
1657 }
1658 return 0;
1659}
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001660EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001661
1662static void pnfs_ld_handle_read_error(struct nfs_read_data *data)
1663{
Fred Isamancd841602012-04-20 14:47:44 -04001664 struct nfs_pgio_header *hdr = data->header;
1665
1666 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
1667 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001668 PNFS_LAYOUTRET_ON_ERROR) {
Fred Isamancd841602012-04-20 14:47:44 -04001669 pnfs_return_layout(hdr->inode);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001670 }
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001671 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1672 data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001673 &hdr->pages,
Benny Halevy78f33272013-02-24 09:55:57 -05001674 hdr->completion_ops,
1675 hdr->dreq);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001676}
1677
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001678/*
Benny Halevyd20581a2011-05-22 19:52:03 +03001679 * Called by non rpc-based layout drivers
1680 */
Peng Tao9b7eecd2011-09-22 21:50:15 -04001681void pnfs_ld_read_done(struct nfs_read_data *data)
Benny Halevyd20581a2011-05-22 19:52:03 +03001682{
Fred Isamancd841602012-04-20 14:47:44 -04001683 struct nfs_pgio_header *hdr = data->header;
1684
Trond Myklebustcc668ab2013-08-14 15:31:28 -04001685 trace_nfs4_pnfs_read(data, hdr->pnfs_error);
Fred Isamancd841602012-04-20 14:47:44 -04001686 if (likely(!hdr->pnfs_error)) {
Benny Halevyd20581a2011-05-22 19:52:03 +03001687 __nfs4_read_done_cb(data);
Fred Isamancd841602012-04-20 14:47:44 -04001688 hdr->mds_ops->rpc_call_done(&data->task, data);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001689 } else
1690 pnfs_ld_handle_read_error(data);
Fred Isamancd841602012-04-20 14:47:44 -04001691 hdr->mds_ops->rpc_release(data);
Benny Halevyd20581a2011-05-22 19:52:03 +03001692}
1693EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
1694
Trond Myklebust493292d2011-07-13 15:58:28 -04001695static void
1696pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1697 struct nfs_read_data *data)
1698{
Fred Isamancd841602012-04-20 14:47:44 -04001699 struct nfs_pgio_header *hdr = data->header;
1700
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001701 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1702 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1703 nfs_pageio_reset_read_mds(desc);
1704 desc->pg_recoalesce = 1;
1705 }
Trond Myklebust493292d2011-07-13 15:58:28 -04001706 nfs_readdata_release(data);
1707}
1708
Benny Halevyd20581a2011-05-22 19:52:03 +03001709/*
Andy Adamson64419a92011-03-01 01:34:16 +00001710 * Call the appropriate parallel I/O subsystem read function.
1711 */
Trond Myklebust493292d2011-07-13 15:58:28 -04001712static enum pnfs_try_status
Andy Adamson64419a92011-03-01 01:34:16 +00001713pnfs_try_to_read_data(struct nfs_read_data *rdata,
Trond Myklebust493292d2011-07-13 15:58:28 -04001714 const struct rpc_call_ops *call_ops,
1715 struct pnfs_layout_segment *lseg)
Andy Adamson64419a92011-03-01 01:34:16 +00001716{
Fred Isamancd841602012-04-20 14:47:44 -04001717 struct nfs_pgio_header *hdr = rdata->header;
1718 struct inode *inode = hdr->inode;
Andy Adamson64419a92011-03-01 01:34:16 +00001719 struct nfs_server *nfss = NFS_SERVER(inode);
1720 enum pnfs_try_status trypnfs;
1721
Fred Isamancd841602012-04-20 14:47:44 -04001722 hdr->mds_ops = call_ops;
Andy Adamson64419a92011-03-01 01:34:16 +00001723
1724 dprintk("%s: Reading ino:%lu %u@%llu\n",
1725 __func__, inode->i_ino, rdata->args.count, rdata->args.offset);
1726
1727 trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001728 if (trypnfs != PNFS_NOT_ATTEMPTED)
Andy Adamson64419a92011-03-01 01:34:16 +00001729 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
Andy Adamson64419a92011-03-01 01:34:16 +00001730 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1731 return trypnfs;
1732}
Andy Adamson863a3c62011-03-23 13:27:54 +00001733
Trond Myklebust493292d2011-07-13 15:58:28 -04001734static void
1735pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head)
1736{
1737 struct nfs_read_data *data;
1738 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1739 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1740
1741 desc->pg_lseg = NULL;
1742 while (!list_empty(head)) {
1743 enum pnfs_try_status trypnfs;
1744
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001745 data = list_first_entry(head, struct nfs_read_data, list);
Trond Myklebust493292d2011-07-13 15:58:28 -04001746 list_del_init(&data->list);
1747
1748 trypnfs = pnfs_try_to_read_data(data, call_ops, lseg);
1749 if (trypnfs == PNFS_NOT_ATTEMPTED)
1750 pnfs_read_through_mds(desc, data);
1751 }
Trond Myklebust9369a432012-09-18 20:57:08 -04001752 pnfs_put_lseg(lseg);
Trond Myklebust493292d2011-07-13 15:58:28 -04001753}
1754
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001755static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
1756{
Trond Myklebust9369a432012-09-18 20:57:08 -04001757 pnfs_put_lseg(hdr->lseg);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001758 nfs_readhdr_free(hdr);
1759}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001760EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001761
Trond Myklebust493292d2011-07-13 15:58:28 -04001762int
1763pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
1764{
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001765 struct nfs_read_header *rhdr;
1766 struct nfs_pgio_header *hdr;
Trond Myklebust493292d2011-07-13 15:58:28 -04001767 int ret;
1768
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001769 rhdr = nfs_readhdr_alloc();
1770 if (!rhdr) {
Fred Isaman061ae2e2012-04-20 14:47:48 -04001771 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001772 ret = -ENOMEM;
Trond Myklebust9369a432012-09-18 20:57:08 -04001773 pnfs_put_lseg(desc->pg_lseg);
Trond Myklebust493292d2011-07-13 15:58:28 -04001774 desc->pg_lseg = NULL;
1775 return ret;
1776 }
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001777 hdr = &rhdr->header;
1778 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
Trond Myklebust9369a432012-09-18 20:57:08 -04001779 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001780 atomic_inc(&hdr->refcnt);
1781 ret = nfs_generic_pagein(desc, hdr);
1782 if (ret != 0) {
Trond Myklebust9369a432012-09-18 20:57:08 -04001783 pnfs_put_lseg(desc->pg_lseg);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001784 desc->pg_lseg = NULL;
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001785 } else
1786 pnfs_do_multiple_reads(desc, &hdr->rpc_list);
1787 if (atomic_dec_and_test(&hdr->refcnt))
Fred Isaman061ae2e2012-04-20 14:47:48 -04001788 hdr->completion_ops->completion(hdr);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001789 return ret;
Trond Myklebust493292d2011-07-13 15:58:28 -04001790}
1791EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
1792
Trond Myklebust71244d92014-01-13 13:34:36 -05001793static void pnfs_clear_layoutcommitting(struct inode *inode)
1794{
1795 unsigned long *bitlock = &NFS_I(inode)->flags;
1796
1797 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
1798 smp_mb__after_clear_bit();
1799 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
1800}
1801
Andy Adamson863a3c62011-03-23 13:27:54 +00001802/*
Peng Taoa9bae562011-07-30 20:52:33 -04001803 * There can be multiple RW segments.
Andy Adamson863a3c62011-03-23 13:27:54 +00001804 */
Peng Taoa9bae562011-07-30 20:52:33 -04001805static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
Andy Adamson863a3c62011-03-23 13:27:54 +00001806{
Peng Taoa9bae562011-07-30 20:52:33 -04001807 struct pnfs_layout_segment *lseg;
Andy Adamson863a3c62011-03-23 13:27:54 +00001808
Peng Taoa9bae562011-07-30 20:52:33 -04001809 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
1810 if (lseg->pls_range.iomode == IOMODE_RW &&
Trond Myklebusta073dbf2013-03-20 12:34:32 -04001811 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
Peng Taoa9bae562011-07-30 20:52:33 -04001812 list_add(&lseg->pls_lc_list, listp);
1813 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001814}
1815
Trond Myklebusta073dbf2013-03-20 12:34:32 -04001816static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
1817{
1818 struct pnfs_layout_segment *lseg, *tmp;
Trond Myklebusta073dbf2013-03-20 12:34:32 -04001819
1820 /* Matched by references in pnfs_set_layoutcommit */
1821 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
1822 list_del_init(&lseg->pls_lc_list);
1823 pnfs_put_lseg(lseg);
1824 }
1825
Trond Myklebust71244d92014-01-13 13:34:36 -05001826 pnfs_clear_layoutcommitting(inode);
Trond Myklebusta073dbf2013-03-20 12:34:32 -04001827}
1828
Peng Tao1b0ae062011-09-22 21:50:12 -04001829void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
1830{
Trond Myklebustb9e028f2012-09-18 16:41:18 -04001831 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
Peng Tao1b0ae062011-09-22 21:50:12 -04001832}
1833EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
1834
Andy Adamson863a3c62011-03-23 13:27:54 +00001835void
1836pnfs_set_layoutcommit(struct nfs_write_data *wdata)
1837{
Fred Isamancd841602012-04-20 14:47:44 -04001838 struct nfs_pgio_header *hdr = wdata->header;
1839 struct inode *inode = hdr->inode;
1840 struct nfs_inode *nfsi = NFS_I(inode);
Vitaliy Gusev4b8ee2b2011-05-20 01:34:46 +04001841 loff_t end_pos = wdata->mds_offset + wdata->res.count;
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001842 bool mark_as_dirty = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001843
Fred Isamancd841602012-04-20 14:47:44 -04001844 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00001845 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001846 mark_as_dirty = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001847 dprintk("%s: Set layoutcommit for inode %lu ",
Fred Isamancd841602012-04-20 14:47:44 -04001848 __func__, inode->i_ino);
Andy Adamson863a3c62011-03-23 13:27:54 +00001849 }
Fred Isamancd841602012-04-20 14:47:44 -04001850 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
Peng Taoa9bae562011-07-30 20:52:33 -04001851 /* references matched in nfs4_layoutcommit_release */
Trond Myklebust9369a432012-09-18 20:57:08 -04001852 pnfs_get_lseg(hdr->lseg);
Peng Taoa9bae562011-07-30 20:52:33 -04001853 }
Peng Taoacff58802011-07-30 20:52:31 -04001854 if (end_pos > nfsi->layout->plh_lwb)
1855 nfsi->layout->plh_lwb = end_pos;
Fred Isamancd841602012-04-20 14:47:44 -04001856 spin_unlock(&inode->i_lock);
Peng Taoacff58802011-07-30 20:52:31 -04001857 dprintk("%s: lseg %p end_pos %llu\n",
Fred Isamancd841602012-04-20 14:47:44 -04001858 __func__, hdr->lseg, nfsi->layout->plh_lwb);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001859
1860 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1861 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1862 if (mark_as_dirty)
Fred Isamancd841602012-04-20 14:47:44 -04001863 mark_inode_dirty_sync(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001864}
1865EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
1866
Andy Adamsondb29c082011-07-30 20:52:38 -04001867void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
1868{
1869 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
1870
1871 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
1872 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
Trond Myklebusta073dbf2013-03-20 12:34:32 -04001873 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
Andy Adamsondb29c082011-07-30 20:52:38 -04001874}
1875
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001876/*
1877 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1878 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1879 * data to disk to allow the server to recover the data if it crashes.
1880 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1881 * is off, and a COMMIT is sent to a data server, or
1882 * if WRITEs to a data server return NFS_DATA_SYNC.
1883 */
Andy Adamson863a3c62011-03-23 13:27:54 +00001884int
Andy Adamsonef311532011-03-12 02:58:10 -05001885pnfs_layoutcommit_inode(struct inode *inode, bool sync)
Andy Adamson863a3c62011-03-23 13:27:54 +00001886{
1887 struct nfs4_layoutcommit_data *data;
1888 struct nfs_inode *nfsi = NFS_I(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001889 loff_t end_pos;
Trond Myklebust71244d92014-01-13 13:34:36 -05001890 int status;
1891
1892 if (!pnfs_layoutcommit_outstanding(inode))
1893 return 0;
Andy Adamson863a3c62011-03-23 13:27:54 +00001894
1895 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
1896
Trond Myklebust71244d92014-01-13 13:34:36 -05001897 status = -EAGAIN;
1898 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
1899 if (!sync)
1900 goto out;
1901 status = wait_on_bit_lock(&nfsi->flags,
1902 NFS_INO_LAYOUTCOMMITTING,
1903 nfs_wait_bit_killable,
1904 TASK_KILLABLE);
1905 if (status)
1906 goto out;
1907 }
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001908
Trond Myklebust71244d92014-01-13 13:34:36 -05001909 status = -ENOMEM;
Andy Adamson863a3c62011-03-23 13:27:54 +00001910 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1911 data = kzalloc(sizeof(*data), GFP_NOFS);
Trond Myklebust71244d92014-01-13 13:34:36 -05001912 if (!data)
1913 goto clear_layoutcommitting;
Andy Adamson863a3c62011-03-23 13:27:54 +00001914
Trond Myklebust71244d92014-01-13 13:34:36 -05001915 status = 0;
1916 spin_lock(&inode->i_lock);
1917 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1918 goto out_unlock;
Peng Tao92407e72011-10-23 20:21:17 -07001919
Peng Taoa9bae562011-07-30 20:52:33 -04001920 INIT_LIST_HEAD(&data->lseg_list);
Peng Taoa9bae562011-07-30 20:52:33 -04001921 pnfs_list_write_lseg(inode, &data->lseg_list);
Andy Adamson863a3c62011-03-23 13:27:54 +00001922
Peng Taoacff58802011-07-30 20:52:31 -04001923 end_pos = nfsi->layout->plh_lwb;
Peng Taoacff58802011-07-30 20:52:31 -04001924 nfsi->layout->plh_lwb = 0;
Andy Adamson863a3c62011-03-23 13:27:54 +00001925
Trond Myklebustf597c532012-03-04 18:13:56 -05001926 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
Andy Adamson863a3c62011-03-23 13:27:54 +00001927 spin_unlock(&inode->i_lock);
1928
1929 data->args.inode = inode;
Peng Tao9fa40752011-07-30 20:52:32 -04001930 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
Andy Adamson863a3c62011-03-23 13:27:54 +00001931 nfs_fattr_init(&data->fattr);
1932 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
1933 data->res.fattr = &data->fattr;
1934 data->args.lastbytewritten = end_pos - 1;
1935 data->res.server = NFS_SERVER(inode);
1936
1937 status = nfs4_proc_layoutcommit(data, sync);
1938out:
Peng Tao92407e72011-10-23 20:21:17 -07001939 if (status)
1940 mark_inode_dirty_sync(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001941 dprintk("<-- %s status %d\n", __func__, status);
1942 return status;
Trond Myklebust71244d92014-01-13 13:34:36 -05001943out_unlock:
1944 spin_unlock(&inode->i_lock);
Peng Tao92407e72011-10-23 20:21:17 -07001945 kfree(data);
Trond Myklebust71244d92014-01-13 13:34:36 -05001946clear_layoutcommitting:
1947 pnfs_clear_layoutcommitting(inode);
Peng Tao92407e72011-10-23 20:21:17 -07001948 goto out;
Andy Adamson863a3c62011-03-23 13:27:54 +00001949}
Andy Adamson82be4172012-05-23 05:02:35 -04001950
1951struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
1952{
1953 struct nfs4_threshold *thp;
1954
1955 thp = kzalloc(sizeof(*thp), GFP_NOFS);
1956 if (!thp) {
1957 dprintk("%s mdsthreshold allocation failed\n", __func__);
1958 return NULL;
1959 }
1960 return thp;
1961}