blob: 26b96de831ea71adcd5dcd99fb7ab63468a9ebdb [file] [log] [blame]
Dean Hildebrand7ab672c2010-10-20 00:18:00 -04001/*
2 * Module for the pnfs nfs4 file layout driver.
3 * Defines all I/O and Policy interface operations, plus code
4 * to register itself with the pNFS client.
5 *
6 * Copyright (c) 2002
7 * The Regents of the University of Michigan
8 * All Rights Reserved
9 *
10 * Dean Hildebrand <dhildebz@umich.edu>
11 *
12 * Permission is granted to use, copy, create derivative works, and
13 * redistribute this software and such derivative works for any purpose,
14 * so long as the name of the University of Michigan is not used in
15 * any advertising or publicity pertaining to the use or distribution
16 * of this software without specific, written prior authorization. If
17 * the above copyright notice or any other identification of the
18 * University of Michigan is included in any copy of any portion of
19 * this software, then the disclaimer below must also be included.
20 *
21 * This software is provided as is, without representation or warranty
22 * of any kind either express or implied, including without limitation
23 * the implied warranties of merchantability, fitness for a particular
24 * purpose, or noninfringement. The Regents of the University of
25 * Michigan shall not be liable for any damages, including special,
26 * indirect, incidental, or consequential damages, with respect to any
27 * claim arising out of or in connection with the use of the software,
28 * even if it has been or is hereafter advised of the possibility of
29 * such damages.
30 */
31
32#include <linux/nfs_fs.h>
Benny Halevy19345cb2011-06-19 18:33:46 -040033#include <linux/nfs_page.h>
Paul Gortmaker143cb492011-07-01 14:23:34 -040034#include <linux/module.h>
Andy Adamson16b374c2010-10-20 00:18:04 -040035
Weston Andros Adamson0a702192012-02-17 13:15:24 -050036#include <linux/sunrpc/metrics.h>
37
Andy Adamson16b374c2010-10-20 00:18:04 -040038#include "internal.h"
Andy Adamson9cb81962012-03-07 10:49:41 -050039#include "delegation.h"
Andy Adamson16b374c2010-10-20 00:18:04 -040040#include "nfs4filelayout.h"
Dean Hildebrand7ab672c2010-10-20 00:18:00 -040041
42#define NFSDBG_FACILITY NFSDBG_PNFS_LD
43
44MODULE_LICENSE("GPL");
45MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
46MODULE_DESCRIPTION("The NFSv4 file layout driver");
47
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +000048#define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
49
Fred Isamancfe7f412011-03-01 01:34:18 +000050static loff_t
51filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
52 loff_t offset)
53{
54 u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
Chris Metcalf3476f112011-08-11 13:54:28 -070055 u64 stripe_no;
56 u32 rem;
Fred Isamancfe7f412011-03-01 01:34:18 +000057
58 offset -= flseg->pattern_offset;
Chris Metcalf3476f112011-08-11 13:54:28 -070059 stripe_no = div_u64(offset, stripe_width);
60 div_u64_rem(offset, flseg->stripe_unit, &rem);
Fred Isamancfe7f412011-03-01 01:34:18 +000061
Chris Metcalf3476f112011-08-11 13:54:28 -070062 return stripe_no * flseg->stripe_unit + rem;
Fred Isamancfe7f412011-03-01 01:34:18 +000063}
64
65/* This function is used by the layout driver to calculate the
66 * offset of the file on the dserver based on whether the
67 * layout type is STRIPE_DENSE or STRIPE_SPARSE
68 */
69static loff_t
70filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
71{
72 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
73
74 switch (flseg->stripe_type) {
75 case STRIPE_SPARSE:
76 return offset;
77
78 case STRIPE_DENSE:
79 return filelayout_get_dense_offset(flseg, offset);
80 }
81
82 BUG();
83}
84
Andy Adamsone7dd79af2012-04-27 17:53:46 -040085static void filelayout_reset_write(struct nfs_write_data *data)
86{
87 struct nfs_pgio_header *hdr = data->header;
Andy Adamsone7dd79af2012-04-27 17:53:46 -040088 struct rpc_task *task = &data->task;
89
90 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
91 dprintk("%s Reset task %5u for i/o through MDS "
92 "(req %s/%lld, %u bytes @ offset %llu)\n", __func__,
93 data->task.tk_pid,
Bryan Schumaker497826a2012-05-22 10:10:03 -040094 hdr->inode->i_sb->s_id,
95 (long long)NFS_FILEID(hdr->inode),
Andy Adamsone7dd79af2012-04-27 17:53:46 -040096 data->args.count,
97 (unsigned long long)data->args.offset);
98
99 task->tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
100 &hdr->pages,
101 hdr->completion_ops);
102 }
103}
104
105static void filelayout_reset_read(struct nfs_read_data *data)
106{
107 struct nfs_pgio_header *hdr = data->header;
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400108 struct rpc_task *task = &data->task;
109
110 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
111 dprintk("%s Reset task %5u for i/o through MDS "
112 "(req %s/%lld, %u bytes @ offset %llu)\n", __func__,
113 data->task.tk_pid,
Bryan Schumaker497826a2012-05-22 10:10:03 -0400114 hdr->inode->i_sb->s_id,
115 (long long)NFS_FILEID(hdr->inode),
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400116 data->args.count,
117 (unsigned long long)data->args.offset);
118
119 task->tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
120 &hdr->pages,
121 hdr->completion_ops);
122 }
123}
124
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000125static int filelayout_async_handle_error(struct rpc_task *task,
126 struct nfs4_state *state,
127 struct nfs_client *clp,
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400128 struct pnfs_layout_segment *lseg)
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000129{
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400130 struct inode *inode = lseg->pls_layout->plh_inode;
131 struct nfs_server *mds_server = NFS_SERVER(inode);
132 struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
Andy Adamson9cb81962012-03-07 10:49:41 -0500133 struct nfs_client *mds_client = mds_server->nfs_client;
Andy Adamson671fb892012-04-27 17:53:49 -0400134 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
Andy Adamson9cb81962012-03-07 10:49:41 -0500135
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000136 if (task->tk_status >= 0)
137 return 0;
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000138
139 switch (task->tk_status) {
Andy Adamson9cb81962012-03-07 10:49:41 -0500140 /* MDS state errors */
141 case -NFS4ERR_DELEG_REVOKED:
142 case -NFS4ERR_ADMIN_REVOKED:
143 case -NFS4ERR_BAD_STATEID:
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400144 if (state == NULL)
145 break;
Andy Adamson2dc31752012-03-08 11:03:53 -0500146 nfs_remove_bad_delegation(state->inode);
Andy Adamson9cb81962012-03-07 10:49:41 -0500147 case -NFS4ERR_OPENMODE:
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400148 if (state == NULL)
149 break;
Andy Adamson9cb81962012-03-07 10:49:41 -0500150 nfs4_schedule_stateid_recovery(mds_server, state);
151 goto wait_on_recovery;
152 case -NFS4ERR_EXPIRED:
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400153 if (state != NULL)
154 nfs4_schedule_stateid_recovery(mds_server, state);
Andy Adamson9cb81962012-03-07 10:49:41 -0500155 nfs4_schedule_lease_recovery(mds_client);
156 goto wait_on_recovery;
157 /* DS session errors */
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000158 case -NFS4ERR_BADSESSION:
159 case -NFS4ERR_BADSLOT:
160 case -NFS4ERR_BAD_HIGH_SLOT:
161 case -NFS4ERR_DEADSESSION:
162 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
163 case -NFS4ERR_SEQ_FALSE_RETRY:
164 case -NFS4ERR_SEQ_MISORDERED:
165 dprintk("%s ERROR %d, Reset session. Exchangeid "
166 "flags 0x%x\n", __func__, task->tk_status,
167 clp->cl_exchange_flags);
Trond Myklebust9f594792012-05-27 13:02:53 -0400168 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000169 break;
170 case -NFS4ERR_DELAY:
171 case -NFS4ERR_GRACE:
172 case -EKEYEXPIRED:
173 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
174 break;
Andy Adamsona8a4ae32011-05-03 13:43:03 -0400175 case -NFS4ERR_RETRY_UNCACHED_REP:
176 break;
Andy Adamson041245c2012-04-27 17:53:53 -0400177 /* Invalidate Layout errors */
178 case -NFS4ERR_PNFS_NO_LAYOUT:
179 case -ESTALE: /* mapped NFS4ERR_STALE */
180 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
181 case -EISDIR: /* mapped NFS4ERR_ISDIR */
182 case -NFS4ERR_FHEXPIRED:
183 case -NFS4ERR_WRONG_TYPE:
184 dprintk("%s Invalid layout error %d\n", __func__,
185 task->tk_status);
186 /*
187 * Destroy layout so new i/o will get a new layout.
188 * Layout will not be destroyed until all current lseg
189 * references are put. Mark layout as invalid to resend failed
190 * i/o and all i/o waiting on the slot table to the MDS until
191 * layout is destroyed and a new valid layout is obtained.
192 */
193 set_bit(NFS_LAYOUT_INVALID,
Andy Adamsond42e7872012-05-22 08:09:28 -0400194 &NFS_I(inode)->layout->plh_flags);
195 pnfs_destroy_layout(NFS_I(inode));
Andy Adamson041245c2012-04-27 17:53:53 -0400196 rpc_wake_up(&tbl->slot_tbl_waitq);
197 goto reset;
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400198 /* RPC connection errors */
199 case -ECONNREFUSED:
200 case -EHOSTDOWN:
201 case -EHOSTUNREACH:
202 case -ENETUNREACH:
203 case -EIO:
204 case -ETIMEDOUT:
205 case -EPIPE:
206 dprintk("%s DS connection error %d\n", __func__,
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000207 task->tk_status);
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400208 filelayout_mark_devid_invalid(devid);
Andy Adamson82c7c7a2012-06-20 15:03:31 -0400209 _pnfs_return_layout(inode);
Andy Adamson671fb892012-04-27 17:53:49 -0400210 rpc_wake_up(&tbl->slot_tbl_waitq);
Andy Adamsonb4a29672012-04-27 17:53:52 -0400211 nfs4_ds_disconnect(clp);
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400212 /* fall through */
213 default:
Andy Adamson041245c2012-04-27 17:53:53 -0400214reset:
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400215 dprintk("%s Retry through MDS. Error %d\n", __func__,
216 task->tk_status);
217 return -NFS4ERR_RESET_TO_MDS;
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000218 }
Andy Adamson9cb81962012-03-07 10:49:41 -0500219out:
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000220 task->tk_status = 0;
221 return -EAGAIN;
Andy Adamson9cb81962012-03-07 10:49:41 -0500222wait_on_recovery:
223 rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
224 if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
225 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
226 goto out;
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000227}
228
229/* NFS_PROTO call done callback routines */
230
231static int filelayout_read_done_cb(struct rpc_task *task,
232 struct nfs_read_data *data)
233{
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400234 struct nfs_pgio_header *hdr = data->header;
235 int err;
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000236
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400237 err = filelayout_async_handle_error(task, data->args.context->state,
238 data->ds_clp, hdr->lseg);
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000239
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400240 switch (err) {
241 case -NFS4ERR_RESET_TO_MDS:
242 filelayout_reset_read(data);
243 return task->tk_status;
244 case -EAGAIN:
Trond Myklebustd00c5d42011-10-19 12:17:29 -0700245 rpc_restart_call_prepare(task);
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000246 return -EAGAIN;
247 }
248
249 return 0;
250}
251
Andy Adamson16b374c2010-10-20 00:18:04 -0400252/*
Andy Adamson863a3c62011-03-23 13:27:54 +0000253 * We reference the rpc_cred of the first WRITE that triggers the need for
254 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
255 * rfc5661 is not clear about which credential should be used.
256 */
257static void
258filelayout_set_layoutcommit(struct nfs_write_data *wdata)
259{
Fred Isamancd841602012-04-20 14:47:44 -0400260 struct nfs_pgio_header *hdr = wdata->header;
261
262 if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds ||
Andy Adamson863a3c62011-03-23 13:27:54 +0000263 wdata->res.verf->committed == NFS_FILE_SYNC)
264 return;
265
266 pnfs_set_layoutcommit(wdata);
Fred Isamancd841602012-04-20 14:47:44 -0400267 dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
268 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
Andy Adamson863a3c62011-03-23 13:27:54 +0000269}
270
271/*
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000272 * Call ops for the async read/write cases
273 * In the case of dense layouts, the offset needs to be reset to its
274 * original value.
275 */
276static void filelayout_read_prepare(struct rpc_task *task, void *data)
277{
Fred Isamancd12ae32012-04-20 14:47:42 -0400278 struct nfs_read_data *rdata = data;
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000279
Andy Adamson041245c2012-04-27 17:53:53 -0400280 if (filelayout_reset_to_mds(rdata->header->lseg)) {
Andy Adamson0ad2f372012-04-27 17:53:48 -0400281 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
282 filelayout_reset_read(rdata);
283 rpc_exit(task, 0);
284 return;
285 }
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000286 rdata->read_done_cb = filelayout_read_done_cb;
287
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000288 if (nfs41_setup_sequence(rdata->ds_clp->cl_session,
289 &rdata->args.seq_args, &rdata->res.seq_res,
Trond Myklebust9d12b212012-01-17 22:04:25 -0500290 task))
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000291 return;
292
293 rpc_call_start(task);
294}
295
296static void filelayout_read_call_done(struct rpc_task *task, void *data)
297{
Fred Isamancd12ae32012-04-20 14:47:42 -0400298 struct nfs_read_data *rdata = data;
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000299
300 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
301
Andy Adamsonbd4aeff2012-05-22 08:09:27 -0400302 if (test_bit(NFS_IOHDR_REDO, &rdata->header->flags) &&
303 task->tk_status == 0)
Andy Adamson0ad2f372012-04-27 17:53:48 -0400304 return;
305
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000306 /* Note this may cause RPC to be resent */
Fred Isamancd841602012-04-20 14:47:44 -0400307 rdata->header->mds_ops->rpc_call_done(task, data);
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000308}
309
Weston Andros Adamson0a702192012-02-17 13:15:24 -0500310static void filelayout_read_count_stats(struct rpc_task *task, void *data)
311{
Fred Isamancd12ae32012-04-20 14:47:42 -0400312 struct nfs_read_data *rdata = data;
Weston Andros Adamson0a702192012-02-17 13:15:24 -0500313
Fred Isamancd841602012-04-20 14:47:44 -0400314 rpc_count_iostats(task, NFS_SERVER(rdata->header->inode)->client->cl_metrics);
Weston Andros Adamson0a702192012-02-17 13:15:24 -0500315}
316
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000317static void filelayout_read_release(void *data)
318{
Fred Isamancd12ae32012-04-20 14:47:42 -0400319 struct nfs_read_data *rdata = data;
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000320
Andy Adamson996074c2012-05-22 08:09:26 -0400321 nfs_put_client(rdata->ds_clp);
Fred Isamancd841602012-04-20 14:47:44 -0400322 rdata->header->mds_ops->rpc_release(data);
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000323}
324
Fred Isamana69aef12011-03-03 15:13:47 +0000325static int filelayout_write_done_cb(struct rpc_task *task,
326 struct nfs_write_data *data)
327{
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400328 struct nfs_pgio_header *hdr = data->header;
329 int err;
Fred Isamana69aef12011-03-03 15:13:47 +0000330
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400331 err = filelayout_async_handle_error(task, data->args.context->state,
332 data->ds_clp, hdr->lseg);
333
334 switch (err) {
335 case -NFS4ERR_RESET_TO_MDS:
336 filelayout_reset_write(data);
337 return task->tk_status;
338 case -EAGAIN:
Trond Myklebustd00c5d42011-10-19 12:17:29 -0700339 rpc_restart_call_prepare(task);
Fred Isamana69aef12011-03-03 15:13:47 +0000340 return -EAGAIN;
341 }
342
Andy Adamson863a3c62011-03-23 13:27:54 +0000343 filelayout_set_layoutcommit(data);
Fred Isamana69aef12011-03-03 15:13:47 +0000344 return 0;
345}
346
Fred Isamane0c2b382011-03-23 13:27:53 +0000347/* Fake up some data that will cause nfs_commit_release to retry the writes. */
Fred Isaman0b7c0152012-04-20 14:47:39 -0400348static void prepare_to_resend_writes(struct nfs_commit_data *data)
Fred Isamane0c2b382011-03-23 13:27:53 +0000349{
350 struct nfs_page *first = nfs_list_entry(data->pages.next);
351
352 data->task.tk_status = 0;
Trond Myklebust2f2c63b2012-06-08 11:56:09 -0400353 memcpy(&data->verf.verifier, &first->wb_verf,
354 sizeof(data->verf.verifier));
355 data->verf.verifier.data[0]++; /* ensure verifier mismatch */
Fred Isamane0c2b382011-03-23 13:27:53 +0000356}
357
358static int filelayout_commit_done_cb(struct rpc_task *task,
Fred Isaman0b7c0152012-04-20 14:47:39 -0400359 struct nfs_commit_data *data)
Fred Isamane0c2b382011-03-23 13:27:53 +0000360{
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400361 int err;
Fred Isamane0c2b382011-03-23 13:27:53 +0000362
Andy Adamsone7dd79af2012-04-27 17:53:46 -0400363 err = filelayout_async_handle_error(task, NULL, data->ds_clp,
364 data->lseg);
365
366 switch (err) {
367 case -NFS4ERR_RESET_TO_MDS:
368 prepare_to_resend_writes(data);
369 return -EAGAIN;
370 case -EAGAIN:
371 rpc_restart_call_prepare(task);
Fred Isamane0c2b382011-03-23 13:27:53 +0000372 return -EAGAIN;
373 }
374
375 return 0;
376}
377
Fred Isamana69aef12011-03-03 15:13:47 +0000378static void filelayout_write_prepare(struct rpc_task *task, void *data)
379{
Fred Isamancd12ae32012-04-20 14:47:42 -0400380 struct nfs_write_data *wdata = data;
Fred Isamana69aef12011-03-03 15:13:47 +0000381
Andy Adamson041245c2012-04-27 17:53:53 -0400382 if (filelayout_reset_to_mds(wdata->header->lseg)) {
Andy Adamson0ad2f372012-04-27 17:53:48 -0400383 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
384 filelayout_reset_write(wdata);
385 rpc_exit(task, 0);
386 return;
387 }
Fred Isamana69aef12011-03-03 15:13:47 +0000388 if (nfs41_setup_sequence(wdata->ds_clp->cl_session,
389 &wdata->args.seq_args, &wdata->res.seq_res,
Trond Myklebust9d12b212012-01-17 22:04:25 -0500390 task))
Fred Isamana69aef12011-03-03 15:13:47 +0000391 return;
392
393 rpc_call_start(task);
394}
395
396static void filelayout_write_call_done(struct rpc_task *task, void *data)
397{
Fred Isamancd12ae32012-04-20 14:47:42 -0400398 struct nfs_write_data *wdata = data;
Fred Isamana69aef12011-03-03 15:13:47 +0000399
Andy Adamsonbd4aeff2012-05-22 08:09:27 -0400400 if (test_bit(NFS_IOHDR_REDO, &wdata->header->flags) &&
401 task->tk_status == 0)
Andy Adamson0ad2f372012-04-27 17:53:48 -0400402 return;
403
Fred Isamana69aef12011-03-03 15:13:47 +0000404 /* Note this may cause RPC to be resent */
Fred Isamancd841602012-04-20 14:47:44 -0400405 wdata->header->mds_ops->rpc_call_done(task, data);
Fred Isamana69aef12011-03-03 15:13:47 +0000406}
407
Weston Andros Adamson0a702192012-02-17 13:15:24 -0500408static void filelayout_write_count_stats(struct rpc_task *task, void *data)
409{
Fred Isamancd12ae32012-04-20 14:47:42 -0400410 struct nfs_write_data *wdata = data;
Weston Andros Adamson0a702192012-02-17 13:15:24 -0500411
Fred Isamancd841602012-04-20 14:47:44 -0400412 rpc_count_iostats(task, NFS_SERVER(wdata->header->inode)->client->cl_metrics);
Weston Andros Adamson0a702192012-02-17 13:15:24 -0500413}
414
Fred Isamana69aef12011-03-03 15:13:47 +0000415static void filelayout_write_release(void *data)
416{
Fred Isamancd12ae32012-04-20 14:47:42 -0400417 struct nfs_write_data *wdata = data;
Fred Isamana69aef12011-03-03 15:13:47 +0000418
Andy Adamson996074c2012-05-22 08:09:26 -0400419 nfs_put_client(wdata->ds_clp);
Fred Isamancd841602012-04-20 14:47:44 -0400420 wdata->header->mds_ops->rpc_release(data);
Fred Isamana69aef12011-03-03 15:13:47 +0000421}
422
Fred Isaman0b7c0152012-04-20 14:47:39 -0400423static void filelayout_commit_prepare(struct rpc_task *task, void *data)
Fred Isamane0c2b382011-03-23 13:27:53 +0000424{
Fred Isaman0b7c0152012-04-20 14:47:39 -0400425 struct nfs_commit_data *wdata = data;
Fred Isamane0c2b382011-03-23 13:27:53 +0000426
Fred Isaman0b7c0152012-04-20 14:47:39 -0400427 if (nfs41_setup_sequence(wdata->ds_clp->cl_session,
428 &wdata->args.seq_args, &wdata->res.seq_res,
429 task))
430 return;
431
432 rpc_call_start(task);
433}
434
435static void filelayout_write_commit_done(struct rpc_task *task, void *data)
436{
437 struct nfs_commit_data *wdata = data;
438
439 /* Note this may cause RPC to be resent */
440 wdata->mds_ops->rpc_call_done(task, data);
441}
442
443static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
444{
445 struct nfs_commit_data *cdata = data;
446
447 rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics);
448}
449
450static void filelayout_commit_release(void *calldata)
451{
452 struct nfs_commit_data *data = calldata;
453
Fred Isamanf453a542012-04-20 14:47:54 -0400454 data->completion_ops->completion(data);
Fred Isaman0b7c0152012-04-20 14:47:39 -0400455 put_lseg(data->lseg);
Andy Adamson3a7936c2012-04-27 17:53:51 -0400456 nfs_put_client(data->ds_clp);
Fred Isaman0b7c0152012-04-20 14:47:39 -0400457 nfs_commitdata_release(data);
Fred Isamane0c2b382011-03-23 13:27:53 +0000458}
459
Trond Myklebust17280172012-03-11 13:11:00 -0400460static const struct rpc_call_ops filelayout_read_call_ops = {
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000461 .rpc_call_prepare = filelayout_read_prepare,
462 .rpc_call_done = filelayout_read_call_done,
Weston Andros Adamson0a702192012-02-17 13:15:24 -0500463 .rpc_count_stats = filelayout_read_count_stats,
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000464 .rpc_release = filelayout_read_release,
465};
466
Trond Myklebust17280172012-03-11 13:11:00 -0400467static const struct rpc_call_ops filelayout_write_call_ops = {
Fred Isamana69aef12011-03-03 15:13:47 +0000468 .rpc_call_prepare = filelayout_write_prepare,
469 .rpc_call_done = filelayout_write_call_done,
Weston Andros Adamson0a702192012-02-17 13:15:24 -0500470 .rpc_count_stats = filelayout_write_count_stats,
Fred Isamana69aef12011-03-03 15:13:47 +0000471 .rpc_release = filelayout_write_release,
472};
473
Trond Myklebust17280172012-03-11 13:11:00 -0400474static const struct rpc_call_ops filelayout_commit_call_ops = {
Fred Isaman0b7c0152012-04-20 14:47:39 -0400475 .rpc_call_prepare = filelayout_commit_prepare,
476 .rpc_call_done = filelayout_write_commit_done,
477 .rpc_count_stats = filelayout_commit_count_stats,
Fred Isamane0c2b382011-03-23 13:27:53 +0000478 .rpc_release = filelayout_commit_release,
479};
480
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000481static enum pnfs_try_status
482filelayout_read_pagelist(struct nfs_read_data *data)
483{
Fred Isamancd841602012-04-20 14:47:44 -0400484 struct nfs_pgio_header *hdr = data->header;
485 struct pnfs_layout_segment *lseg = hdr->lseg;
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000486 struct nfs4_pnfs_ds *ds;
487 loff_t offset = data->args.offset;
488 u32 j, idx;
489 struct nfs_fh *fh;
490 int status;
491
492 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
Fred Isamancd841602012-04-20 14:47:44 -0400493 __func__, hdr->inode->i_ino,
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000494 data->args.pgbase, (size_t)data->args.count, offset);
495
496 /* Retrieve the correct rpc_client for the byte range */
497 j = nfs4_fl_calc_j_index(lseg, offset);
498 idx = nfs4_fl_calc_ds_index(lseg, j);
499 ds = nfs4_fl_prepare_ds(lseg, idx);
Andy Adamson90fecfc2012-04-27 17:53:43 -0400500 if (!ds)
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000501 return PNFS_NOT_ATTEMPTED;
Andy Adamson3a7936c2012-04-27 17:53:51 -0400502 dprintk("%s USE DS: %s cl_count %d\n", __func__,
503 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000504
505 /* No multipath support. Use first DS */
Andy Adamson3a7936c2012-04-27 17:53:51 -0400506 atomic_inc(&ds->ds_clp->cl_count);
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000507 data->ds_clp = ds->ds_clp;
508 fh = nfs4_fl_select_ds_fh(lseg, j);
509 if (fh)
510 data->args.fh = fh;
511
512 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
513 data->mds_offset = offset;
514
515 /* Perform an asynchronous read to ds */
Fred Isamanc5996c42012-04-20 14:47:41 -0400516 status = nfs_initiate_read(ds->ds_clp->cl_rpcclient, data,
Andy Adamson9f0ec1762012-04-27 17:53:44 -0400517 &filelayout_read_call_ops, RPC_TASK_SOFTCONN);
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000518 BUG_ON(status != 0);
519 return PNFS_ATTEMPTED;
520}
521
Fred Isamana69aef12011-03-03 15:13:47 +0000522/* Perform async writes. */
Andy Adamson0382b742011-03-03 15:13:45 +0000523static enum pnfs_try_status
524filelayout_write_pagelist(struct nfs_write_data *data, int sync)
525{
Fred Isamancd841602012-04-20 14:47:44 -0400526 struct nfs_pgio_header *hdr = data->header;
527 struct pnfs_layout_segment *lseg = hdr->lseg;
Fred Isamana69aef12011-03-03 15:13:47 +0000528 struct nfs4_pnfs_ds *ds;
529 loff_t offset = data->args.offset;
530 u32 j, idx;
531 struct nfs_fh *fh;
532 int status;
533
534 /* Retrieve the correct rpc_client for the byte range */
535 j = nfs4_fl_calc_j_index(lseg, offset);
536 idx = nfs4_fl_calc_ds_index(lseg, j);
537 ds = nfs4_fl_prepare_ds(lseg, idx);
Andy Adamson90fecfc2012-04-27 17:53:43 -0400538 if (!ds)
Fred Isamana69aef12011-03-03 15:13:47 +0000539 return PNFS_NOT_ATTEMPTED;
Andy Adamson3a7936c2012-04-27 17:53:51 -0400540 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d\n",
541 __func__, hdr->inode->i_ino, sync, (size_t) data->args.count,
542 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
Fred Isamana69aef12011-03-03 15:13:47 +0000543
Fred Isamana69aef12011-03-03 15:13:47 +0000544 data->write_done_cb = filelayout_write_done_cb;
Andy Adamson3a7936c2012-04-27 17:53:51 -0400545 atomic_inc(&ds->ds_clp->cl_count);
Fred Isamana69aef12011-03-03 15:13:47 +0000546 data->ds_clp = ds->ds_clp;
547 fh = nfs4_fl_select_ds_fh(lseg, j);
548 if (fh)
549 data->args.fh = fh;
550 /*
551 * Get the file offset on the dserver. Set the write offset to
552 * this offset and save the original offset.
553 */
554 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
Fred Isamana69aef12011-03-03 15:13:47 +0000555
556 /* Perform an asynchronous write */
Fred Isamanc5996c42012-04-20 14:47:41 -0400557 status = nfs_initiate_write(ds->ds_clp->cl_rpcclient, data,
Andy Adamson9f0ec1762012-04-27 17:53:44 -0400558 &filelayout_write_call_ops, sync,
559 RPC_TASK_SOFTCONN);
Fred Isamana69aef12011-03-03 15:13:47 +0000560 BUG_ON(status != 0);
561 return PNFS_ATTEMPTED;
Andy Adamson0382b742011-03-03 15:13:45 +0000562}
563
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000564/*
Andy Adamson16b374c2010-10-20 00:18:04 -0400565 * filelayout_check_layout()
566 *
567 * Make sure layout segment parameters are sane WRT the device.
568 * At this point no generic layer initialization of the lseg has occurred,
569 * and nothing has been added to the layout_hdr cache.
570 *
571 */
572static int
573filelayout_check_layout(struct pnfs_layout_hdr *lo,
574 struct nfs4_filelayout_segment *fl,
575 struct nfs4_layoutget_res *lgr,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400576 struct nfs4_deviceid *id,
577 gfp_t gfp_flags)
Andy Adamson16b374c2010-10-20 00:18:04 -0400578{
Benny Halevya1eaecb2011-05-19 22:14:47 -0400579 struct nfs4_deviceid_node *d;
Andy Adamson16b374c2010-10-20 00:18:04 -0400580 struct nfs4_file_layout_dsaddr *dsaddr;
581 int status = -EINVAL;
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000582 struct nfs_server *nfss = NFS_SERVER(lo->plh_inode);
Andy Adamson16b374c2010-10-20 00:18:04 -0400583
584 dprintk("--> %s\n", __func__);
585
Andy Adamson7c24d942011-06-13 18:22:38 -0400586 /* FIXME: remove this check when layout segment support is added */
587 if (lgr->range.offset != 0 ||
588 lgr->range.length != NFS4_MAX_UINT64) {
589 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
590 __func__);
591 goto out;
592 }
593
Andy Adamson16b374c2010-10-20 00:18:04 -0400594 if (fl->pattern_offset > lgr->range.offset) {
Jim Rees3b6445a2011-02-22 19:31:57 -0500595 dprintk("%s pattern_offset %lld too large\n",
Andy Adamson16b374c2010-10-20 00:18:04 -0400596 __func__, fl->pattern_offset);
597 goto out;
598 }
599
Benny Halevy75247af2011-02-22 15:56:01 -0800600 if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) {
601 dprintk("%s Invalid stripe unit (%u)\n",
Andy Adamson16b374c2010-10-20 00:18:04 -0400602 __func__, fl->stripe_unit);
603 goto out;
604 }
605
606 /* find and reference the deviceid */
Benny Halevy35c8bb52011-05-24 18:04:02 +0300607 d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld,
608 NFS_SERVER(lo->plh_inode)->nfs_client, id);
Benny Halevya1eaecb2011-05-19 22:14:47 -0400609 if (d == NULL) {
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400610 dsaddr = get_device_info(lo->plh_inode, id, gfp_flags);
Andy Adamson16b374c2010-10-20 00:18:04 -0400611 if (dsaddr == NULL)
612 goto out;
Benny Halevya1eaecb2011-05-19 22:14:47 -0400613 } else
614 dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
Andy Adamsonc47abcf2011-06-15 17:52:40 -0400615 /* Found deviceid is being reaped */
616 if (test_bit(NFS_DEVICEID_INVALID, &dsaddr->id_node.flags))
617 goto out_put;
618
Andy Adamson16b374c2010-10-20 00:18:04 -0400619 fl->dsaddr = dsaddr;
620
Chuck Levere4149662011-10-25 12:18:03 -0400621 if (fl->first_stripe_index >= dsaddr->stripe_count) {
622 dprintk("%s Bad first_stripe_index %u\n",
Andy Adamson16b374c2010-10-20 00:18:04 -0400623 __func__, fl->first_stripe_index);
624 goto out_put;
625 }
626
627 if ((fl->stripe_type == STRIPE_SPARSE &&
628 fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
629 (fl->stripe_type == STRIPE_DENSE &&
630 fl->num_fh != dsaddr->stripe_count)) {
631 dprintk("%s num_fh %u not valid for given packing\n",
632 __func__, fl->num_fh);
633 goto out_put;
634 }
635
636 if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
637 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
638 "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
639 nfss->wsize);
640 }
641
642 status = 0;
643out:
644 dprintk("--> %s returns %d\n", __func__, status);
645 return status;
646out_put:
Christoph Hellwigea8eecd2011-03-01 01:34:21 +0000647 nfs4_fl_put_deviceid(dsaddr);
Andy Adamson16b374c2010-10-20 00:18:04 -0400648 goto out;
649}
650
651static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
652{
653 int i;
654
655 for (i = 0; i < fl->num_fh; i++) {
656 if (!fl->fh_array[i])
657 break;
658 kfree(fl->fh_array[i]);
659 }
660 kfree(fl->fh_array);
661 fl->fh_array = NULL;
662}
663
664static void
665_filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
666{
667 filelayout_free_fh_array(fl);
668 kfree(fl);
669}
670
671static int
672filelayout_decode_layout(struct pnfs_layout_hdr *flo,
673 struct nfs4_filelayout_segment *fl,
674 struct nfs4_layoutget_res *lgr,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400675 struct nfs4_deviceid *id,
676 gfp_t gfp_flags)
Andy Adamson16b374c2010-10-20 00:18:04 -0400677{
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400678 struct xdr_stream stream;
Benny Halevyf7da7a12011-05-19 14:16:47 -0400679 struct xdr_buf buf;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400680 struct page *scratch;
681 __be32 *p;
Andy Adamson16b374c2010-10-20 00:18:04 -0400682 uint32_t nfl_util;
683 int i;
684
685 dprintk("%s: set_layout_map Begin\n", __func__);
686
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400687 scratch = alloc_page(gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400688 if (!scratch)
689 return -ENOMEM;
690
Benny Halevyf7da7a12011-05-19 14:16:47 -0400691 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400692 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
693
694 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
695 * num_fh (4) */
696 p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
697 if (unlikely(!p))
698 goto out_err;
699
Andy Adamson16b374c2010-10-20 00:18:04 -0400700 memcpy(id, p, sizeof(*id));
701 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
Benny Halevya1eaecb2011-05-19 22:14:47 -0400702 nfs4_print_deviceid(id);
Andy Adamson16b374c2010-10-20 00:18:04 -0400703
704 nfl_util = be32_to_cpup(p++);
705 if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
706 fl->commit_through_mds = 1;
707 if (nfl_util & NFL4_UFLG_DENSE)
708 fl->stripe_type = STRIPE_DENSE;
709 else
710 fl->stripe_type = STRIPE_SPARSE;
711 fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
712
713 fl->first_stripe_index = be32_to_cpup(p++);
714 p = xdr_decode_hyper(p, &fl->pattern_offset);
715 fl->num_fh = be32_to_cpup(p++);
716
717 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
718 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
719 fl->pattern_offset);
720
Andy Adamsoncec765c2011-06-13 18:36:17 -0400721 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
722 * Futher checking is done in filelayout_check_layout */
Chuck Levere4149662011-10-25 12:18:03 -0400723 if (fl->num_fh >
Andy Adamsoncec765c2011-06-13 18:36:17 -0400724 max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400725 goto out_err;
726
Andy Adamsoncec765c2011-06-13 18:36:17 -0400727 if (fl->num_fh > 0) {
728 fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *),
729 gfp_flags);
730 if (!fl->fh_array)
731 goto out_err;
732 }
Andy Adamson16b374c2010-10-20 00:18:04 -0400733
734 for (i = 0; i < fl->num_fh; i++) {
735 /* Do we want to use a mempool here? */
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400736 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400737 if (!fl->fh_array[i])
738 goto out_err_free;
739
740 p = xdr_inline_decode(&stream, 4);
741 if (unlikely(!p))
742 goto out_err_free;
Andy Adamson16b374c2010-10-20 00:18:04 -0400743 fl->fh_array[i]->size = be32_to_cpup(p++);
744 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
Weston Andros Adamsonf9fd2d92012-01-26 13:32:22 -0500745 printk(KERN_ERR "NFS: Too big fh %d received %d\n",
Andy Adamson16b374c2010-10-20 00:18:04 -0400746 i, fl->fh_array[i]->size);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400747 goto out_err_free;
Andy Adamson16b374c2010-10-20 00:18:04 -0400748 }
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400749
750 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
751 if (unlikely(!p))
752 goto out_err_free;
Andy Adamson16b374c2010-10-20 00:18:04 -0400753 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
Andy Adamson16b374c2010-10-20 00:18:04 -0400754 dprintk("DEBUG: %s: fh len %d\n", __func__,
755 fl->fh_array[i]->size);
756 }
757
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400758 __free_page(scratch);
Andy Adamson16b374c2010-10-20 00:18:04 -0400759 return 0;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400760
761out_err_free:
762 filelayout_free_fh_array(fl);
763out_err:
764 __free_page(scratch);
765 return -EIO;
Andy Adamson16b374c2010-10-20 00:18:04 -0400766}
767
Fred Isamanc8795132011-03-23 13:27:49 +0000768static void
769filelayout_free_lseg(struct pnfs_layout_segment *lseg)
770{
771 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
772
773 dprintk("--> %s\n", __func__);
774 nfs4_fl_put_deviceid(fl->dsaddr);
Fred Isaman799ba8d2012-04-20 14:47:38 -0400775 /* This assumes a single RW lseg */
776 if (lseg->pls_range.iomode == IOMODE_RW) {
777 struct nfs4_filelayout *flo;
778
779 flo = FILELAYOUT_FROM_HDR(lseg->pls_layout);
780 flo->commit_info.nbuckets = 0;
781 kfree(flo->commit_info.buckets);
782 flo->commit_info.buckets = NULL;
783 }
Fred Isamanc8795132011-03-23 13:27:49 +0000784 _filelayout_free_lseg(fl);
785}
786
Fred Isaman799ba8d2012-04-20 14:47:38 -0400787static int
788filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg,
Fred Isamanea2cf222012-04-20 14:47:53 -0400789 struct nfs_commit_info *cinfo,
Fred Isaman799ba8d2012-04-20 14:47:38 -0400790 gfp_t gfp_flags)
791{
792 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
Fred Isamanea2cf222012-04-20 14:47:53 -0400793 struct pnfs_commit_bucket *buckets;
Fred Isaman799ba8d2012-04-20 14:47:38 -0400794 int size;
795
796 if (fl->commit_through_mds)
797 return 0;
Fred Isamanea2cf222012-04-20 14:47:53 -0400798 if (cinfo->ds->nbuckets != 0) {
Fred Isaman799ba8d2012-04-20 14:47:38 -0400799 /* This assumes there is only one IOMODE_RW lseg. What
800 * we really want to do is have a layout_hdr level
801 * dictionary of <multipath_list4, fh> keys, each
802 * associated with a struct list_head, populated by calls
803 * to filelayout_write_pagelist().
804 * */
805 return 0;
806 }
807
808 size = (fl->stripe_type == STRIPE_SPARSE) ?
809 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
810
Fred Isamanea2cf222012-04-20 14:47:53 -0400811 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
Fred Isaman799ba8d2012-04-20 14:47:38 -0400812 gfp_flags);
813 if (!buckets)
814 return -ENOMEM;
815 else {
816 int i;
817
Fred Isamanea2cf222012-04-20 14:47:53 -0400818 spin_lock(cinfo->lock);
819 if (cinfo->ds->nbuckets != 0)
Fred Isaman799ba8d2012-04-20 14:47:38 -0400820 kfree(buckets);
821 else {
Fred Isamanea2cf222012-04-20 14:47:53 -0400822 cinfo->ds->buckets = buckets;
823 cinfo->ds->nbuckets = size;
Fred Isaman799ba8d2012-04-20 14:47:38 -0400824 for (i = 0; i < size; i++) {
825 INIT_LIST_HEAD(&buckets[i].written);
826 INIT_LIST_HEAD(&buckets[i].committing);
827 }
828 }
Fred Isamanea2cf222012-04-20 14:47:53 -0400829 spin_unlock(cinfo->lock);
Fred Isaman799ba8d2012-04-20 14:47:38 -0400830 return 0;
831 }
832}
833
Andy Adamson16b374c2010-10-20 00:18:04 -0400834static struct pnfs_layout_segment *
835filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400836 struct nfs4_layoutget_res *lgr,
837 gfp_t gfp_flags)
Andy Adamson16b374c2010-10-20 00:18:04 -0400838{
839 struct nfs4_filelayout_segment *fl;
840 int rc;
841 struct nfs4_deviceid id;
842
843 dprintk("--> %s\n", __func__);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400844 fl = kzalloc(sizeof(*fl), gfp_flags);
Andy Adamson16b374c2010-10-20 00:18:04 -0400845 if (!fl)
846 return NULL;
847
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400848 rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
849 if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
Andy Adamson16b374c2010-10-20 00:18:04 -0400850 _filelayout_free_lseg(fl);
851 return NULL;
852 }
853 return &fl->generic_hdr;
854}
855
Fred Isaman94ad1c82011-03-01 01:34:14 +0000856/*
857 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
858 *
Benny Halevy18ad0a92011-05-25 21:03:56 +0300859 * return true : coalesce page
860 * return false : don't coalesce page
Fred Isaman94ad1c82011-03-01 01:34:14 +0000861 */
Trond Myklebust1751c362011-06-10 13:30:23 -0400862static bool
Fred Isaman94ad1c82011-03-01 01:34:14 +0000863filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
864 struct nfs_page *req)
865{
866 u64 p_stripe, r_stripe;
867 u32 stripe_unit;
868
Benny Halevy19345cb2011-06-19 18:33:46 -0400869 if (!pnfs_generic_pg_test(pgio, prev, req) ||
870 !nfs_generic_pg_test(pgio, prev, req))
871 return false;
Benny Halevy89a58e32011-05-25 20:54:40 +0300872
Fred Isamanb5542842012-04-20 14:47:43 -0400873 p_stripe = (u64)req_offset(prev);
874 r_stripe = (u64)req_offset(req);
Fred Isaman94ad1c82011-03-01 01:34:14 +0000875 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
876
877 do_div(p_stripe, stripe_unit);
878 do_div(r_stripe, stripe_unit);
879
880 return (p_stripe == r_stripe);
881}
882
Trond Myklebust17280172012-03-11 13:11:00 -0400883static void
Andy Adamson7c24d942011-06-13 18:22:38 -0400884filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
885 struct nfs_page *req)
886{
887 BUG_ON(pgio->pg_lseg != NULL);
888
Fred Isaman1825a0d2012-04-20 19:55:31 -0400889 if (req->wb_offset != req->wb_pgbase) {
890 /*
891 * Handling unaligned pages is difficult, because have to
892 * somehow split a req in two in certain cases in the
893 * pg.test code. Avoid this by just not using pnfs
894 * in this case.
895 */
896 nfs_pageio_reset_read_mds(pgio);
897 return;
898 }
Andy Adamson7c24d942011-06-13 18:22:38 -0400899 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
900 req->wb_context,
901 0,
902 NFS4_MAX_UINT64,
903 IOMODE_READ,
904 GFP_KERNEL);
905 /* If no lseg, fall back to read through mds */
906 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -0400907 nfs_pageio_reset_read_mds(pgio);
Andy Adamson7c24d942011-06-13 18:22:38 -0400908}
909
Trond Myklebust17280172012-03-11 13:11:00 -0400910static void
Andy Adamson7c24d942011-06-13 18:22:38 -0400911filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
912 struct nfs_page *req)
913{
Fred Isamanea2cf222012-04-20 14:47:53 -0400914 struct nfs_commit_info cinfo;
Fred Isaman799ba8d2012-04-20 14:47:38 -0400915 int status;
916
Andy Adamson7c24d942011-06-13 18:22:38 -0400917 BUG_ON(pgio->pg_lseg != NULL);
918
Fred Isaman1825a0d2012-04-20 19:55:31 -0400919 if (req->wb_offset != req->wb_pgbase)
920 goto out_mds;
Andy Adamson7c24d942011-06-13 18:22:38 -0400921 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
922 req->wb_context,
923 0,
924 NFS4_MAX_UINT64,
925 IOMODE_RW,
926 GFP_NOFS);
927 /* If no lseg, fall back to write through mds */
928 if (pgio->pg_lseg == NULL)
Fred Isaman799ba8d2012-04-20 14:47:38 -0400929 goto out_mds;
Fred Isamanea2cf222012-04-20 14:47:53 -0400930 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
931 status = filelayout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
Fred Isaman799ba8d2012-04-20 14:47:38 -0400932 if (status < 0) {
933 put_lseg(pgio->pg_lseg);
934 pgio->pg_lseg = NULL;
935 goto out_mds;
936 }
937 return;
938out_mds:
939 nfs_pageio_reset_write_mds(pgio);
Andy Adamson7c24d942011-06-13 18:22:38 -0400940}
941
Trond Myklebust1751c362011-06-10 13:30:23 -0400942static const struct nfs_pageio_ops filelayout_pg_read_ops = {
Andy Adamson7c24d942011-06-13 18:22:38 -0400943 .pg_init = filelayout_pg_init_read,
Trond Myklebust1751c362011-06-10 13:30:23 -0400944 .pg_test = filelayout_pg_test,
Trond Myklebust493292d2011-07-13 15:58:28 -0400945 .pg_doio = pnfs_generic_pg_readpages,
Trond Myklebust1751c362011-06-10 13:30:23 -0400946};
947
948static const struct nfs_pageio_ops filelayout_pg_write_ops = {
Andy Adamson7c24d942011-06-13 18:22:38 -0400949 .pg_init = filelayout_pg_init_write,
Trond Myklebust1751c362011-06-10 13:30:23 -0400950 .pg_test = filelayout_pg_test,
Trond Myklebustdce81292011-07-13 15:59:19 -0400951 .pg_doio = pnfs_generic_pg_writepages,
Trond Myklebust1751c362011-06-10 13:30:23 -0400952};
953
Fred Isamane0c2b382011-03-23 13:27:53 +0000954static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
955{
956 if (fl->stripe_type == STRIPE_SPARSE)
957 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
958 else
959 return j;
960}
961
Fred Isamand6d6dc72012-03-08 17:29:35 -0500962/* The generic layer is about to remove the req from the commit list.
963 * If this will make the bucket empty, it will need to put the lseg reference.
Fred Isamand6d6dc72012-03-08 17:29:35 -0500964 */
Trond Myklebust8dd37752012-03-15 17:16:40 -0400965static void
Fred Isamanea2cf222012-04-20 14:47:53 -0400966filelayout_clear_request_commit(struct nfs_page *req,
967 struct nfs_commit_info *cinfo)
Fred Isamane0c2b382011-03-23 13:27:53 +0000968{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400969 struct pnfs_layout_segment *freeme = NULL;
Trond Myklebust8dd37752012-03-15 17:16:40 -0400970
Fred Isamanea2cf222012-04-20 14:47:53 -0400971 spin_lock(cinfo->lock);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400972 if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
973 goto out;
Fred Isamanea2cf222012-04-20 14:47:53 -0400974 cinfo->ds->nwritten--;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500975 if (list_is_singular(&req->wb_list)) {
Fred Isamanea2cf222012-04-20 14:47:53 -0400976 struct pnfs_commit_bucket *bucket;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500977
Fred Isaman799ba8d2012-04-20 14:47:38 -0400978 bucket = list_first_entry(&req->wb_list,
Fred Isamanea2cf222012-04-20 14:47:53 -0400979 struct pnfs_commit_bucket,
Fred Isaman799ba8d2012-04-20 14:47:38 -0400980 written);
981 freeme = bucket->wlseg;
982 bucket->wlseg = NULL;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500983 }
Trond Myklebust8dd37752012-03-15 17:16:40 -0400984out:
Fred Isamanea2cf222012-04-20 14:47:53 -0400985 nfs_request_remove_commit_list(req, cinfo);
986 spin_unlock(cinfo->lock);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400987 put_lseg(freeme);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500988}
989
990static struct list_head *
991filelayout_choose_commit_list(struct nfs_page *req,
Fred Isamanea2cf222012-04-20 14:47:53 -0400992 struct pnfs_layout_segment *lseg,
993 struct nfs_commit_info *cinfo)
Fred Isamand6d6dc72012-03-08 17:29:35 -0500994{
Fred Isamane0c2b382011-03-23 13:27:53 +0000995 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
996 u32 i, j;
997 struct list_head *list;
Fred Isamanea2cf222012-04-20 14:47:53 -0400998 struct pnfs_commit_bucket *buckets;
Fred Isamane0c2b382011-03-23 13:27:53 +0000999
Fred Isamand6d6dc72012-03-08 17:29:35 -05001000 if (fl->commit_through_mds)
Fred Isamanea2cf222012-04-20 14:47:53 -04001001 return &cinfo->mds->list;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001002
Fred Isamane0c2b382011-03-23 13:27:53 +00001003 /* Note that we are calling nfs4_fl_calc_j_index on each page
1004 * that ends up being committed to a data server. An attractive
1005 * alternative is to add a field to nfs_write_data and nfs_page
1006 * to store the value calculated in filelayout_write_pagelist
1007 * and just use that here.
1008 */
Fred Isamanb5542842012-04-20 14:47:43 -04001009 j = nfs4_fl_calc_j_index(lseg, req_offset(req));
Fred Isamane0c2b382011-03-23 13:27:53 +00001010 i = select_bucket_index(fl, j);
Fred Isamanea2cf222012-04-20 14:47:53 -04001011 buckets = cinfo->ds->buckets;
Fred Isaman799ba8d2012-04-20 14:47:38 -04001012 list = &buckets[i].written;
Fred Isamane0c2b382011-03-23 13:27:53 +00001013 if (list_empty(list)) {
Fred Isamand6d6dc72012-03-08 17:29:35 -05001014 /* Non-empty buckets hold a reference on the lseg. That ref
1015 * is normally transferred to the COMMIT call and released
1016 * there. It could also be released if the last req is pulled
1017 * off due to a rewrite, in which case it will be done in
Fred Isaman799ba8d2012-04-20 14:47:38 -04001018 * filelayout_clear_request_commit
Fred Isamand6d6dc72012-03-08 17:29:35 -05001019 */
Fred Isaman799ba8d2012-04-20 14:47:38 -04001020 buckets[i].wlseg = get_lseg(lseg);
Fred Isamane0c2b382011-03-23 13:27:53 +00001021 }
Trond Myklebust8dd37752012-03-15 17:16:40 -04001022 set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
Fred Isamanea2cf222012-04-20 14:47:53 -04001023 cinfo->ds->nwritten++;
Fred Isamane0c2b382011-03-23 13:27:53 +00001024 return list;
1025}
1026
Trond Myklebust8dd37752012-03-15 17:16:40 -04001027static void
1028filelayout_mark_request_commit(struct nfs_page *req,
Fred Isamanea2cf222012-04-20 14:47:53 -04001029 struct pnfs_layout_segment *lseg,
1030 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -04001031{
1032 struct list_head *list;
1033
Fred Isamanea2cf222012-04-20 14:47:53 -04001034 list = filelayout_choose_commit_list(req, lseg, cinfo);
1035 nfs_request_add_commit_list(req, list, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -04001036}
1037
Fred Isamane0c2b382011-03-23 13:27:53 +00001038static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1039{
1040 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1041
1042 if (flseg->stripe_type == STRIPE_SPARSE)
1043 return i;
1044 else
1045 return nfs4_fl_calc_ds_index(lseg, i);
1046}
1047
1048static struct nfs_fh *
1049select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1050{
1051 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1052
1053 if (flseg->stripe_type == STRIPE_SPARSE) {
1054 if (flseg->num_fh == 1)
1055 i = 0;
1056 else if (flseg->num_fh == 0)
1057 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
1058 return NULL;
1059 }
1060 return flseg->fh_array[i];
1061}
1062
Fred Isaman0b7c0152012-04-20 14:47:39 -04001063static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
Fred Isamane0c2b382011-03-23 13:27:53 +00001064{
1065 struct pnfs_layout_segment *lseg = data->lseg;
1066 struct nfs4_pnfs_ds *ds;
1067 u32 idx;
1068 struct nfs_fh *fh;
1069
1070 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1071 ds = nfs4_fl_prepare_ds(lseg, idx);
1072 if (!ds) {
Fred Isamane0c2b382011-03-23 13:27:53 +00001073 prepare_to_resend_writes(data);
Trond Myklebust8dd37752012-03-15 17:16:40 -04001074 filelayout_commit_release(data);
Fred Isamane0c2b382011-03-23 13:27:53 +00001075 return -EAGAIN;
1076 }
Andy Adamson3a7936c2012-04-27 17:53:51 -04001077 dprintk("%s ino %lu, how %d cl_count %d\n", __func__,
1078 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count));
Fred Isaman0b7c0152012-04-20 14:47:39 -04001079 data->commit_done_cb = filelayout_commit_done_cb;
Andy Adamson3a7936c2012-04-27 17:53:51 -04001080 atomic_inc(&ds->ds_clp->cl_count);
Fred Isamane0c2b382011-03-23 13:27:53 +00001081 data->ds_clp = ds->ds_clp;
1082 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1083 if (fh)
1084 data->args.fh = fh;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001085 return nfs_initiate_commit(ds->ds_clp->cl_rpcclient, data,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001086 &filelayout_commit_call_ops, how,
1087 RPC_TASK_SOFTCONN);
Fred Isamane0c2b382011-03-23 13:27:53 +00001088}
1089
Trond Myklebust8dd37752012-03-15 17:16:40 -04001090static int
Fred Isaman1763da12012-04-20 14:47:57 -04001091transfer_commit_list(struct list_head *src, struct list_head *dst,
1092 struct nfs_commit_info *cinfo, int max)
Trond Myklebust8dd37752012-03-15 17:16:40 -04001093{
Trond Myklebust8dd37752012-03-15 17:16:40 -04001094 struct nfs_page *req, *tmp;
1095 int ret = 0;
1096
1097 list_for_each_entry_safe(req, tmp, src, wb_list) {
1098 if (!nfs_lock_request(req))
1099 continue;
Trond Myklebust53b8ee32012-05-22 16:36:27 -04001100 kref_get(&req->wb_kref);
Fred Isamanea2cf222012-04-20 14:47:53 -04001101 if (cond_resched_lock(cinfo->lock))
Trond Myklebust3b3be882012-03-17 11:59:30 -04001102 list_safe_reset_next(req, tmp, wb_list);
Fred Isamanea2cf222012-04-20 14:47:53 -04001103 nfs_request_remove_commit_list(req, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -04001104 clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1105 nfs_list_add_request(req, dst);
1106 ret++;
Fred Isaman1763da12012-04-20 14:47:57 -04001107 if ((ret == max) && !cinfo->dreq)
Trond Myklebust8dd37752012-03-15 17:16:40 -04001108 break;
1109 }
Fred Isaman1763da12012-04-20 14:47:57 -04001110 return ret;
1111}
1112
1113static int
1114filelayout_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
1115 struct nfs_commit_info *cinfo,
1116 int max)
1117{
1118 struct list_head *src = &bucket->written;
1119 struct list_head *dst = &bucket->committing;
1120 int ret;
1121
1122 ret = transfer_commit_list(src, dst, cinfo, max);
Fred Isaman799ba8d2012-04-20 14:47:38 -04001123 if (ret) {
Fred Isamanea2cf222012-04-20 14:47:53 -04001124 cinfo->ds->nwritten -= ret;
1125 cinfo->ds->ncommitting += ret;
Fred Isaman799ba8d2012-04-20 14:47:38 -04001126 bucket->clseg = bucket->wlseg;
1127 if (list_empty(src))
1128 bucket->wlseg = NULL;
1129 else
1130 get_lseg(bucket->clseg);
1131 }
Trond Myklebust8dd37752012-03-15 17:16:40 -04001132 return ret;
1133}
1134
Fred Isamand6d6dc72012-03-08 17:29:35 -05001135/* Move reqs from written to committing lists, returning count of number moved.
Fred Isamanea2cf222012-04-20 14:47:53 -04001136 * Note called with cinfo->lock held.
Fred Isamand6d6dc72012-03-08 17:29:35 -05001137 */
Fred Isamanea2cf222012-04-20 14:47:53 -04001138static int filelayout_scan_commit_lists(struct nfs_commit_info *cinfo,
1139 int max)
Fred Isamand6d6dc72012-03-08 17:29:35 -05001140{
Fred Isamand6d6dc72012-03-08 17:29:35 -05001141 int i, rv = 0, cnt;
1142
Fred Isamanea2cf222012-04-20 14:47:53 -04001143 for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
1144 cnt = filelayout_scan_ds_commit_list(&cinfo->ds->buckets[i],
1145 cinfo, max);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001146 max -= cnt;
1147 rv += cnt;
1148 }
Fred Isamand6d6dc72012-03-08 17:29:35 -05001149 return rv;
1150}
1151
Fred Isaman1763da12012-04-20 14:47:57 -04001152/* Pull everything off the committing lists and dump into @dst */
1153static void filelayout_recover_commit_reqs(struct list_head *dst,
1154 struct nfs_commit_info *cinfo)
1155{
1156 struct pnfs_commit_bucket *b;
1157 int i;
1158
1159 /* NOTE cinfo->lock is NOT held, relying on fact that this is
1160 * only called on single thread per dreq.
1161 * Can't take the lock because need to do put_lseg
1162 */
1163 for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
1164 if (transfer_commit_list(&b->written, dst, cinfo, 0)) {
1165 BUG_ON(!list_empty(&b->written));
1166 put_lseg(b->wlseg);
1167 b->wlseg = NULL;
1168 }
1169 }
1170 cinfo->ds->nwritten = 0;
1171}
1172
Trond Myklebust9390f422012-03-16 13:52:45 -04001173static unsigned int
Fred Isamanea2cf222012-04-20 14:47:53 -04001174alloc_ds_commits(struct nfs_commit_info *cinfo, struct list_head *list)
Fred Isamane0c2b382011-03-23 13:27:53 +00001175{
Fred Isamanea2cf222012-04-20 14:47:53 -04001176 struct pnfs_ds_commit_info *fl_cinfo;
1177 struct pnfs_commit_bucket *bucket;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001178 struct nfs_commit_data *data;
Fred Isamane0c2b382011-03-23 13:27:53 +00001179 int i, j;
Trond Myklebust9390f422012-03-16 13:52:45 -04001180 unsigned int nreq = 0;
Fred Isamane0c2b382011-03-23 13:27:53 +00001181
Fred Isamanea2cf222012-04-20 14:47:53 -04001182 fl_cinfo = cinfo->ds;
Fred Isaman799ba8d2012-04-20 14:47:38 -04001183 bucket = fl_cinfo->buckets;
1184 for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) {
1185 if (list_empty(&bucket->committing))
Fred Isamane0c2b382011-03-23 13:27:53 +00001186 continue;
1187 data = nfs_commitdata_alloc();
1188 if (!data)
Trond Myklebust9390f422012-03-16 13:52:45 -04001189 break;
Fred Isamane0c2b382011-03-23 13:27:53 +00001190 data->ds_commit_index = i;
Fred Isaman799ba8d2012-04-20 14:47:38 -04001191 data->lseg = bucket->clseg;
1192 bucket->clseg = NULL;
Fred Isamane0c2b382011-03-23 13:27:53 +00001193 list_add(&data->pages, list);
Trond Myklebust9390f422012-03-16 13:52:45 -04001194 nreq++;
Fred Isamane0c2b382011-03-23 13:27:53 +00001195 }
Fred Isamane0c2b382011-03-23 13:27:53 +00001196
Trond Myklebust9390f422012-03-16 13:52:45 -04001197 /* Clean up on error */
Fred Isaman799ba8d2012-04-20 14:47:38 -04001198 for (j = i; j < fl_cinfo->nbuckets; j++, bucket++) {
1199 if (list_empty(&bucket->committing))
Fred Isamane0c2b382011-03-23 13:27:53 +00001200 continue;
Fred Isamanea2cf222012-04-20 14:47:53 -04001201 nfs_retry_commit(&bucket->committing, bucket->clseg, cinfo);
Fred Isaman799ba8d2012-04-20 14:47:38 -04001202 put_lseg(bucket->clseg);
1203 bucket->clseg = NULL;
Fred Isamane0c2b382011-03-23 13:27:53 +00001204 }
Fred Isamane0c2b382011-03-23 13:27:53 +00001205 /* Caller will clean up entries put on list */
Trond Myklebust9390f422012-03-16 13:52:45 -04001206 return nreq;
Fred Isamane0c2b382011-03-23 13:27:53 +00001207}
1208
1209/* This follows nfs_commit_list pretty closely */
1210static int
1211filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
Fred Isamanea2cf222012-04-20 14:47:53 -04001212 int how, struct nfs_commit_info *cinfo)
Fred Isamane0c2b382011-03-23 13:27:53 +00001213{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001214 struct nfs_commit_data *data, *tmp;
Fred Isamane0c2b382011-03-23 13:27:53 +00001215 LIST_HEAD(list);
Trond Myklebust9390f422012-03-16 13:52:45 -04001216 unsigned int nreq = 0;
Fred Isamane0c2b382011-03-23 13:27:53 +00001217
1218 if (!list_empty(mds_pages)) {
1219 data = nfs_commitdata_alloc();
Trond Myklebust9390f422012-03-16 13:52:45 -04001220 if (data != NULL) {
1221 data->lseg = NULL;
1222 list_add(&data->pages, &list);
1223 nreq++;
1224 } else
Fred Isamanea2cf222012-04-20 14:47:53 -04001225 nfs_retry_commit(mds_pages, NULL, cinfo);
Fred Isamane0c2b382011-03-23 13:27:53 +00001226 }
1227
Fred Isamanea2cf222012-04-20 14:47:53 -04001228 nreq += alloc_ds_commits(cinfo, &list);
Trond Myklebust9390f422012-03-16 13:52:45 -04001229
1230 if (nreq == 0) {
Fred Isamanf453a542012-04-20 14:47:54 -04001231 cinfo->completion_ops->error_cleanup(NFS_I(inode));
Trond Myklebust9390f422012-03-16 13:52:45 -04001232 goto out;
1233 }
1234
Fred Isamanea2cf222012-04-20 14:47:53 -04001235 atomic_add(nreq, &cinfo->mds->rpcs_out);
Fred Isamane0c2b382011-03-23 13:27:53 +00001236
1237 list_for_each_entry_safe(data, tmp, &list, pages) {
1238 list_del_init(&data->pages);
Fred Isamane0c2b382011-03-23 13:27:53 +00001239 if (!data->lseg) {
Fred Isamanf453a542012-04-20 14:47:54 -04001240 nfs_init_commit(data, mds_pages, NULL, cinfo);
Fred Isaman0b7c0152012-04-20 14:47:39 -04001241 nfs_initiate_commit(NFS_CLIENT(inode), data,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001242 data->mds_ops, how, 0);
Fred Isamane0c2b382011-03-23 13:27:53 +00001243 } else {
Fred Isamanea2cf222012-04-20 14:47:53 -04001244 struct pnfs_commit_bucket *buckets;
Fred Isaman799ba8d2012-04-20 14:47:38 -04001245
Fred Isamanea2cf222012-04-20 14:47:53 -04001246 buckets = cinfo->ds->buckets;
Fred Isamanf453a542012-04-20 14:47:54 -04001247 nfs_init_commit(data, &buckets[data->ds_commit_index].committing, data->lseg, cinfo);
Fred Isamane0c2b382011-03-23 13:27:53 +00001248 filelayout_initiate_commit(data, how);
1249 }
1250 }
Trond Myklebust9390f422012-03-16 13:52:45 -04001251out:
Fred Isamanea2cf222012-04-20 14:47:53 -04001252 cinfo->ds->ncommitting = 0;
Trond Myklebust9390f422012-03-16 13:52:45 -04001253 return PNFS_ATTEMPTED;
Fred Isamane0c2b382011-03-23 13:27:53 +00001254}
1255
Benny Halevy1775bc32011-05-20 13:47:33 +02001256static void
1257filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
1258{
1259 nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1260}
1261
Fred Isaman799ba8d2012-04-20 14:47:38 -04001262static struct pnfs_layout_hdr *
1263filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
1264{
1265 struct nfs4_filelayout *flo;
1266
1267 flo = kzalloc(sizeof(*flo), gfp_flags);
1268 return &flo->generic_hdr;
1269}
1270
1271static void
1272filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
1273{
1274 kfree(FILELAYOUT_FROM_HDR(lo));
1275}
1276
Fred Isamanea2cf222012-04-20 14:47:53 -04001277static struct pnfs_ds_commit_info *
1278filelayout_get_ds_info(struct inode *inode)
1279{
Fred Isamandf011742012-04-24 14:50:34 -04001280 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1281
1282 if (layout == NULL)
1283 return NULL;
1284 else
1285 return &FILELAYOUT_FROM_HDR(layout)->commit_info;
Fred Isamanea2cf222012-04-20 14:47:53 -04001286}
1287
Dean Hildebrand7ab672c2010-10-20 00:18:00 -04001288static struct pnfs_layoutdriver_type filelayout_type = {
Christoph Hellwigea8eecd2011-03-01 01:34:21 +00001289 .id = LAYOUT_NFSV4_1_FILES,
1290 .name = "LAYOUT_NFSV4_1_FILES",
1291 .owner = THIS_MODULE,
Fred Isaman799ba8d2012-04-20 14:47:38 -04001292 .alloc_layout_hdr = filelayout_alloc_layout_hdr,
1293 .free_layout_hdr = filelayout_free_layout_hdr,
Christoph Hellwigea8eecd2011-03-01 01:34:21 +00001294 .alloc_lseg = filelayout_alloc_lseg,
1295 .free_lseg = filelayout_free_lseg,
Trond Myklebust1751c362011-06-10 13:30:23 -04001296 .pg_read_ops = &filelayout_pg_read_ops,
1297 .pg_write_ops = &filelayout_pg_write_ops,
Fred Isamanea2cf222012-04-20 14:47:53 -04001298 .get_ds_info = &filelayout_get_ds_info,
Trond Myklebust8dd37752012-03-15 17:16:40 -04001299 .mark_request_commit = filelayout_mark_request_commit,
1300 .clear_request_commit = filelayout_clear_request_commit,
Fred Isamand6d6dc72012-03-08 17:29:35 -05001301 .scan_commit_lists = filelayout_scan_commit_lists,
Fred Isaman1763da12012-04-20 14:47:57 -04001302 .recover_commit_reqs = filelayout_recover_commit_reqs,
Fred Isamane0c2b382011-03-23 13:27:53 +00001303 .commit_pagelist = filelayout_commit_pagelist,
Andy Adamsondc70d7b2011-03-01 01:34:19 +00001304 .read_pagelist = filelayout_read_pagelist,
Andy Adamson0382b742011-03-03 15:13:45 +00001305 .write_pagelist = filelayout_write_pagelist,
Benny Halevy1775bc32011-05-20 13:47:33 +02001306 .free_deviceid_node = filelayout_free_deveiceid_node,
Dean Hildebrand7ab672c2010-10-20 00:18:00 -04001307};
1308
1309static int __init nfs4filelayout_init(void)
1310{
1311 printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1312 __func__);
1313 return pnfs_register_layoutdriver(&filelayout_type);
1314}
1315
1316static void __exit nfs4filelayout_exit(void)
1317{
1318 printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1319 __func__);
1320 pnfs_unregister_layoutdriver(&filelayout_type);
1321}
1322
J. Bruce Fieldsf85ef692011-07-15 19:18:42 -04001323MODULE_ALIAS("nfs-layouttype4-1");
1324
Dean Hildebrand7ab672c2010-10-20 00:18:00 -04001325module_init(nfs4filelayout_init);
1326module_exit(nfs4filelayout_exit);