Commit | Line | Data |
---|---|---|
7ab672ce DH |
1 | /* |
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> | |
19345cb2 | 33 | #include <linux/nfs_page.h> |
143cb494 | 34 | #include <linux/module.h> |
16b374ca | 35 | |
0a702195 WAA |
36 | #include <linux/sunrpc/metrics.h> |
37 | ||
b5968725 TH |
38 | #include "../nfs4session.h" |
39 | #include "../internal.h" | |
40 | #include "../delegation.h" | |
41 | #include "filelayout.h" | |
42 | #include "../nfs4trace.h" | |
7ab672ce DH |
43 | |
44 | #define NFSDBG_FACILITY NFSDBG_PNFS_LD | |
45 | ||
46 | MODULE_LICENSE("GPL"); | |
47 | MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>"); | |
48 | MODULE_DESCRIPTION("The NFSv4 file layout driver"); | |
49 | ||
cbdabc7f AA |
50 | #define FILELAYOUT_POLL_RETRY_MAX (15*HZ) |
51 | ||
cfe7f412 FI |
52 | static loff_t |
53 | filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg, | |
54 | loff_t offset) | |
55 | { | |
56 | u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count; | |
3476f114 CM |
57 | u64 stripe_no; |
58 | u32 rem; | |
cfe7f412 FI |
59 | |
60 | offset -= flseg->pattern_offset; | |
3476f114 CM |
61 | stripe_no = div_u64(offset, stripe_width); |
62 | div_u64_rem(offset, flseg->stripe_unit, &rem); | |
cfe7f412 | 63 | |
3476f114 | 64 | return stripe_no * flseg->stripe_unit + rem; |
cfe7f412 FI |
65 | } |
66 | ||
67 | /* This function is used by the layout driver to calculate the | |
68 | * offset of the file on the dserver based on whether the | |
69 | * layout type is STRIPE_DENSE or STRIPE_SPARSE | |
70 | */ | |
71 | static loff_t | |
72 | filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset) | |
73 | { | |
74 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
75 | ||
76 | switch (flseg->stripe_type) { | |
77 | case STRIPE_SPARSE: | |
78 | return offset; | |
79 | ||
80 | case STRIPE_DENSE: | |
81 | return filelayout_get_dense_offset(flseg, offset); | |
82 | } | |
83 | ||
84 | BUG(); | |
85 | } | |
86 | ||
d45f60c6 | 87 | static void filelayout_reset_write(struct nfs_pgio_header *hdr) |
e7dd79af | 88 | { |
d45f60c6 | 89 | struct rpc_task *task = &hdr->task; |
e7dd79af AA |
90 | |
91 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { | |
92 | dprintk("%s Reset task %5u for i/o through MDS " | |
1e8968c5 | 93 | "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, |
d45f60c6 | 94 | hdr->task.tk_pid, |
497826af | 95 | hdr->inode->i_sb->s_id, |
1e8968c5 | 96 | (unsigned long long)NFS_FILEID(hdr->inode), |
d45f60c6 WAA |
97 | hdr->args.count, |
98 | (unsigned long long)hdr->args.offset); | |
e7dd79af | 99 | |
53113ad3 | 100 | task->tk_status = pnfs_write_done_resend_to_mds(hdr); |
e7dd79af AA |
101 | } |
102 | } | |
103 | ||
d45f60c6 | 104 | static void filelayout_reset_read(struct nfs_pgio_header *hdr) |
e7dd79af | 105 | { |
d45f60c6 | 106 | struct rpc_task *task = &hdr->task; |
e7dd79af AA |
107 | |
108 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { | |
109 | dprintk("%s Reset task %5u for i/o through MDS " | |
1e8968c5 | 110 | "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, |
d45f60c6 | 111 | hdr->task.tk_pid, |
497826af | 112 | hdr->inode->i_sb->s_id, |
1e8968c5 | 113 | (unsigned long long)NFS_FILEID(hdr->inode), |
d45f60c6 WAA |
114 | hdr->args.count, |
115 | (unsigned long long)hdr->args.offset); | |
e7dd79af | 116 | |
53113ad3 | 117 | task->tk_status = pnfs_read_done_resend_to_mds(hdr); |
e7dd79af AA |
118 | } |
119 | } | |
120 | ||
cbdabc7f AA |
121 | static int filelayout_async_handle_error(struct rpc_task *task, |
122 | struct nfs4_state *state, | |
123 | struct nfs_client *clp, | |
e7dd79af | 124 | struct pnfs_layout_segment *lseg) |
cbdabc7f | 125 | { |
d527e5c1 TM |
126 | struct pnfs_layout_hdr *lo = lseg->pls_layout; |
127 | struct inode *inode = lo->plh_inode; | |
e7dd79af AA |
128 | struct nfs_server *mds_server = NFS_SERVER(inode); |
129 | struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg); | |
9cb81968 | 130 | struct nfs_client *mds_client = mds_server->nfs_client; |
671fb896 | 131 | struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table; |
9cb81968 | 132 | |
cbdabc7f AA |
133 | if (task->tk_status >= 0) |
134 | return 0; | |
cbdabc7f AA |
135 | |
136 | switch (task->tk_status) { | |
9cb81968 AA |
137 | /* MDS state errors */ |
138 | case -NFS4ERR_DELEG_REVOKED: | |
139 | case -NFS4ERR_ADMIN_REVOKED: | |
140 | case -NFS4ERR_BAD_STATEID: | |
9cb81968 | 141 | case -NFS4ERR_OPENMODE: |
e7dd79af AA |
142 | if (state == NULL) |
143 | break; | |
5d422301 TM |
144 | if (nfs4_schedule_stateid_recovery(mds_server, state) < 0) |
145 | goto out_bad_stateid; | |
9cb81968 AA |
146 | goto wait_on_recovery; |
147 | case -NFS4ERR_EXPIRED: | |
5d422301 TM |
148 | if (state != NULL) { |
149 | if (nfs4_schedule_stateid_recovery(mds_server, state) < 0) | |
150 | goto out_bad_stateid; | |
151 | } | |
9cb81968 AA |
152 | nfs4_schedule_lease_recovery(mds_client); |
153 | goto wait_on_recovery; | |
154 | /* DS session errors */ | |
cbdabc7f AA |
155 | case -NFS4ERR_BADSESSION: |
156 | case -NFS4ERR_BADSLOT: | |
157 | case -NFS4ERR_BAD_HIGH_SLOT: | |
158 | case -NFS4ERR_DEADSESSION: | |
159 | case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: | |
160 | case -NFS4ERR_SEQ_FALSE_RETRY: | |
161 | case -NFS4ERR_SEQ_MISORDERED: | |
162 | dprintk("%s ERROR %d, Reset session. Exchangeid " | |
163 | "flags 0x%x\n", __func__, task->tk_status, | |
164 | clp->cl_exchange_flags); | |
9f594791 | 165 | nfs4_schedule_session_recovery(clp->cl_session, task->tk_status); |
cbdabc7f AA |
166 | break; |
167 | case -NFS4ERR_DELAY: | |
168 | case -NFS4ERR_GRACE: | |
cbdabc7f AA |
169 | rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX); |
170 | break; | |
a8a4ae3a AA |
171 | case -NFS4ERR_RETRY_UNCACHED_REP: |
172 | break; | |
041245c8 AA |
173 | /* Invalidate Layout errors */ |
174 | case -NFS4ERR_PNFS_NO_LAYOUT: | |
175 | case -ESTALE: /* mapped NFS4ERR_STALE */ | |
176 | case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */ | |
177 | case -EISDIR: /* mapped NFS4ERR_ISDIR */ | |
178 | case -NFS4ERR_FHEXPIRED: | |
179 | case -NFS4ERR_WRONG_TYPE: | |
180 | dprintk("%s Invalid layout error %d\n", __func__, | |
181 | task->tk_status); | |
182 | /* | |
183 | * Destroy layout so new i/o will get a new layout. | |
184 | * Layout will not be destroyed until all current lseg | |
185 | * references are put. Mark layout as invalid to resend failed | |
186 | * i/o and all i/o waiting on the slot table to the MDS until | |
187 | * layout is destroyed and a new valid layout is obtained. | |
188 | */ | |
d42e7873 | 189 | pnfs_destroy_layout(NFS_I(inode)); |
041245c8 AA |
190 | rpc_wake_up(&tbl->slot_tbl_waitq); |
191 | goto reset; | |
e7dd79af AA |
192 | /* RPC connection errors */ |
193 | case -ECONNREFUSED: | |
194 | case -EHOSTDOWN: | |
195 | case -EHOSTUNREACH: | |
196 | case -ENETUNREACH: | |
197 | case -EIO: | |
198 | case -ETIMEDOUT: | |
199 | case -EPIPE: | |
200 | dprintk("%s DS connection error %d\n", __func__, | |
201 | task->tk_status); | |
1dfed273 | 202 | nfs4_mark_deviceid_unavailable(devid); |
c220106f | 203 | pnfs_error_mark_layout_for_return(inode, lseg); |
671fb896 | 204 | rpc_wake_up(&tbl->slot_tbl_waitq); |
e7dd79af | 205 | /* fall through */ |
cbdabc7f | 206 | default: |
041245c8 | 207 | reset: |
e7dd79af | 208 | dprintk("%s Retry through MDS. Error %d\n", __func__, |
cbdabc7f | 209 | task->tk_status); |
e7dd79af | 210 | return -NFS4ERR_RESET_TO_MDS; |
cbdabc7f | 211 | } |
9cb81968 | 212 | out: |
cbdabc7f AA |
213 | task->tk_status = 0; |
214 | return -EAGAIN; | |
5d422301 TM |
215 | out_bad_stateid: |
216 | task->tk_status = -EIO; | |
217 | return 0; | |
9cb81968 AA |
218 | wait_on_recovery: |
219 | rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL); | |
220 | if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0) | |
221 | rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task); | |
222 | goto out; | |
cbdabc7f AA |
223 | } |
224 | ||
225 | /* NFS_PROTO call done callback routines */ | |
226 | ||
227 | static int filelayout_read_done_cb(struct rpc_task *task, | |
d45f60c6 | 228 | struct nfs_pgio_header *hdr) |
cbdabc7f | 229 | { |
e7dd79af | 230 | int err; |
cbdabc7f | 231 | |
d45f60c6 WAA |
232 | trace_nfs4_pnfs_read(hdr, task->tk_status); |
233 | err = filelayout_async_handle_error(task, hdr->args.context->state, | |
234 | hdr->ds_clp, hdr->lseg); | |
cbdabc7f | 235 | |
e7dd79af AA |
236 | switch (err) { |
237 | case -NFS4ERR_RESET_TO_MDS: | |
d45f60c6 | 238 | filelayout_reset_read(hdr); |
e7dd79af AA |
239 | return task->tk_status; |
240 | case -EAGAIN: | |
d00c5d43 | 241 | rpc_restart_call_prepare(task); |
cbdabc7f AA |
242 | return -EAGAIN; |
243 | } | |
244 | ||
245 | return 0; | |
246 | } | |
247 | ||
863a3c6c AA |
248 | /* |
249 | * We reference the rpc_cred of the first WRITE that triggers the need for | |
250 | * a LAYOUTCOMMIT, and use it to send the layoutcommit compound. | |
251 | * rfc5661 is not clear about which credential should be used. | |
252 | */ | |
253 | static void | |
d45f60c6 | 254 | filelayout_set_layoutcommit(struct nfs_pgio_header *hdr) |
863a3c6c | 255 | { |
cd841605 FI |
256 | |
257 | if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds || | |
bc7d4b8f | 258 | hdr->res.verf->committed != NFS_DATA_SYNC) |
863a3c6c AA |
259 | return; |
260 | ||
d45f60c6 | 261 | pnfs_set_layoutcommit(hdr); |
f383b7e8 | 262 | dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino, |
cd841605 | 263 | (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb); |
863a3c6c AA |
264 | } |
265 | ||
1dfed273 TM |
266 | bool |
267 | filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node) | |
268 | { | |
269 | return filelayout_test_devid_invalid(node) || | |
270 | nfs4_test_deviceid_unavailable(node); | |
271 | } | |
272 | ||
273 | static bool | |
274 | filelayout_reset_to_mds(struct pnfs_layout_segment *lseg) | |
275 | { | |
276 | struct nfs4_deviceid_node *node = FILELAYOUT_DEVID_NODE(lseg); | |
277 | ||
8006bfba | 278 | return filelayout_test_devid_unavailable(node); |
1dfed273 TM |
279 | } |
280 | ||
dc70d7b3 AA |
281 | /* |
282 | * Call ops for the async read/write cases | |
283 | * In the case of dense layouts, the offset needs to be reset to its | |
284 | * original value. | |
285 | */ | |
286 | static void filelayout_read_prepare(struct rpc_task *task, void *data) | |
287 | { | |
d45f60c6 | 288 | struct nfs_pgio_header *hdr = data; |
dc70d7b3 | 289 | |
d45f60c6 | 290 | if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { |
c58c8441 TM |
291 | rpc_exit(task, -EIO); |
292 | return; | |
293 | } | |
d45f60c6 | 294 | if (filelayout_reset_to_mds(hdr->lseg)) { |
0ad2f378 | 295 | dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid); |
d45f60c6 | 296 | filelayout_reset_read(hdr); |
0ad2f378 AA |
297 | rpc_exit(task, 0); |
298 | return; | |
299 | } | |
d45f60c6 | 300 | hdr->pgio_done_cb = filelayout_read_done_cb; |
cbdabc7f | 301 | |
d45f60c6 WAA |
302 | if (nfs41_setup_sequence(hdr->ds_clp->cl_session, |
303 | &hdr->args.seq_args, | |
304 | &hdr->res.seq_res, | |
9b206149 TM |
305 | task)) |
306 | return; | |
d45f60c6 WAA |
307 | if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context, |
308 | hdr->args.lock_context, FMODE_READ) == -EIO) | |
869a9d37 | 309 | rpc_exit(task, -EIO); /* lost lock, terminate I/O */ |
dc70d7b3 AA |
310 | } |
311 | ||
312 | static void filelayout_read_call_done(struct rpc_task *task, void *data) | |
313 | { | |
d45f60c6 | 314 | struct nfs_pgio_header *hdr = data; |
dc70d7b3 AA |
315 | |
316 | dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status); | |
317 | ||
d45f60c6 | 318 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && |
f9c96fcc | 319 | task->tk_status == 0) { |
d45f60c6 | 320 | nfs41_sequence_done(task, &hdr->res.seq_res); |
0ad2f378 | 321 | return; |
f9c96fcc | 322 | } |
0ad2f378 | 323 | |
dc70d7b3 | 324 | /* Note this may cause RPC to be resent */ |
d45f60c6 | 325 | hdr->mds_ops->rpc_call_done(task, data); |
dc70d7b3 AA |
326 | } |
327 | ||
0a702195 WAA |
328 | static void filelayout_read_count_stats(struct rpc_task *task, void *data) |
329 | { | |
d45f60c6 | 330 | struct nfs_pgio_header *hdr = data; |
0a702195 | 331 | |
d45f60c6 | 332 | rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics); |
0a702195 WAA |
333 | } |
334 | ||
a69aef14 | 335 | static int filelayout_write_done_cb(struct rpc_task *task, |
d45f60c6 | 336 | struct nfs_pgio_header *hdr) |
a69aef14 | 337 | { |
e7dd79af AA |
338 | int err; |
339 | ||
d45f60c6 WAA |
340 | trace_nfs4_pnfs_write(hdr, task->tk_status); |
341 | err = filelayout_async_handle_error(task, hdr->args.context->state, | |
342 | hdr->ds_clp, hdr->lseg); | |
e7dd79af AA |
343 | |
344 | switch (err) { | |
345 | case -NFS4ERR_RESET_TO_MDS: | |
d45f60c6 | 346 | filelayout_reset_write(hdr); |
e7dd79af AA |
347 | return task->tk_status; |
348 | case -EAGAIN: | |
d00c5d43 | 349 | rpc_restart_call_prepare(task); |
a69aef14 FI |
350 | return -EAGAIN; |
351 | } | |
352 | ||
d45f60c6 | 353 | filelayout_set_layoutcommit(hdr); |
a69aef14 FI |
354 | return 0; |
355 | } | |
356 | ||
e0c2b380 | 357 | static int filelayout_commit_done_cb(struct rpc_task *task, |
0b7c0153 | 358 | struct nfs_commit_data *data) |
e0c2b380 | 359 | { |
e7dd79af AA |
360 | int err; |
361 | ||
cc668ab3 | 362 | trace_nfs4_pnfs_commit_ds(data, task->tk_status); |
e7dd79af AA |
363 | err = filelayout_async_handle_error(task, NULL, data->ds_clp, |
364 | data->lseg); | |
365 | ||
366 | switch (err) { | |
367 | case -NFS4ERR_RESET_TO_MDS: | |
f54bcf2e | 368 | pnfs_generic_prepare_to_resend_writes(data); |
e7dd79af AA |
369 | return -EAGAIN; |
370 | case -EAGAIN: | |
371 | rpc_restart_call_prepare(task); | |
e0c2b380 FI |
372 | return -EAGAIN; |
373 | } | |
374 | ||
bc7d4b8f PT |
375 | if (data->verf.committed == NFS_UNSTABLE) |
376 | pnfs_commit_set_layoutcommit(data); | |
377 | ||
e0c2b380 FI |
378 | return 0; |
379 | } | |
380 | ||
a69aef14 FI |
381 | static void filelayout_write_prepare(struct rpc_task *task, void *data) |
382 | { | |
d45f60c6 | 383 | struct nfs_pgio_header *hdr = data; |
a69aef14 | 384 | |
d45f60c6 | 385 | if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { |
c58c8441 TM |
386 | rpc_exit(task, -EIO); |
387 | return; | |
388 | } | |
d45f60c6 | 389 | if (filelayout_reset_to_mds(hdr->lseg)) { |
0ad2f378 | 390 | dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid); |
d45f60c6 | 391 | filelayout_reset_write(hdr); |
0ad2f378 AA |
392 | rpc_exit(task, 0); |
393 | return; | |
394 | } | |
d45f60c6 WAA |
395 | if (nfs41_setup_sequence(hdr->ds_clp->cl_session, |
396 | &hdr->args.seq_args, | |
397 | &hdr->res.seq_res, | |
9b206149 TM |
398 | task)) |
399 | return; | |
d45f60c6 WAA |
400 | if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context, |
401 | hdr->args.lock_context, FMODE_WRITE) == -EIO) | |
869a9d37 | 402 | rpc_exit(task, -EIO); /* lost lock, terminate I/O */ |
a69aef14 FI |
403 | } |
404 | ||
405 | static void filelayout_write_call_done(struct rpc_task *task, void *data) | |
406 | { | |
d45f60c6 | 407 | struct nfs_pgio_header *hdr = data; |
a69aef14 | 408 | |
d45f60c6 | 409 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && |
f9c96fcc | 410 | task->tk_status == 0) { |
d45f60c6 | 411 | nfs41_sequence_done(task, &hdr->res.seq_res); |
0ad2f378 | 412 | return; |
f9c96fcc | 413 | } |
0ad2f378 | 414 | |
a69aef14 | 415 | /* Note this may cause RPC to be resent */ |
d45f60c6 | 416 | hdr->mds_ops->rpc_call_done(task, data); |
a69aef14 FI |
417 | } |
418 | ||
0a702195 WAA |
419 | static void filelayout_write_count_stats(struct rpc_task *task, void *data) |
420 | { | |
d45f60c6 | 421 | struct nfs_pgio_header *hdr = data; |
0a702195 | 422 | |
d45f60c6 | 423 | rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics); |
0a702195 WAA |
424 | } |
425 | ||
0b7c0153 | 426 | static void filelayout_commit_prepare(struct rpc_task *task, void *data) |
e0c2b380 | 427 | { |
0b7c0153 | 428 | struct nfs_commit_data *wdata = data; |
e0c2b380 | 429 | |
d9afbd1b TM |
430 | nfs41_setup_sequence(wdata->ds_clp->cl_session, |
431 | &wdata->args.seq_args, | |
432 | &wdata->res.seq_res, | |
433 | task); | |
0b7c0153 FI |
434 | } |
435 | ||
0b7c0153 FI |
436 | static void filelayout_commit_count_stats(struct rpc_task *task, void *data) |
437 | { | |
438 | struct nfs_commit_data *cdata = data; | |
439 | ||
440 | rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics); | |
441 | } | |
442 | ||
17280175 | 443 | static const struct rpc_call_ops filelayout_read_call_ops = { |
dc70d7b3 AA |
444 | .rpc_call_prepare = filelayout_read_prepare, |
445 | .rpc_call_done = filelayout_read_call_done, | |
0a702195 | 446 | .rpc_count_stats = filelayout_read_count_stats, |
f54bcf2e | 447 | .rpc_release = pnfs_generic_rw_release, |
dc70d7b3 AA |
448 | }; |
449 | ||
17280175 | 450 | static const struct rpc_call_ops filelayout_write_call_ops = { |
a69aef14 FI |
451 | .rpc_call_prepare = filelayout_write_prepare, |
452 | .rpc_call_done = filelayout_write_call_done, | |
0a702195 | 453 | .rpc_count_stats = filelayout_write_count_stats, |
f54bcf2e | 454 | .rpc_release = pnfs_generic_rw_release, |
a69aef14 FI |
455 | }; |
456 | ||
17280175 | 457 | static const struct rpc_call_ops filelayout_commit_call_ops = { |
0b7c0153 | 458 | .rpc_call_prepare = filelayout_commit_prepare, |
f54bcf2e | 459 | .rpc_call_done = pnfs_generic_write_commit_done, |
0b7c0153 | 460 | .rpc_count_stats = filelayout_commit_count_stats, |
f54bcf2e | 461 | .rpc_release = pnfs_generic_commit_release, |
e0c2b380 FI |
462 | }; |
463 | ||
dc70d7b3 | 464 | static enum pnfs_try_status |
d45f60c6 | 465 | filelayout_read_pagelist(struct nfs_pgio_header *hdr) |
dc70d7b3 | 466 | { |
cd841605 | 467 | struct pnfs_layout_segment *lseg = hdr->lseg; |
dc70d7b3 | 468 | struct nfs4_pnfs_ds *ds; |
0e20162e | 469 | struct rpc_clnt *ds_clnt; |
d45f60c6 | 470 | loff_t offset = hdr->args.offset; |
dc70d7b3 AA |
471 | u32 j, idx; |
472 | struct nfs_fh *fh; | |
dc70d7b3 AA |
473 | |
474 | dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n", | |
cd841605 | 475 | __func__, hdr->inode->i_ino, |
d45f60c6 | 476 | hdr->args.pgbase, (size_t)hdr->args.count, offset); |
dc70d7b3 AA |
477 | |
478 | /* Retrieve the correct rpc_client for the byte range */ | |
479 | j = nfs4_fl_calc_j_index(lseg, offset); | |
480 | idx = nfs4_fl_calc_ds_index(lseg, j); | |
481 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
90fecfcb | 482 | if (!ds) |
dc70d7b3 | 483 | return PNFS_NOT_ATTEMPTED; |
0e20162e AA |
484 | |
485 | ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode); | |
486 | if (IS_ERR(ds_clnt)) | |
487 | return PNFS_NOT_ATTEMPTED; | |
488 | ||
3a7936c3 AA |
489 | dprintk("%s USE DS: %s cl_count %d\n", __func__, |
490 | ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count)); | |
dc70d7b3 AA |
491 | |
492 | /* No multipath support. Use first DS */ | |
3a7936c3 | 493 | atomic_inc(&ds->ds_clp->cl_count); |
d45f60c6 | 494 | hdr->ds_clp = ds->ds_clp; |
6cccbb6f | 495 | hdr->ds_commit_idx = idx; |
dc70d7b3 AA |
496 | fh = nfs4_fl_select_ds_fh(lseg, j); |
497 | if (fh) | |
d45f60c6 | 498 | hdr->args.fh = fh; |
dc70d7b3 | 499 | |
d45f60c6 WAA |
500 | hdr->args.offset = filelayout_get_dserver_offset(lseg, offset); |
501 | hdr->mds_offset = offset; | |
dc70d7b3 AA |
502 | |
503 | /* Perform an asynchronous read to ds */ | |
46a5ab47 PT |
504 | nfs_initiate_pgio(ds_clnt, hdr, hdr->cred, |
505 | NFS_PROTO(hdr->inode), &filelayout_read_call_ops, | |
506 | 0, RPC_TASK_SOFTCONN); | |
dc70d7b3 AA |
507 | return PNFS_ATTEMPTED; |
508 | } | |
509 | ||
a69aef14 | 510 | /* Perform async writes. */ |
0382b744 | 511 | static enum pnfs_try_status |
d45f60c6 | 512 | filelayout_write_pagelist(struct nfs_pgio_header *hdr, int sync) |
0382b744 | 513 | { |
cd841605 | 514 | struct pnfs_layout_segment *lseg = hdr->lseg; |
a69aef14 | 515 | struct nfs4_pnfs_ds *ds; |
0e20162e | 516 | struct rpc_clnt *ds_clnt; |
d45f60c6 | 517 | loff_t offset = hdr->args.offset; |
a69aef14 FI |
518 | u32 j, idx; |
519 | struct nfs_fh *fh; | |
a69aef14 FI |
520 | |
521 | /* Retrieve the correct rpc_client for the byte range */ | |
522 | j = nfs4_fl_calc_j_index(lseg, offset); | |
523 | idx = nfs4_fl_calc_ds_index(lseg, j); | |
524 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
90fecfcb | 525 | if (!ds) |
a69aef14 | 526 | return PNFS_NOT_ATTEMPTED; |
0e20162e AA |
527 | |
528 | ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode); | |
529 | if (IS_ERR(ds_clnt)) | |
530 | return PNFS_NOT_ATTEMPTED; | |
531 | ||
3a7936c3 | 532 | dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d\n", |
d45f60c6 | 533 | __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count, |
3a7936c3 | 534 | offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count)); |
a69aef14 | 535 | |
d45f60c6 | 536 | hdr->pgio_done_cb = filelayout_write_done_cb; |
3a7936c3 | 537 | atomic_inc(&ds->ds_clp->cl_count); |
d45f60c6 | 538 | hdr->ds_clp = ds->ds_clp; |
6cccbb6f | 539 | hdr->ds_commit_idx = idx; |
a69aef14 FI |
540 | fh = nfs4_fl_select_ds_fh(lseg, j); |
541 | if (fh) | |
d45f60c6 WAA |
542 | hdr->args.fh = fh; |
543 | hdr->args.offset = filelayout_get_dserver_offset(lseg, offset); | |
a69aef14 FI |
544 | |
545 | /* Perform an asynchronous write */ | |
46a5ab47 PT |
546 | nfs_initiate_pgio(ds_clnt, hdr, hdr->cred, |
547 | NFS_PROTO(hdr->inode), &filelayout_write_call_ops, | |
548 | sync, RPC_TASK_SOFTCONN); | |
a69aef14 | 549 | return PNFS_ATTEMPTED; |
0382b744 AA |
550 | } |
551 | ||
16b374ca AA |
552 | /* |
553 | * filelayout_check_layout() | |
554 | * | |
555 | * Make sure layout segment parameters are sane WRT the device. | |
556 | * At this point no generic layer initialization of the lseg has occurred, | |
557 | * and nothing has been added to the layout_hdr cache. | |
558 | * | |
559 | */ | |
560 | static int | |
561 | filelayout_check_layout(struct pnfs_layout_hdr *lo, | |
562 | struct nfs4_filelayout_segment *fl, | |
563 | struct nfs4_layoutget_res *lgr, | |
a75b9df9 TM |
564 | struct nfs4_deviceid *id, |
565 | gfp_t gfp_flags) | |
16b374ca | 566 | { |
a1eaecbc | 567 | struct nfs4_deviceid_node *d; |
16b374ca AA |
568 | struct nfs4_file_layout_dsaddr *dsaddr; |
569 | int status = -EINVAL; | |
16b374ca AA |
570 | |
571 | dprintk("--> %s\n", __func__); | |
572 | ||
7c24d948 AA |
573 | /* FIXME: remove this check when layout segment support is added */ |
574 | if (lgr->range.offset != 0 || | |
575 | lgr->range.length != NFS4_MAX_UINT64) { | |
576 | dprintk("%s Only whole file layouts supported. Use MDS i/o\n", | |
577 | __func__); | |
578 | goto out; | |
579 | } | |
580 | ||
16b374ca | 581 | if (fl->pattern_offset > lgr->range.offset) { |
3b6445a6 | 582 | dprintk("%s pattern_offset %lld too large\n", |
16b374ca AA |
583 | __func__, fl->pattern_offset); |
584 | goto out; | |
585 | } | |
586 | ||
c6194271 | 587 | if (!fl->stripe_unit) { |
75247aff | 588 | dprintk("%s Invalid stripe unit (%u)\n", |
16b374ca AA |
589 | __func__, fl->stripe_unit); |
590 | goto out; | |
591 | } | |
592 | ||
593 | /* find and reference the deviceid */ | |
661373b1 CH |
594 | d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), id, |
595 | lo->plh_lc_cred, gfp_flags); | |
596 | if (d == NULL) | |
597 | goto out; | |
598 | ||
599 | dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node); | |
1dfed273 TM |
600 | /* Found deviceid is unavailable */ |
601 | if (filelayout_test_devid_unavailable(&dsaddr->id_node)) | |
661373b1 | 602 | goto out_put; |
c47abcf8 | 603 | |
16b374ca AA |
604 | fl->dsaddr = dsaddr; |
605 | ||
e414966b CL |
606 | if (fl->first_stripe_index >= dsaddr->stripe_count) { |
607 | dprintk("%s Bad first_stripe_index %u\n", | |
16b374ca AA |
608 | __func__, fl->first_stripe_index); |
609 | goto out_put; | |
610 | } | |
611 | ||
612 | if ((fl->stripe_type == STRIPE_SPARSE && | |
613 | fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) || | |
614 | (fl->stripe_type == STRIPE_DENSE && | |
615 | fl->num_fh != dsaddr->stripe_count)) { | |
616 | dprintk("%s num_fh %u not valid for given packing\n", | |
617 | __func__, fl->num_fh); | |
618 | goto out_put; | |
619 | } | |
620 | ||
16b374ca AA |
621 | status = 0; |
622 | out: | |
623 | dprintk("--> %s returns %d\n", __func__, status); | |
624 | return status; | |
625 | out_put: | |
ea8eecdd | 626 | nfs4_fl_put_deviceid(dsaddr); |
16b374ca AA |
627 | goto out; |
628 | } | |
629 | ||
630 | static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl) | |
631 | { | |
632 | int i; | |
633 | ||
634 | for (i = 0; i < fl->num_fh; i++) { | |
635 | if (!fl->fh_array[i]) | |
636 | break; | |
637 | kfree(fl->fh_array[i]); | |
638 | } | |
639 | kfree(fl->fh_array); | |
640 | fl->fh_array = NULL; | |
641 | } | |
642 | ||
643 | static void | |
644 | _filelayout_free_lseg(struct nfs4_filelayout_segment *fl) | |
645 | { | |
646 | filelayout_free_fh_array(fl); | |
647 | kfree(fl); | |
648 | } | |
649 | ||
650 | static int | |
651 | filelayout_decode_layout(struct pnfs_layout_hdr *flo, | |
652 | struct nfs4_filelayout_segment *fl, | |
653 | struct nfs4_layoutget_res *lgr, | |
a75b9df9 TM |
654 | struct nfs4_deviceid *id, |
655 | gfp_t gfp_flags) | |
16b374ca | 656 | { |
35124a09 | 657 | struct xdr_stream stream; |
f7da7a12 | 658 | struct xdr_buf buf; |
35124a09 WAA |
659 | struct page *scratch; |
660 | __be32 *p; | |
16b374ca AA |
661 | uint32_t nfl_util; |
662 | int i; | |
663 | ||
664 | dprintk("%s: set_layout_map Begin\n", __func__); | |
665 | ||
a75b9df9 | 666 | scratch = alloc_page(gfp_flags); |
35124a09 WAA |
667 | if (!scratch) |
668 | return -ENOMEM; | |
669 | ||
f7da7a12 | 670 | xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len); |
35124a09 WAA |
671 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); |
672 | ||
673 | /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8), | |
674 | * num_fh (4) */ | |
675 | p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20); | |
676 | if (unlikely(!p)) | |
677 | goto out_err; | |
678 | ||
16b374ca AA |
679 | memcpy(id, p, sizeof(*id)); |
680 | p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE); | |
a1eaecbc | 681 | nfs4_print_deviceid(id); |
16b374ca AA |
682 | |
683 | nfl_util = be32_to_cpup(p++); | |
684 | if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS) | |
685 | fl->commit_through_mds = 1; | |
686 | if (nfl_util & NFL4_UFLG_DENSE) | |
687 | fl->stripe_type = STRIPE_DENSE; | |
688 | else | |
689 | fl->stripe_type = STRIPE_SPARSE; | |
690 | fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK; | |
691 | ||
692 | fl->first_stripe_index = be32_to_cpup(p++); | |
693 | p = xdr_decode_hyper(p, &fl->pattern_offset); | |
694 | fl->num_fh = be32_to_cpup(p++); | |
695 | ||
696 | dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n", | |
697 | __func__, nfl_util, fl->num_fh, fl->first_stripe_index, | |
698 | fl->pattern_offset); | |
699 | ||
cec765cf AA |
700 | /* Note that a zero value for num_fh is legal for STRIPE_SPARSE. |
701 | * Futher checking is done in filelayout_check_layout */ | |
e414966b | 702 | if (fl->num_fh > |
cec765cf | 703 | max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT)) |
35124a09 WAA |
704 | goto out_err; |
705 | ||
cec765cf | 706 | if (fl->num_fh > 0) { |
1813badd | 707 | fl->fh_array = kcalloc(fl->num_fh, sizeof(fl->fh_array[0]), |
cec765cf AA |
708 | gfp_flags); |
709 | if (!fl->fh_array) | |
710 | goto out_err; | |
711 | } | |
16b374ca AA |
712 | |
713 | for (i = 0; i < fl->num_fh; i++) { | |
714 | /* Do we want to use a mempool here? */ | |
a75b9df9 | 715 | fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags); |
35124a09 WAA |
716 | if (!fl->fh_array[i]) |
717 | goto out_err_free; | |
718 | ||
719 | p = xdr_inline_decode(&stream, 4); | |
720 | if (unlikely(!p)) | |
721 | goto out_err_free; | |
16b374ca AA |
722 | fl->fh_array[i]->size = be32_to_cpup(p++); |
723 | if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) { | |
f9fd2d9c | 724 | printk(KERN_ERR "NFS: Too big fh %d received %d\n", |
16b374ca | 725 | i, fl->fh_array[i]->size); |
35124a09 | 726 | goto out_err_free; |
16b374ca | 727 | } |
35124a09 WAA |
728 | |
729 | p = xdr_inline_decode(&stream, fl->fh_array[i]->size); | |
730 | if (unlikely(!p)) | |
731 | goto out_err_free; | |
16b374ca | 732 | memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size); |
16b374ca AA |
733 | dprintk("DEBUG: %s: fh len %d\n", __func__, |
734 | fl->fh_array[i]->size); | |
735 | } | |
736 | ||
35124a09 | 737 | __free_page(scratch); |
16b374ca | 738 | return 0; |
35124a09 WAA |
739 | |
740 | out_err_free: | |
741 | filelayout_free_fh_array(fl); | |
742 | out_err: | |
743 | __free_page(scratch); | |
744 | return -EIO; | |
16b374ca AA |
745 | } |
746 | ||
c879513e FI |
747 | static void |
748 | filelayout_free_lseg(struct pnfs_layout_segment *lseg) | |
749 | { | |
750 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); | |
751 | ||
752 | dprintk("--> %s\n", __func__); | |
753 | nfs4_fl_put_deviceid(fl->dsaddr); | |
799ba8d5 FI |
754 | /* This assumes a single RW lseg */ |
755 | if (lseg->pls_range.iomode == IOMODE_RW) { | |
756 | struct nfs4_filelayout *flo; | |
757 | ||
758 | flo = FILELAYOUT_FROM_HDR(lseg->pls_layout); | |
759 | flo->commit_info.nbuckets = 0; | |
760 | kfree(flo->commit_info.buckets); | |
761 | flo->commit_info.buckets = NULL; | |
762 | } | |
c879513e FI |
763 | _filelayout_free_lseg(fl); |
764 | } | |
765 | ||
799ba8d5 FI |
766 | static int |
767 | filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg, | |
ea2cf228 | 768 | struct nfs_commit_info *cinfo, |
799ba8d5 FI |
769 | gfp_t gfp_flags) |
770 | { | |
771 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); | |
ea2cf228 | 772 | struct pnfs_commit_bucket *buckets; |
dd7663e7 | 773 | int size, i; |
799ba8d5 FI |
774 | |
775 | if (fl->commit_through_mds) | |
776 | return 0; | |
dd7663e7 WAA |
777 | |
778 | size = (fl->stripe_type == STRIPE_SPARSE) ? | |
779 | fl->dsaddr->ds_num : fl->dsaddr->stripe_count; | |
780 | ||
781 | if (cinfo->ds->nbuckets >= size) { | |
799ba8d5 FI |
782 | /* This assumes there is only one IOMODE_RW lseg. What |
783 | * we really want to do is have a layout_hdr level | |
784 | * dictionary of <multipath_list4, fh> keys, each | |
785 | * associated with a struct list_head, populated by calls | |
786 | * to filelayout_write_pagelist(). | |
787 | * */ | |
788 | return 0; | |
789 | } | |
790 | ||
ea2cf228 | 791 | buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket), |
799ba8d5 FI |
792 | gfp_flags); |
793 | if (!buckets) | |
794 | return -ENOMEM; | |
dd7663e7 WAA |
795 | for (i = 0; i < size; i++) { |
796 | INIT_LIST_HEAD(&buckets[i].written); | |
797 | INIT_LIST_HEAD(&buckets[i].committing); | |
5002c586 WAA |
798 | /* mark direct verifier as unset */ |
799 | buckets[i].direct_verf.committed = NFS_INVALID_STABLE_HOW; | |
799ba8d5 | 800 | } |
dd7663e7 WAA |
801 | |
802 | spin_lock(cinfo->lock); | |
803 | if (cinfo->ds->nbuckets >= size) | |
804 | goto out; | |
805 | for (i = 0; i < cinfo->ds->nbuckets; i++) { | |
806 | list_splice(&cinfo->ds->buckets[i].written, | |
807 | &buckets[i].written); | |
808 | list_splice(&cinfo->ds->buckets[i].committing, | |
809 | &buckets[i].committing); | |
5002c586 WAA |
810 | buckets[i].direct_verf.committed = |
811 | cinfo->ds->buckets[i].direct_verf.committed; | |
dd7663e7 WAA |
812 | buckets[i].wlseg = cinfo->ds->buckets[i].wlseg; |
813 | buckets[i].clseg = cinfo->ds->buckets[i].clseg; | |
814 | } | |
815 | swap(cinfo->ds->buckets, buckets); | |
816 | cinfo->ds->nbuckets = size; | |
817 | out: | |
818 | spin_unlock(cinfo->lock); | |
819 | kfree(buckets); | |
820 | return 0; | |
799ba8d5 FI |
821 | } |
822 | ||
16b374ca AA |
823 | static struct pnfs_layout_segment * |
824 | filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid, | |
a75b9df9 TM |
825 | struct nfs4_layoutget_res *lgr, |
826 | gfp_t gfp_flags) | |
16b374ca AA |
827 | { |
828 | struct nfs4_filelayout_segment *fl; | |
829 | int rc; | |
830 | struct nfs4_deviceid id; | |
831 | ||
832 | dprintk("--> %s\n", __func__); | |
a75b9df9 | 833 | fl = kzalloc(sizeof(*fl), gfp_flags); |
16b374ca AA |
834 | if (!fl) |
835 | return NULL; | |
836 | ||
a75b9df9 TM |
837 | rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags); |
838 | if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) { | |
16b374ca AA |
839 | _filelayout_free_lseg(fl); |
840 | return NULL; | |
841 | } | |
842 | return &fl->generic_hdr; | |
843 | } | |
844 | ||
94ad1c80 FI |
845 | /* |
846 | * filelayout_pg_test(). Called by nfs_can_coalesce_requests() | |
847 | * | |
b4fdac1a WAA |
848 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number |
849 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
94ad1c80 | 850 | */ |
b4fdac1a | 851 | static size_t |
94ad1c80 FI |
852 | filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
853 | struct nfs_page *req) | |
854 | { | |
0f9c429e | 855 | unsigned int size; |
94ad1c80 | 856 | u64 p_stripe, r_stripe; |
c6194271 WAA |
857 | u32 stripe_offset; |
858 | u64 segment_offset = pgio->pg_lseg->pls_range.offset; | |
859 | u32 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit; | |
94ad1c80 | 860 | |
0f9c429e WAA |
861 | /* calls nfs_generic_pg_test */ |
862 | size = pnfs_generic_pg_test(pgio, prev, req); | |
863 | if (!size) | |
b4fdac1a | 864 | return 0; |
89a58e32 | 865 | |
c6194271 | 866 | /* see if req and prev are in the same stripe */ |
0f9c429e | 867 | if (prev) { |
c6194271 WAA |
868 | p_stripe = (u64)req_offset(prev) - segment_offset; |
869 | r_stripe = (u64)req_offset(req) - segment_offset; | |
0f9c429e WAA |
870 | do_div(p_stripe, stripe_unit); |
871 | do_div(r_stripe, stripe_unit); | |
94ad1c80 | 872 | |
0f9c429e WAA |
873 | if (p_stripe != r_stripe) |
874 | return 0; | |
875 | } | |
c6194271 WAA |
876 | |
877 | /* calculate remaining bytes in the current stripe */ | |
878 | div_u64_rem((u64)req_offset(req) - segment_offset, | |
879 | stripe_unit, | |
880 | &stripe_offset); | |
881 | WARN_ON_ONCE(stripe_offset > stripe_unit); | |
882 | if (stripe_offset >= stripe_unit) | |
883 | return 0; | |
884 | return min(stripe_unit - (unsigned int)stripe_offset, size); | |
94ad1c80 FI |
885 | } |
886 | ||
17280175 | 887 | static void |
7c24d948 AA |
888 | filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio, |
889 | struct nfs_page *req) | |
890 | { | |
c6194271 WAA |
891 | if (!pgio->pg_lseg) |
892 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
7c24d948 AA |
893 | req->wb_context, |
894 | 0, | |
895 | NFS4_MAX_UINT64, | |
896 | IOMODE_READ, | |
897 | GFP_KERNEL); | |
898 | /* If no lseg, fall back to read through mds */ | |
899 | if (pgio->pg_lseg == NULL) | |
1f945357 | 900 | nfs_pageio_reset_read_mds(pgio); |
7c24d948 AA |
901 | } |
902 | ||
17280175 | 903 | static void |
7c24d948 AA |
904 | filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio, |
905 | struct nfs_page *req) | |
906 | { | |
ea2cf228 | 907 | struct nfs_commit_info cinfo; |
799ba8d5 FI |
908 | int status; |
909 | ||
c6194271 WAA |
910 | if (!pgio->pg_lseg) |
911 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
7c24d948 AA |
912 | req->wb_context, |
913 | 0, | |
914 | NFS4_MAX_UINT64, | |
915 | IOMODE_RW, | |
916 | GFP_NOFS); | |
917 | /* If no lseg, fall back to write through mds */ | |
918 | if (pgio->pg_lseg == NULL) | |
799ba8d5 | 919 | goto out_mds; |
ea2cf228 FI |
920 | nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq); |
921 | status = filelayout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS); | |
799ba8d5 | 922 | if (status < 0) { |
9369a431 | 923 | pnfs_put_lseg(pgio->pg_lseg); |
799ba8d5 FI |
924 | pgio->pg_lseg = NULL; |
925 | goto out_mds; | |
926 | } | |
927 | return; | |
928 | out_mds: | |
929 | nfs_pageio_reset_write_mds(pgio); | |
7c24d948 AA |
930 | } |
931 | ||
1751c363 | 932 | static const struct nfs_pageio_ops filelayout_pg_read_ops = { |
7c24d948 | 933 | .pg_init = filelayout_pg_init_read, |
1751c363 | 934 | .pg_test = filelayout_pg_test, |
493292dd | 935 | .pg_doio = pnfs_generic_pg_readpages, |
180bb5ec | 936 | .pg_cleanup = pnfs_generic_pg_cleanup, |
1751c363 TM |
937 | }; |
938 | ||
939 | static const struct nfs_pageio_ops filelayout_pg_write_ops = { | |
7c24d948 | 940 | .pg_init = filelayout_pg_init_write, |
1751c363 | 941 | .pg_test = filelayout_pg_test, |
dce81290 | 942 | .pg_doio = pnfs_generic_pg_writepages, |
180bb5ec | 943 | .pg_cleanup = pnfs_generic_pg_cleanup, |
1751c363 TM |
944 | }; |
945 | ||
e0c2b380 FI |
946 | static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j) |
947 | { | |
948 | if (fl->stripe_type == STRIPE_SPARSE) | |
949 | return nfs4_fl_calc_ds_index(&fl->generic_hdr, j); | |
950 | else | |
951 | return j; | |
952 | } | |
953 | ||
c8a3292d PT |
954 | static void |
955 | filelayout_mark_request_commit(struct nfs_page *req, | |
956 | struct pnfs_layout_segment *lseg, | |
b57ff130 WAA |
957 | struct nfs_commit_info *cinfo, |
958 | u32 ds_commit_idx) | |
c8a3292d | 959 | |
e0c2b380 | 960 | { |
e0c2b380 FI |
961 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); |
962 | u32 i, j; | |
e0c2b380 | 963 | |
c8a3292d | 964 | if (fl->commit_through_mds) { |
338d00cf TH |
965 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
966 | } else { | |
967 | /* Note that we are calling nfs4_fl_calc_j_index on each page | |
968 | * that ends up being committed to a data server. An attractive | |
969 | * alternative is to add a field to nfs_write_data and nfs_page | |
970 | * to store the value calculated in filelayout_write_pagelist | |
971 | * and just use that here. | |
d6d6dc7c | 972 | */ |
338d00cf TH |
973 | j = nfs4_fl_calc_j_index(lseg, req_offset(req)); |
974 | i = select_bucket_index(fl, j); | |
975 | pnfs_layout_mark_request_commit(req, lseg, cinfo, i); | |
e0c2b380 | 976 | } |
8dd37758 TM |
977 | } |
978 | ||
e0c2b380 FI |
979 | static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i) |
980 | { | |
981 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
982 | ||
983 | if (flseg->stripe_type == STRIPE_SPARSE) | |
984 | return i; | |
985 | else | |
986 | return nfs4_fl_calc_ds_index(lseg, i); | |
987 | } | |
988 | ||
989 | static struct nfs_fh * | |
990 | select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i) | |
991 | { | |
992 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
993 | ||
994 | if (flseg->stripe_type == STRIPE_SPARSE) { | |
995 | if (flseg->num_fh == 1) | |
996 | i = 0; | |
997 | else if (flseg->num_fh == 0) | |
998 | /* Use the MDS OPEN fh set in nfs_read_rpcsetup */ | |
999 | return NULL; | |
1000 | } | |
1001 | return flseg->fh_array[i]; | |
1002 | } | |
1003 | ||
0b7c0153 | 1004 | static int filelayout_initiate_commit(struct nfs_commit_data *data, int how) |
e0c2b380 FI |
1005 | { |
1006 | struct pnfs_layout_segment *lseg = data->lseg; | |
1007 | struct nfs4_pnfs_ds *ds; | |
0e20162e | 1008 | struct rpc_clnt *ds_clnt; |
e0c2b380 FI |
1009 | u32 idx; |
1010 | struct nfs_fh *fh; | |
1011 | ||
1012 | idx = calc_ds_index_from_commit(lseg, data->ds_commit_index); | |
1013 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
0e20162e AA |
1014 | if (!ds) |
1015 | goto out_err; | |
1016 | ||
1017 | ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, data->inode); | |
1018 | if (IS_ERR(ds_clnt)) | |
1019 | goto out_err; | |
1020 | ||
3a7936c3 AA |
1021 | dprintk("%s ino %lu, how %d cl_count %d\n", __func__, |
1022 | data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count)); | |
0b7c0153 | 1023 | data->commit_done_cb = filelayout_commit_done_cb; |
3a7936c3 | 1024 | atomic_inc(&ds->ds_clp->cl_count); |
e0c2b380 FI |
1025 | data->ds_clp = ds->ds_clp; |
1026 | fh = select_ds_fh_from_commit(lseg, data->ds_commit_index); | |
1027 | if (fh) | |
1028 | data->args.fh = fh; | |
c36aae9a | 1029 | return nfs_initiate_commit(ds_clnt, data, NFS_PROTO(data->inode), |
9f0ec176 AA |
1030 | &filelayout_commit_call_ops, how, |
1031 | RPC_TASK_SOFTCONN); | |
0e20162e | 1032 | out_err: |
f54bcf2e TH |
1033 | pnfs_generic_prepare_to_resend_writes(data); |
1034 | pnfs_generic_commit_release(data); | |
0e20162e | 1035 | return -EAGAIN; |
e0c2b380 FI |
1036 | } |
1037 | ||
02d1426c WAA |
1038 | /* filelayout_search_commit_reqs - Search lists in @cinfo for the head reqest |
1039 | * for @page | |
1040 | * @cinfo - commit info for current inode | |
1041 | * @page - page to search for matching head request | |
1042 | * | |
1043 | * Returns a the head request if one is found, otherwise returns NULL. | |
1044 | */ | |
1045 | static struct nfs_page * | |
1046 | filelayout_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page) | |
1047 | { | |
1048 | struct nfs_page *freq, *t; | |
1049 | struct pnfs_commit_bucket *b; | |
1050 | int i; | |
1051 | ||
1052 | /* Linearly search the commit lists for each bucket until a matching | |
1053 | * request is found */ | |
1054 | for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) { | |
1055 | list_for_each_entry_safe(freq, t, &b->written, wb_list) { | |
1056 | if (freq->wb_page == page) | |
1057 | return freq->wb_head; | |
1058 | } | |
1059 | list_for_each_entry_safe(freq, t, &b->committing, wb_list) { | |
1060 | if (freq->wb_page == page) | |
1061 | return freq->wb_head; | |
1062 | } | |
1063 | } | |
1064 | ||
1065 | return NULL; | |
1066 | } | |
1067 | ||
e0c2b380 FI |
1068 | static int |
1069 | filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, | |
ea2cf228 | 1070 | int how, struct nfs_commit_info *cinfo) |
e0c2b380 | 1071 | { |
f54bcf2e TH |
1072 | return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo, |
1073 | filelayout_initiate_commit); | |
e0c2b380 | 1074 | } |
f54bcf2e | 1075 | |
661373b1 CH |
1076 | static struct nfs4_deviceid_node * |
1077 | filelayout_alloc_deviceid_node(struct nfs_server *server, | |
1078 | struct pnfs_device *pdev, gfp_t gfp_flags) | |
1079 | { | |
1080 | struct nfs4_file_layout_dsaddr *dsaddr; | |
1081 | ||
1082 | dsaddr = nfs4_fl_alloc_deviceid_node(server, pdev, gfp_flags); | |
1083 | if (!dsaddr) | |
1084 | return NULL; | |
1085 | return &dsaddr->id_node; | |
1086 | } | |
e0c2b380 | 1087 | |
1775bc34 BH |
1088 | static void |
1089 | filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d) | |
1090 | { | |
1091 | nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node)); | |
1092 | } | |
1093 | ||
799ba8d5 FI |
1094 | static struct pnfs_layout_hdr * |
1095 | filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags) | |
1096 | { | |
1097 | struct nfs4_filelayout *flo; | |
1098 | ||
1099 | flo = kzalloc(sizeof(*flo), gfp_flags); | |
6df200f5 | 1100 | return flo != NULL ? &flo->generic_hdr : NULL; |
799ba8d5 FI |
1101 | } |
1102 | ||
1103 | static void | |
1104 | filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
1105 | { | |
1106 | kfree(FILELAYOUT_FROM_HDR(lo)); | |
1107 | } | |
1108 | ||
ea2cf228 FI |
1109 | static struct pnfs_ds_commit_info * |
1110 | filelayout_get_ds_info(struct inode *inode) | |
1111 | { | |
df011748 FI |
1112 | struct pnfs_layout_hdr *layout = NFS_I(inode)->layout; |
1113 | ||
1114 | if (layout == NULL) | |
1115 | return NULL; | |
1116 | else | |
1117 | return &FILELAYOUT_FROM_HDR(layout)->commit_info; | |
ea2cf228 FI |
1118 | } |
1119 | ||
7ab672ce | 1120 | static struct pnfs_layoutdriver_type filelayout_type = { |
ea8eecdd CH |
1121 | .id = LAYOUT_NFSV4_1_FILES, |
1122 | .name = "LAYOUT_NFSV4_1_FILES", | |
1123 | .owner = THIS_MODULE, | |
799ba8d5 FI |
1124 | .alloc_layout_hdr = filelayout_alloc_layout_hdr, |
1125 | .free_layout_hdr = filelayout_free_layout_hdr, | |
ea8eecdd CH |
1126 | .alloc_lseg = filelayout_alloc_lseg, |
1127 | .free_lseg = filelayout_free_lseg, | |
1751c363 TM |
1128 | .pg_read_ops = &filelayout_pg_read_ops, |
1129 | .pg_write_ops = &filelayout_pg_write_ops, | |
ea2cf228 | 1130 | .get_ds_info = &filelayout_get_ds_info, |
8dd37758 | 1131 | .mark_request_commit = filelayout_mark_request_commit, |
f54bcf2e TH |
1132 | .clear_request_commit = pnfs_generic_clear_request_commit, |
1133 | .scan_commit_lists = pnfs_generic_scan_commit_lists, | |
1134 | .recover_commit_reqs = pnfs_generic_recover_commit_reqs, | |
02d1426c | 1135 | .search_commit_reqs = filelayout_search_commit_reqs, |
e0c2b380 | 1136 | .commit_pagelist = filelayout_commit_pagelist, |
dc70d7b3 | 1137 | .read_pagelist = filelayout_read_pagelist, |
0382b744 | 1138 | .write_pagelist = filelayout_write_pagelist, |
661373b1 | 1139 | .alloc_deviceid_node = filelayout_alloc_deviceid_node, |
1775bc34 | 1140 | .free_deviceid_node = filelayout_free_deveiceid_node, |
7ab672ce DH |
1141 | }; |
1142 | ||
1143 | static int __init nfs4filelayout_init(void) | |
1144 | { | |
1145 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n", | |
1146 | __func__); | |
1147 | return pnfs_register_layoutdriver(&filelayout_type); | |
1148 | } | |
1149 | ||
1150 | static void __exit nfs4filelayout_exit(void) | |
1151 | { | |
1152 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n", | |
1153 | __func__); | |
1154 | pnfs_unregister_layoutdriver(&filelayout_type); | |
1155 | } | |
1156 | ||
f85ef69c BF |
1157 | MODULE_ALIAS("nfs-layouttype4-1"); |
1158 | ||
7ab672ce DH |
1159 | module_init(nfs4filelayout_init); |
1160 | module_exit(nfs4filelayout_exit); |