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 | ||
16b374ca | 38 | #include "internal.h" |
9cb81968 | 39 | #include "delegation.h" |
16b374ca | 40 | #include "nfs4filelayout.h" |
7ab672ce DH |
41 | |
42 | #define NFSDBG_FACILITY NFSDBG_PNFS_LD | |
43 | ||
44 | MODULE_LICENSE("GPL"); | |
45 | MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>"); | |
46 | MODULE_DESCRIPTION("The NFSv4 file layout driver"); | |
47 | ||
cbdabc7f AA |
48 | #define FILELAYOUT_POLL_RETRY_MAX (15*HZ) |
49 | ||
cfe7f412 FI |
50 | static loff_t |
51 | filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg, | |
52 | loff_t offset) | |
53 | { | |
54 | u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count; | |
3476f114 CM |
55 | u64 stripe_no; |
56 | u32 rem; | |
cfe7f412 FI |
57 | |
58 | offset -= flseg->pattern_offset; | |
3476f114 CM |
59 | stripe_no = div_u64(offset, stripe_width); |
60 | div_u64_rem(offset, flseg->stripe_unit, &rem); | |
cfe7f412 | 61 | |
3476f114 | 62 | return stripe_no * flseg->stripe_unit + rem; |
cfe7f412 FI |
63 | } |
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 | */ | |
69 | static loff_t | |
70 | filelayout_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 | ||
e7dd79af AA |
85 | static void filelayout_reset_write(struct nfs_write_data *data) |
86 | { | |
87 | struct nfs_pgio_header *hdr = data->header; | |
e7dd79af AA |
88 | 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, | |
497826af BS |
94 | hdr->inode->i_sb->s_id, |
95 | (long long)NFS_FILEID(hdr->inode), | |
e7dd79af AA |
96 | 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 | ||
105 | static void filelayout_reset_read(struct nfs_read_data *data) | |
106 | { | |
107 | struct nfs_pgio_header *hdr = data->header; | |
e7dd79af AA |
108 | 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, | |
497826af BS |
114 | hdr->inode->i_sb->s_id, |
115 | (long long)NFS_FILEID(hdr->inode), | |
e7dd79af AA |
116 | 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 | ||
cbdabc7f AA |
125 | static int filelayout_async_handle_error(struct rpc_task *task, |
126 | struct nfs4_state *state, | |
127 | struct nfs_client *clp, | |
e7dd79af | 128 | struct pnfs_layout_segment *lseg) |
cbdabc7f | 129 | { |
e7dd79af AA |
130 | 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); | |
9cb81968 | 133 | struct nfs_client *mds_client = mds_server->nfs_client; |
671fb896 | 134 | struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table; |
9cb81968 | 135 | |
cbdabc7f AA |
136 | if (task->tk_status >= 0) |
137 | return 0; | |
cbdabc7f AA |
138 | |
139 | switch (task->tk_status) { | |
9cb81968 AA |
140 | /* MDS state errors */ |
141 | case -NFS4ERR_DELEG_REVOKED: | |
142 | case -NFS4ERR_ADMIN_REVOKED: | |
143 | case -NFS4ERR_BAD_STATEID: | |
e7dd79af AA |
144 | if (state == NULL) |
145 | break; | |
2dc31756 | 146 | nfs_remove_bad_delegation(state->inode); |
9cb81968 | 147 | case -NFS4ERR_OPENMODE: |
e7dd79af AA |
148 | if (state == NULL) |
149 | break; | |
9cb81968 AA |
150 | nfs4_schedule_stateid_recovery(mds_server, state); |
151 | goto wait_on_recovery; | |
152 | case -NFS4ERR_EXPIRED: | |
e7dd79af AA |
153 | if (state != NULL) |
154 | nfs4_schedule_stateid_recovery(mds_server, state); | |
9cb81968 AA |
155 | nfs4_schedule_lease_recovery(mds_client); |
156 | goto wait_on_recovery; | |
157 | /* DS session errors */ | |
cbdabc7f AA |
158 | 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); | |
9f594791 | 168 | nfs4_schedule_session_recovery(clp->cl_session, task->tk_status); |
cbdabc7f AA |
169 | break; |
170 | case -NFS4ERR_DELAY: | |
171 | case -NFS4ERR_GRACE: | |
172 | case -EKEYEXPIRED: | |
173 | rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX); | |
174 | break; | |
a8a4ae3a AA |
175 | case -NFS4ERR_RETRY_UNCACHED_REP: |
176 | break; | |
041245c8 AA |
177 | /* 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, | |
d42e7873 AA |
194 | &NFS_I(inode)->layout->plh_flags); |
195 | pnfs_destroy_layout(NFS_I(inode)); | |
041245c8 AA |
196 | rpc_wake_up(&tbl->slot_tbl_waitq); |
197 | goto reset; | |
e7dd79af AA |
198 | /* 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__, | |
207 | task->tk_status); | |
0a57cdac | 208 | if (!filelayout_test_devid_invalid(devid)) |
d42e7873 | 209 | _pnfs_return_layout(inode); |
e7dd79af | 210 | filelayout_mark_devid_invalid(devid); |
671fb896 | 211 | rpc_wake_up(&tbl->slot_tbl_waitq); |
b4a2967e | 212 | nfs4_ds_disconnect(clp); |
e7dd79af | 213 | /* fall through */ |
cbdabc7f | 214 | default: |
041245c8 | 215 | reset: |
e7dd79af | 216 | dprintk("%s Retry through MDS. Error %d\n", __func__, |
cbdabc7f | 217 | task->tk_status); |
e7dd79af | 218 | return -NFS4ERR_RESET_TO_MDS; |
cbdabc7f | 219 | } |
9cb81968 | 220 | out: |
cbdabc7f AA |
221 | task->tk_status = 0; |
222 | return -EAGAIN; | |
9cb81968 AA |
223 | wait_on_recovery: |
224 | rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL); | |
225 | if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0) | |
226 | rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task); | |
227 | goto out; | |
cbdabc7f AA |
228 | } |
229 | ||
230 | /* NFS_PROTO call done callback routines */ | |
231 | ||
232 | static int filelayout_read_done_cb(struct rpc_task *task, | |
233 | struct nfs_read_data *data) | |
234 | { | |
e7dd79af AA |
235 | struct nfs_pgio_header *hdr = data->header; |
236 | int err; | |
cbdabc7f | 237 | |
e7dd79af AA |
238 | err = filelayout_async_handle_error(task, data->args.context->state, |
239 | data->ds_clp, hdr->lseg); | |
cbdabc7f | 240 | |
e7dd79af AA |
241 | switch (err) { |
242 | case -NFS4ERR_RESET_TO_MDS: | |
243 | filelayout_reset_read(data); | |
244 | return task->tk_status; | |
245 | case -EAGAIN: | |
d00c5d43 | 246 | rpc_restart_call_prepare(task); |
cbdabc7f AA |
247 | return -EAGAIN; |
248 | } | |
249 | ||
250 | return 0; | |
251 | } | |
252 | ||
863a3c6c AA |
253 | /* |
254 | * We reference the rpc_cred of the first WRITE that triggers the need for | |
255 | * a LAYOUTCOMMIT, and use it to send the layoutcommit compound. | |
256 | * rfc5661 is not clear about which credential should be used. | |
257 | */ | |
258 | static void | |
259 | filelayout_set_layoutcommit(struct nfs_write_data *wdata) | |
260 | { | |
cd841605 FI |
261 | struct nfs_pgio_header *hdr = wdata->header; |
262 | ||
263 | if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds || | |
863a3c6c AA |
264 | wdata->res.verf->committed == NFS_FILE_SYNC) |
265 | return; | |
266 | ||
267 | pnfs_set_layoutcommit(wdata); | |
cd841605 FI |
268 | dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino, |
269 | (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb); | |
863a3c6c AA |
270 | } |
271 | ||
dc70d7b3 AA |
272 | /* |
273 | * Call ops for the async read/write cases | |
274 | * In the case of dense layouts, the offset needs to be reset to its | |
275 | * original value. | |
276 | */ | |
277 | static void filelayout_read_prepare(struct rpc_task *task, void *data) | |
278 | { | |
cd12ae32 | 279 | struct nfs_read_data *rdata = data; |
dc70d7b3 | 280 | |
041245c8 | 281 | if (filelayout_reset_to_mds(rdata->header->lseg)) { |
0ad2f378 AA |
282 | dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid); |
283 | filelayout_reset_read(rdata); | |
284 | rpc_exit(task, 0); | |
285 | return; | |
286 | } | |
cbdabc7f AA |
287 | rdata->read_done_cb = filelayout_read_done_cb; |
288 | ||
dc70d7b3 AA |
289 | if (nfs41_setup_sequence(rdata->ds_clp->cl_session, |
290 | &rdata->args.seq_args, &rdata->res.seq_res, | |
9d12b216 | 291 | task)) |
dc70d7b3 AA |
292 | return; |
293 | ||
294 | rpc_call_start(task); | |
295 | } | |
296 | ||
297 | static void filelayout_read_call_done(struct rpc_task *task, void *data) | |
298 | { | |
cd12ae32 | 299 | struct nfs_read_data *rdata = data; |
dc70d7b3 AA |
300 | |
301 | dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status); | |
302 | ||
bd4aeffb AA |
303 | if (test_bit(NFS_IOHDR_REDO, &rdata->header->flags) && |
304 | task->tk_status == 0) | |
0ad2f378 AA |
305 | return; |
306 | ||
dc70d7b3 | 307 | /* Note this may cause RPC to be resent */ |
cd841605 | 308 | rdata->header->mds_ops->rpc_call_done(task, data); |
dc70d7b3 AA |
309 | } |
310 | ||
0a702195 WAA |
311 | static void filelayout_read_count_stats(struct rpc_task *task, void *data) |
312 | { | |
cd12ae32 | 313 | struct nfs_read_data *rdata = data; |
0a702195 | 314 | |
cd841605 | 315 | rpc_count_iostats(task, NFS_SERVER(rdata->header->inode)->client->cl_metrics); |
0a702195 WAA |
316 | } |
317 | ||
dc70d7b3 AA |
318 | static void filelayout_read_release(void *data) |
319 | { | |
cd12ae32 | 320 | struct nfs_read_data *rdata = data; |
dc70d7b3 | 321 | |
996074cb | 322 | nfs_put_client(rdata->ds_clp); |
cd841605 | 323 | rdata->header->mds_ops->rpc_release(data); |
dc70d7b3 AA |
324 | } |
325 | ||
a69aef14 FI |
326 | static int filelayout_write_done_cb(struct rpc_task *task, |
327 | struct nfs_write_data *data) | |
328 | { | |
e7dd79af AA |
329 | struct nfs_pgio_header *hdr = data->header; |
330 | int err; | |
331 | ||
332 | err = filelayout_async_handle_error(task, data->args.context->state, | |
333 | data->ds_clp, hdr->lseg); | |
334 | ||
335 | switch (err) { | |
336 | case -NFS4ERR_RESET_TO_MDS: | |
337 | filelayout_reset_write(data); | |
338 | return task->tk_status; | |
339 | case -EAGAIN: | |
d00c5d43 | 340 | rpc_restart_call_prepare(task); |
a69aef14 FI |
341 | return -EAGAIN; |
342 | } | |
343 | ||
863a3c6c | 344 | filelayout_set_layoutcommit(data); |
a69aef14 FI |
345 | return 0; |
346 | } | |
347 | ||
e0c2b380 | 348 | /* Fake up some data that will cause nfs_commit_release to retry the writes. */ |
0b7c0153 | 349 | static void prepare_to_resend_writes(struct nfs_commit_data *data) |
e0c2b380 FI |
350 | { |
351 | struct nfs_page *first = nfs_list_entry(data->pages.next); | |
352 | ||
353 | data->task.tk_status = 0; | |
354 | memcpy(data->verf.verifier, first->wb_verf.verifier, | |
355 | sizeof(first->wb_verf.verifier)); | |
356 | data->verf.verifier[0]++; /* ensure verifier mismatch */ | |
357 | } | |
358 | ||
359 | static int filelayout_commit_done_cb(struct rpc_task *task, | |
0b7c0153 | 360 | struct nfs_commit_data *data) |
e0c2b380 | 361 | { |
e7dd79af AA |
362 | int err; |
363 | ||
364 | err = filelayout_async_handle_error(task, NULL, data->ds_clp, | |
365 | data->lseg); | |
366 | ||
367 | switch (err) { | |
368 | case -NFS4ERR_RESET_TO_MDS: | |
369 | prepare_to_resend_writes(data); | |
370 | return -EAGAIN; | |
371 | case -EAGAIN: | |
372 | rpc_restart_call_prepare(task); | |
e0c2b380 FI |
373 | return -EAGAIN; |
374 | } | |
375 | ||
376 | return 0; | |
377 | } | |
378 | ||
a69aef14 FI |
379 | static void filelayout_write_prepare(struct rpc_task *task, void *data) |
380 | { | |
cd12ae32 | 381 | struct nfs_write_data *wdata = data; |
a69aef14 | 382 | |
041245c8 | 383 | if (filelayout_reset_to_mds(wdata->header->lseg)) { |
0ad2f378 AA |
384 | dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid); |
385 | filelayout_reset_write(wdata); | |
386 | rpc_exit(task, 0); | |
387 | return; | |
388 | } | |
a69aef14 FI |
389 | if (nfs41_setup_sequence(wdata->ds_clp->cl_session, |
390 | &wdata->args.seq_args, &wdata->res.seq_res, | |
9d12b216 | 391 | task)) |
a69aef14 FI |
392 | return; |
393 | ||
394 | rpc_call_start(task); | |
395 | } | |
396 | ||
397 | static void filelayout_write_call_done(struct rpc_task *task, void *data) | |
398 | { | |
cd12ae32 | 399 | struct nfs_write_data *wdata = data; |
a69aef14 | 400 | |
bd4aeffb AA |
401 | if (test_bit(NFS_IOHDR_REDO, &wdata->header->flags) && |
402 | task->tk_status == 0) | |
0ad2f378 AA |
403 | return; |
404 | ||
a69aef14 | 405 | /* Note this may cause RPC to be resent */ |
cd841605 | 406 | wdata->header->mds_ops->rpc_call_done(task, data); |
a69aef14 FI |
407 | } |
408 | ||
0a702195 WAA |
409 | static void filelayout_write_count_stats(struct rpc_task *task, void *data) |
410 | { | |
cd12ae32 | 411 | struct nfs_write_data *wdata = data; |
0a702195 | 412 | |
cd841605 | 413 | rpc_count_iostats(task, NFS_SERVER(wdata->header->inode)->client->cl_metrics); |
0a702195 WAA |
414 | } |
415 | ||
a69aef14 FI |
416 | static void filelayout_write_release(void *data) |
417 | { | |
cd12ae32 | 418 | struct nfs_write_data *wdata = data; |
a69aef14 | 419 | |
996074cb | 420 | nfs_put_client(wdata->ds_clp); |
cd841605 | 421 | wdata->header->mds_ops->rpc_release(data); |
a69aef14 FI |
422 | } |
423 | ||
0b7c0153 | 424 | static void filelayout_commit_prepare(struct rpc_task *task, void *data) |
e0c2b380 | 425 | { |
0b7c0153 | 426 | struct nfs_commit_data *wdata = data; |
e0c2b380 | 427 | |
0b7c0153 FI |
428 | if (nfs41_setup_sequence(wdata->ds_clp->cl_session, |
429 | &wdata->args.seq_args, &wdata->res.seq_res, | |
430 | task)) | |
431 | return; | |
432 | ||
433 | rpc_call_start(task); | |
434 | } | |
435 | ||
436 | static void filelayout_write_commit_done(struct rpc_task *task, void *data) | |
437 | { | |
438 | struct nfs_commit_data *wdata = data; | |
439 | ||
440 | /* Note this may cause RPC to be resent */ | |
441 | wdata->mds_ops->rpc_call_done(task, data); | |
442 | } | |
443 | ||
444 | static void filelayout_commit_count_stats(struct rpc_task *task, void *data) | |
445 | { | |
446 | struct nfs_commit_data *cdata = data; | |
447 | ||
448 | rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics); | |
449 | } | |
450 | ||
451 | static void filelayout_commit_release(void *calldata) | |
452 | { | |
453 | struct nfs_commit_data *data = calldata; | |
454 | ||
f453a54a | 455 | data->completion_ops->completion(data); |
0b7c0153 | 456 | put_lseg(data->lseg); |
3a7936c3 | 457 | nfs_put_client(data->ds_clp); |
0b7c0153 | 458 | nfs_commitdata_release(data); |
e0c2b380 FI |
459 | } |
460 | ||
17280175 | 461 | static const struct rpc_call_ops filelayout_read_call_ops = { |
dc70d7b3 AA |
462 | .rpc_call_prepare = filelayout_read_prepare, |
463 | .rpc_call_done = filelayout_read_call_done, | |
0a702195 | 464 | .rpc_count_stats = filelayout_read_count_stats, |
dc70d7b3 AA |
465 | .rpc_release = filelayout_read_release, |
466 | }; | |
467 | ||
17280175 | 468 | static const struct rpc_call_ops filelayout_write_call_ops = { |
a69aef14 FI |
469 | .rpc_call_prepare = filelayout_write_prepare, |
470 | .rpc_call_done = filelayout_write_call_done, | |
0a702195 | 471 | .rpc_count_stats = filelayout_write_count_stats, |
a69aef14 FI |
472 | .rpc_release = filelayout_write_release, |
473 | }; | |
474 | ||
17280175 | 475 | static const struct rpc_call_ops filelayout_commit_call_ops = { |
0b7c0153 FI |
476 | .rpc_call_prepare = filelayout_commit_prepare, |
477 | .rpc_call_done = filelayout_write_commit_done, | |
478 | .rpc_count_stats = filelayout_commit_count_stats, | |
e0c2b380 FI |
479 | .rpc_release = filelayout_commit_release, |
480 | }; | |
481 | ||
dc70d7b3 AA |
482 | static enum pnfs_try_status |
483 | filelayout_read_pagelist(struct nfs_read_data *data) | |
484 | { | |
cd841605 FI |
485 | struct nfs_pgio_header *hdr = data->header; |
486 | struct pnfs_layout_segment *lseg = hdr->lseg; | |
dc70d7b3 AA |
487 | struct nfs4_pnfs_ds *ds; |
488 | loff_t offset = data->args.offset; | |
489 | u32 j, idx; | |
490 | struct nfs_fh *fh; | |
491 | int status; | |
492 | ||
493 | dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n", | |
cd841605 | 494 | __func__, hdr->inode->i_ino, |
dc70d7b3 AA |
495 | data->args.pgbase, (size_t)data->args.count, offset); |
496 | ||
497 | /* Retrieve the correct rpc_client for the byte range */ | |
498 | j = nfs4_fl_calc_j_index(lseg, offset); | |
499 | idx = nfs4_fl_calc_ds_index(lseg, j); | |
500 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
90fecfcb | 501 | if (!ds) |
dc70d7b3 | 502 | return PNFS_NOT_ATTEMPTED; |
3a7936c3 AA |
503 | dprintk("%s USE DS: %s cl_count %d\n", __func__, |
504 | ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count)); | |
dc70d7b3 AA |
505 | |
506 | /* No multipath support. Use first DS */ | |
3a7936c3 | 507 | atomic_inc(&ds->ds_clp->cl_count); |
dc70d7b3 AA |
508 | data->ds_clp = ds->ds_clp; |
509 | fh = nfs4_fl_select_ds_fh(lseg, j); | |
510 | if (fh) | |
511 | data->args.fh = fh; | |
512 | ||
513 | data->args.offset = filelayout_get_dserver_offset(lseg, offset); | |
514 | data->mds_offset = offset; | |
515 | ||
516 | /* Perform an asynchronous read to ds */ | |
c5996c4e | 517 | status = nfs_initiate_read(ds->ds_clp->cl_rpcclient, data, |
9f0ec176 | 518 | &filelayout_read_call_ops, RPC_TASK_SOFTCONN); |
dc70d7b3 AA |
519 | BUG_ON(status != 0); |
520 | return PNFS_ATTEMPTED; | |
521 | } | |
522 | ||
a69aef14 | 523 | /* Perform async writes. */ |
0382b744 AA |
524 | static enum pnfs_try_status |
525 | filelayout_write_pagelist(struct nfs_write_data *data, int sync) | |
526 | { | |
cd841605 FI |
527 | struct nfs_pgio_header *hdr = data->header; |
528 | struct pnfs_layout_segment *lseg = hdr->lseg; | |
a69aef14 FI |
529 | struct nfs4_pnfs_ds *ds; |
530 | loff_t offset = data->args.offset; | |
531 | u32 j, idx; | |
532 | struct nfs_fh *fh; | |
533 | int status; | |
534 | ||
535 | /* Retrieve the correct rpc_client for the byte range */ | |
536 | j = nfs4_fl_calc_j_index(lseg, offset); | |
537 | idx = nfs4_fl_calc_ds_index(lseg, j); | |
538 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
90fecfcb | 539 | if (!ds) |
a69aef14 | 540 | return PNFS_NOT_ATTEMPTED; |
3a7936c3 AA |
541 | dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d\n", |
542 | __func__, hdr->inode->i_ino, sync, (size_t) data->args.count, | |
543 | offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count)); | |
a69aef14 | 544 | |
a69aef14 | 545 | data->write_done_cb = filelayout_write_done_cb; |
3a7936c3 | 546 | atomic_inc(&ds->ds_clp->cl_count); |
a69aef14 FI |
547 | data->ds_clp = ds->ds_clp; |
548 | fh = nfs4_fl_select_ds_fh(lseg, j); | |
549 | if (fh) | |
550 | data->args.fh = fh; | |
551 | /* | |
552 | * Get the file offset on the dserver. Set the write offset to | |
553 | * this offset and save the original offset. | |
554 | */ | |
555 | data->args.offset = filelayout_get_dserver_offset(lseg, offset); | |
a69aef14 FI |
556 | |
557 | /* Perform an asynchronous write */ | |
c5996c4e | 558 | status = nfs_initiate_write(ds->ds_clp->cl_rpcclient, data, |
9f0ec176 AA |
559 | &filelayout_write_call_ops, sync, |
560 | RPC_TASK_SOFTCONN); | |
a69aef14 FI |
561 | BUG_ON(status != 0); |
562 | return PNFS_ATTEMPTED; | |
0382b744 AA |
563 | } |
564 | ||
16b374ca AA |
565 | /* |
566 | * filelayout_check_layout() | |
567 | * | |
568 | * Make sure layout segment parameters are sane WRT the device. | |
569 | * At this point no generic layer initialization of the lseg has occurred, | |
570 | * and nothing has been added to the layout_hdr cache. | |
571 | * | |
572 | */ | |
573 | static int | |
574 | filelayout_check_layout(struct pnfs_layout_hdr *lo, | |
575 | struct nfs4_filelayout_segment *fl, | |
576 | struct nfs4_layoutget_res *lgr, | |
a75b9df9 TM |
577 | struct nfs4_deviceid *id, |
578 | gfp_t gfp_flags) | |
16b374ca | 579 | { |
a1eaecbc | 580 | struct nfs4_deviceid_node *d; |
16b374ca AA |
581 | struct nfs4_file_layout_dsaddr *dsaddr; |
582 | int status = -EINVAL; | |
b7edfaa1 | 583 | struct nfs_server *nfss = NFS_SERVER(lo->plh_inode); |
16b374ca AA |
584 | |
585 | dprintk("--> %s\n", __func__); | |
586 | ||
7c24d948 AA |
587 | /* FIXME: remove this check when layout segment support is added */ |
588 | if (lgr->range.offset != 0 || | |
589 | lgr->range.length != NFS4_MAX_UINT64) { | |
590 | dprintk("%s Only whole file layouts supported. Use MDS i/o\n", | |
591 | __func__); | |
592 | goto out; | |
593 | } | |
594 | ||
16b374ca | 595 | if (fl->pattern_offset > lgr->range.offset) { |
3b6445a6 | 596 | dprintk("%s pattern_offset %lld too large\n", |
16b374ca AA |
597 | __func__, fl->pattern_offset); |
598 | goto out; | |
599 | } | |
600 | ||
75247aff BH |
601 | if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) { |
602 | dprintk("%s Invalid stripe unit (%u)\n", | |
16b374ca AA |
603 | __func__, fl->stripe_unit); |
604 | goto out; | |
605 | } | |
606 | ||
607 | /* find and reference the deviceid */ | |
35c8bb54 BH |
608 | d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld, |
609 | NFS_SERVER(lo->plh_inode)->nfs_client, id); | |
a1eaecbc | 610 | if (d == NULL) { |
a75b9df9 | 611 | dsaddr = get_device_info(lo->plh_inode, id, gfp_flags); |
16b374ca AA |
612 | if (dsaddr == NULL) |
613 | goto out; | |
a1eaecbc BH |
614 | } else |
615 | dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node); | |
c47abcf8 AA |
616 | /* Found deviceid is being reaped */ |
617 | if (test_bit(NFS_DEVICEID_INVALID, &dsaddr->id_node.flags)) | |
618 | goto out_put; | |
619 | ||
16b374ca AA |
620 | fl->dsaddr = dsaddr; |
621 | ||
e414966b CL |
622 | if (fl->first_stripe_index >= dsaddr->stripe_count) { |
623 | dprintk("%s Bad first_stripe_index %u\n", | |
16b374ca AA |
624 | __func__, fl->first_stripe_index); |
625 | goto out_put; | |
626 | } | |
627 | ||
628 | if ((fl->stripe_type == STRIPE_SPARSE && | |
629 | fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) || | |
630 | (fl->stripe_type == STRIPE_DENSE && | |
631 | fl->num_fh != dsaddr->stripe_count)) { | |
632 | dprintk("%s num_fh %u not valid for given packing\n", | |
633 | __func__, fl->num_fh); | |
634 | goto out_put; | |
635 | } | |
636 | ||
637 | if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) { | |
638 | dprintk("%s Stripe unit (%u) not aligned with rsize %u " | |
639 | "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize, | |
640 | nfss->wsize); | |
641 | } | |
642 | ||
643 | status = 0; | |
644 | out: | |
645 | dprintk("--> %s returns %d\n", __func__, status); | |
646 | return status; | |
647 | out_put: | |
ea8eecdd | 648 | nfs4_fl_put_deviceid(dsaddr); |
16b374ca AA |
649 | goto out; |
650 | } | |
651 | ||
652 | static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl) | |
653 | { | |
654 | int i; | |
655 | ||
656 | for (i = 0; i < fl->num_fh; i++) { | |
657 | if (!fl->fh_array[i]) | |
658 | break; | |
659 | kfree(fl->fh_array[i]); | |
660 | } | |
661 | kfree(fl->fh_array); | |
662 | fl->fh_array = NULL; | |
663 | } | |
664 | ||
665 | static void | |
666 | _filelayout_free_lseg(struct nfs4_filelayout_segment *fl) | |
667 | { | |
668 | filelayout_free_fh_array(fl); | |
669 | kfree(fl); | |
670 | } | |
671 | ||
672 | static int | |
673 | filelayout_decode_layout(struct pnfs_layout_hdr *flo, | |
674 | struct nfs4_filelayout_segment *fl, | |
675 | struct nfs4_layoutget_res *lgr, | |
a75b9df9 TM |
676 | struct nfs4_deviceid *id, |
677 | gfp_t gfp_flags) | |
16b374ca | 678 | { |
35124a09 | 679 | struct xdr_stream stream; |
f7da7a12 | 680 | struct xdr_buf buf; |
35124a09 WAA |
681 | struct page *scratch; |
682 | __be32 *p; | |
16b374ca AA |
683 | uint32_t nfl_util; |
684 | int i; | |
685 | ||
686 | dprintk("%s: set_layout_map Begin\n", __func__); | |
687 | ||
a75b9df9 | 688 | scratch = alloc_page(gfp_flags); |
35124a09 WAA |
689 | if (!scratch) |
690 | return -ENOMEM; | |
691 | ||
f7da7a12 | 692 | xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len); |
35124a09 WAA |
693 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); |
694 | ||
695 | /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8), | |
696 | * num_fh (4) */ | |
697 | p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20); | |
698 | if (unlikely(!p)) | |
699 | goto out_err; | |
700 | ||
16b374ca AA |
701 | memcpy(id, p, sizeof(*id)); |
702 | p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE); | |
a1eaecbc | 703 | nfs4_print_deviceid(id); |
16b374ca AA |
704 | |
705 | nfl_util = be32_to_cpup(p++); | |
706 | if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS) | |
707 | fl->commit_through_mds = 1; | |
708 | if (nfl_util & NFL4_UFLG_DENSE) | |
709 | fl->stripe_type = STRIPE_DENSE; | |
710 | else | |
711 | fl->stripe_type = STRIPE_SPARSE; | |
712 | fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK; | |
713 | ||
714 | fl->first_stripe_index = be32_to_cpup(p++); | |
715 | p = xdr_decode_hyper(p, &fl->pattern_offset); | |
716 | fl->num_fh = be32_to_cpup(p++); | |
717 | ||
718 | dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n", | |
719 | __func__, nfl_util, fl->num_fh, fl->first_stripe_index, | |
720 | fl->pattern_offset); | |
721 | ||
cec765cf AA |
722 | /* Note that a zero value for num_fh is legal for STRIPE_SPARSE. |
723 | * Futher checking is done in filelayout_check_layout */ | |
e414966b | 724 | if (fl->num_fh > |
cec765cf | 725 | max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT)) |
35124a09 WAA |
726 | goto out_err; |
727 | ||
cec765cf AA |
728 | if (fl->num_fh > 0) { |
729 | fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *), | |
730 | gfp_flags); | |
731 | if (!fl->fh_array) | |
732 | goto out_err; | |
733 | } | |
16b374ca AA |
734 | |
735 | for (i = 0; i < fl->num_fh; i++) { | |
736 | /* Do we want to use a mempool here? */ | |
a75b9df9 | 737 | fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags); |
35124a09 WAA |
738 | if (!fl->fh_array[i]) |
739 | goto out_err_free; | |
740 | ||
741 | p = xdr_inline_decode(&stream, 4); | |
742 | if (unlikely(!p)) | |
743 | goto out_err_free; | |
16b374ca AA |
744 | fl->fh_array[i]->size = be32_to_cpup(p++); |
745 | if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) { | |
f9fd2d9c | 746 | printk(KERN_ERR "NFS: Too big fh %d received %d\n", |
16b374ca | 747 | i, fl->fh_array[i]->size); |
35124a09 | 748 | goto out_err_free; |
16b374ca | 749 | } |
35124a09 WAA |
750 | |
751 | p = xdr_inline_decode(&stream, fl->fh_array[i]->size); | |
752 | if (unlikely(!p)) | |
753 | goto out_err_free; | |
16b374ca | 754 | memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size); |
16b374ca AA |
755 | dprintk("DEBUG: %s: fh len %d\n", __func__, |
756 | fl->fh_array[i]->size); | |
757 | } | |
758 | ||
35124a09 | 759 | __free_page(scratch); |
16b374ca | 760 | return 0; |
35124a09 WAA |
761 | |
762 | out_err_free: | |
763 | filelayout_free_fh_array(fl); | |
764 | out_err: | |
765 | __free_page(scratch); | |
766 | return -EIO; | |
16b374ca AA |
767 | } |
768 | ||
c879513e FI |
769 | static void |
770 | filelayout_free_lseg(struct pnfs_layout_segment *lseg) | |
771 | { | |
772 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); | |
773 | ||
774 | dprintk("--> %s\n", __func__); | |
775 | nfs4_fl_put_deviceid(fl->dsaddr); | |
799ba8d5 FI |
776 | /* This assumes a single RW lseg */ |
777 | if (lseg->pls_range.iomode == IOMODE_RW) { | |
778 | struct nfs4_filelayout *flo; | |
779 | ||
780 | flo = FILELAYOUT_FROM_HDR(lseg->pls_layout); | |
781 | flo->commit_info.nbuckets = 0; | |
782 | kfree(flo->commit_info.buckets); | |
783 | flo->commit_info.buckets = NULL; | |
784 | } | |
c879513e FI |
785 | _filelayout_free_lseg(fl); |
786 | } | |
787 | ||
799ba8d5 FI |
788 | static int |
789 | filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg, | |
ea2cf228 | 790 | struct nfs_commit_info *cinfo, |
799ba8d5 FI |
791 | gfp_t gfp_flags) |
792 | { | |
793 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); | |
ea2cf228 | 794 | struct pnfs_commit_bucket *buckets; |
799ba8d5 FI |
795 | int size; |
796 | ||
797 | if (fl->commit_through_mds) | |
798 | return 0; | |
ea2cf228 | 799 | if (cinfo->ds->nbuckets != 0) { |
799ba8d5 FI |
800 | /* This assumes there is only one IOMODE_RW lseg. What |
801 | * we really want to do is have a layout_hdr level | |
802 | * dictionary of <multipath_list4, fh> keys, each | |
803 | * associated with a struct list_head, populated by calls | |
804 | * to filelayout_write_pagelist(). | |
805 | * */ | |
806 | return 0; | |
807 | } | |
808 | ||
809 | size = (fl->stripe_type == STRIPE_SPARSE) ? | |
810 | fl->dsaddr->ds_num : fl->dsaddr->stripe_count; | |
811 | ||
ea2cf228 | 812 | buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket), |
799ba8d5 FI |
813 | gfp_flags); |
814 | if (!buckets) | |
815 | return -ENOMEM; | |
816 | else { | |
817 | int i; | |
818 | ||
ea2cf228 FI |
819 | spin_lock(cinfo->lock); |
820 | if (cinfo->ds->nbuckets != 0) | |
799ba8d5 FI |
821 | kfree(buckets); |
822 | else { | |
ea2cf228 FI |
823 | cinfo->ds->buckets = buckets; |
824 | cinfo->ds->nbuckets = size; | |
799ba8d5 FI |
825 | for (i = 0; i < size; i++) { |
826 | INIT_LIST_HEAD(&buckets[i].written); | |
827 | INIT_LIST_HEAD(&buckets[i].committing); | |
828 | } | |
829 | } | |
ea2cf228 | 830 | spin_unlock(cinfo->lock); |
799ba8d5 FI |
831 | return 0; |
832 | } | |
833 | } | |
834 | ||
16b374ca AA |
835 | static struct pnfs_layout_segment * |
836 | filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid, | |
a75b9df9 TM |
837 | struct nfs4_layoutget_res *lgr, |
838 | gfp_t gfp_flags) | |
16b374ca AA |
839 | { |
840 | struct nfs4_filelayout_segment *fl; | |
841 | int rc; | |
842 | struct nfs4_deviceid id; | |
843 | ||
844 | dprintk("--> %s\n", __func__); | |
a75b9df9 | 845 | fl = kzalloc(sizeof(*fl), gfp_flags); |
16b374ca AA |
846 | if (!fl) |
847 | return NULL; | |
848 | ||
a75b9df9 TM |
849 | rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags); |
850 | if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) { | |
16b374ca AA |
851 | _filelayout_free_lseg(fl); |
852 | return NULL; | |
853 | } | |
854 | return &fl->generic_hdr; | |
855 | } | |
856 | ||
94ad1c80 FI |
857 | /* |
858 | * filelayout_pg_test(). Called by nfs_can_coalesce_requests() | |
859 | * | |
18ad0a9f BH |
860 | * return true : coalesce page |
861 | * return false : don't coalesce page | |
94ad1c80 | 862 | */ |
1751c363 | 863 | static bool |
94ad1c80 FI |
864 | filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
865 | struct nfs_page *req) | |
866 | { | |
867 | u64 p_stripe, r_stripe; | |
868 | u32 stripe_unit; | |
869 | ||
19345cb2 BH |
870 | if (!pnfs_generic_pg_test(pgio, prev, req) || |
871 | !nfs_generic_pg_test(pgio, prev, req)) | |
872 | return false; | |
89a58e32 | 873 | |
b5542849 FI |
874 | p_stripe = (u64)req_offset(prev); |
875 | r_stripe = (u64)req_offset(req); | |
94ad1c80 FI |
876 | stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit; |
877 | ||
878 | do_div(p_stripe, stripe_unit); | |
879 | do_div(r_stripe, stripe_unit); | |
880 | ||
881 | return (p_stripe == r_stripe); | |
882 | } | |
883 | ||
17280175 | 884 | static void |
7c24d948 AA |
885 | filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio, |
886 | struct nfs_page *req) | |
887 | { | |
888 | BUG_ON(pgio->pg_lseg != NULL); | |
889 | ||
1825a0d0 FI |
890 | if (req->wb_offset != req->wb_pgbase) { |
891 | /* | |
892 | * Handling unaligned pages is difficult, because have to | |
893 | * somehow split a req in two in certain cases in the | |
894 | * pg.test code. Avoid this by just not using pnfs | |
895 | * in this case. | |
896 | */ | |
897 | nfs_pageio_reset_read_mds(pgio); | |
898 | return; | |
899 | } | |
7c24d948 AA |
900 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
901 | req->wb_context, | |
902 | 0, | |
903 | NFS4_MAX_UINT64, | |
904 | IOMODE_READ, | |
905 | GFP_KERNEL); | |
906 | /* If no lseg, fall back to read through mds */ | |
907 | if (pgio->pg_lseg == NULL) | |
1f945357 | 908 | nfs_pageio_reset_read_mds(pgio); |
7c24d948 AA |
909 | } |
910 | ||
17280175 | 911 | static void |
7c24d948 AA |
912 | filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio, |
913 | struct nfs_page *req) | |
914 | { | |
ea2cf228 | 915 | struct nfs_commit_info cinfo; |
799ba8d5 FI |
916 | int status; |
917 | ||
7c24d948 AA |
918 | BUG_ON(pgio->pg_lseg != NULL); |
919 | ||
1825a0d0 FI |
920 | if (req->wb_offset != req->wb_pgbase) |
921 | goto out_mds; | |
7c24d948 AA |
922 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
923 | req->wb_context, | |
924 | 0, | |
925 | NFS4_MAX_UINT64, | |
926 | IOMODE_RW, | |
927 | GFP_NOFS); | |
928 | /* If no lseg, fall back to write through mds */ | |
929 | if (pgio->pg_lseg == NULL) | |
799ba8d5 | 930 | goto out_mds; |
ea2cf228 FI |
931 | nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq); |
932 | status = filelayout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS); | |
799ba8d5 FI |
933 | if (status < 0) { |
934 | put_lseg(pgio->pg_lseg); | |
935 | pgio->pg_lseg = NULL; | |
936 | goto out_mds; | |
937 | } | |
938 | return; | |
939 | out_mds: | |
940 | nfs_pageio_reset_write_mds(pgio); | |
7c24d948 AA |
941 | } |
942 | ||
1751c363 | 943 | static const struct nfs_pageio_ops filelayout_pg_read_ops = { |
7c24d948 | 944 | .pg_init = filelayout_pg_init_read, |
1751c363 | 945 | .pg_test = filelayout_pg_test, |
493292dd | 946 | .pg_doio = pnfs_generic_pg_readpages, |
1751c363 TM |
947 | }; |
948 | ||
949 | static const struct nfs_pageio_ops filelayout_pg_write_ops = { | |
7c24d948 | 950 | .pg_init = filelayout_pg_init_write, |
1751c363 | 951 | .pg_test = filelayout_pg_test, |
dce81290 | 952 | .pg_doio = pnfs_generic_pg_writepages, |
1751c363 TM |
953 | }; |
954 | ||
e0c2b380 FI |
955 | static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j) |
956 | { | |
957 | if (fl->stripe_type == STRIPE_SPARSE) | |
958 | return nfs4_fl_calc_ds_index(&fl->generic_hdr, j); | |
959 | else | |
960 | return j; | |
961 | } | |
962 | ||
d6d6dc7c FI |
963 | /* The generic layer is about to remove the req from the commit list. |
964 | * If this will make the bucket empty, it will need to put the lseg reference. | |
d6d6dc7c | 965 | */ |
8dd37758 | 966 | static void |
ea2cf228 FI |
967 | filelayout_clear_request_commit(struct nfs_page *req, |
968 | struct nfs_commit_info *cinfo) | |
d6d6dc7c | 969 | { |
8dd37758 | 970 | struct pnfs_layout_segment *freeme = NULL; |
8dd37758 | 971 | |
ea2cf228 | 972 | spin_lock(cinfo->lock); |
8dd37758 TM |
973 | if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags)) |
974 | goto out; | |
ea2cf228 | 975 | cinfo->ds->nwritten--; |
d6d6dc7c | 976 | if (list_is_singular(&req->wb_list)) { |
ea2cf228 | 977 | struct pnfs_commit_bucket *bucket; |
d6d6dc7c | 978 | |
799ba8d5 | 979 | bucket = list_first_entry(&req->wb_list, |
ea2cf228 | 980 | struct pnfs_commit_bucket, |
799ba8d5 FI |
981 | written); |
982 | freeme = bucket->wlseg; | |
983 | bucket->wlseg = NULL; | |
d6d6dc7c | 984 | } |
8dd37758 | 985 | out: |
ea2cf228 FI |
986 | nfs_request_remove_commit_list(req, cinfo); |
987 | spin_unlock(cinfo->lock); | |
8dd37758 | 988 | put_lseg(freeme); |
d6d6dc7c FI |
989 | } |
990 | ||
991 | static struct list_head * | |
992 | filelayout_choose_commit_list(struct nfs_page *req, | |
ea2cf228 FI |
993 | struct pnfs_layout_segment *lseg, |
994 | struct nfs_commit_info *cinfo) | |
e0c2b380 | 995 | { |
e0c2b380 FI |
996 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); |
997 | u32 i, j; | |
998 | struct list_head *list; | |
ea2cf228 | 999 | struct pnfs_commit_bucket *buckets; |
e0c2b380 | 1000 | |
d6d6dc7c | 1001 | if (fl->commit_through_mds) |
ea2cf228 | 1002 | return &cinfo->mds->list; |
d6d6dc7c | 1003 | |
e0c2b380 FI |
1004 | /* Note that we are calling nfs4_fl_calc_j_index on each page |
1005 | * that ends up being committed to a data server. An attractive | |
1006 | * alternative is to add a field to nfs_write_data and nfs_page | |
1007 | * to store the value calculated in filelayout_write_pagelist | |
1008 | * and just use that here. | |
1009 | */ | |
b5542849 | 1010 | j = nfs4_fl_calc_j_index(lseg, req_offset(req)); |
e0c2b380 | 1011 | i = select_bucket_index(fl, j); |
ea2cf228 | 1012 | buckets = cinfo->ds->buckets; |
799ba8d5 | 1013 | list = &buckets[i].written; |
e0c2b380 | 1014 | if (list_empty(list)) { |
d6d6dc7c FI |
1015 | /* Non-empty buckets hold a reference on the lseg. That ref |
1016 | * is normally transferred to the COMMIT call and released | |
1017 | * there. It could also be released if the last req is pulled | |
1018 | * off due to a rewrite, in which case it will be done in | |
799ba8d5 | 1019 | * filelayout_clear_request_commit |
d6d6dc7c | 1020 | */ |
799ba8d5 | 1021 | buckets[i].wlseg = get_lseg(lseg); |
e0c2b380 | 1022 | } |
8dd37758 | 1023 | set_bit(PG_COMMIT_TO_DS, &req->wb_flags); |
ea2cf228 | 1024 | cinfo->ds->nwritten++; |
e0c2b380 FI |
1025 | return list; |
1026 | } | |
1027 | ||
8dd37758 TM |
1028 | static void |
1029 | filelayout_mark_request_commit(struct nfs_page *req, | |
ea2cf228 FI |
1030 | struct pnfs_layout_segment *lseg, |
1031 | struct nfs_commit_info *cinfo) | |
8dd37758 TM |
1032 | { |
1033 | struct list_head *list; | |
1034 | ||
ea2cf228 FI |
1035 | list = filelayout_choose_commit_list(req, lseg, cinfo); |
1036 | nfs_request_add_commit_list(req, list, cinfo); | |
8dd37758 TM |
1037 | } |
1038 | ||
e0c2b380 FI |
1039 | static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i) |
1040 | { | |
1041 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
1042 | ||
1043 | if (flseg->stripe_type == STRIPE_SPARSE) | |
1044 | return i; | |
1045 | else | |
1046 | return nfs4_fl_calc_ds_index(lseg, i); | |
1047 | } | |
1048 | ||
1049 | static struct nfs_fh * | |
1050 | select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i) | |
1051 | { | |
1052 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
1053 | ||
1054 | if (flseg->stripe_type == STRIPE_SPARSE) { | |
1055 | if (flseg->num_fh == 1) | |
1056 | i = 0; | |
1057 | else if (flseg->num_fh == 0) | |
1058 | /* Use the MDS OPEN fh set in nfs_read_rpcsetup */ | |
1059 | return NULL; | |
1060 | } | |
1061 | return flseg->fh_array[i]; | |
1062 | } | |
1063 | ||
0b7c0153 | 1064 | static int filelayout_initiate_commit(struct nfs_commit_data *data, int how) |
e0c2b380 FI |
1065 | { |
1066 | struct pnfs_layout_segment *lseg = data->lseg; | |
1067 | struct nfs4_pnfs_ds *ds; | |
1068 | u32 idx; | |
1069 | struct nfs_fh *fh; | |
1070 | ||
1071 | idx = calc_ds_index_from_commit(lseg, data->ds_commit_index); | |
1072 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
1073 | if (!ds) { | |
e0c2b380 | 1074 | prepare_to_resend_writes(data); |
8dd37758 | 1075 | filelayout_commit_release(data); |
e0c2b380 FI |
1076 | return -EAGAIN; |
1077 | } | |
3a7936c3 AA |
1078 | dprintk("%s ino %lu, how %d cl_count %d\n", __func__, |
1079 | data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count)); | |
0b7c0153 | 1080 | data->commit_done_cb = filelayout_commit_done_cb; |
3a7936c3 | 1081 | atomic_inc(&ds->ds_clp->cl_count); |
e0c2b380 FI |
1082 | data->ds_clp = ds->ds_clp; |
1083 | fh = select_ds_fh_from_commit(lseg, data->ds_commit_index); | |
1084 | if (fh) | |
1085 | data->args.fh = fh; | |
0b7c0153 | 1086 | return nfs_initiate_commit(ds->ds_clp->cl_rpcclient, data, |
9f0ec176 AA |
1087 | &filelayout_commit_call_ops, how, |
1088 | RPC_TASK_SOFTCONN); | |
e0c2b380 FI |
1089 | } |
1090 | ||
8dd37758 | 1091 | static int |
1763da12 FI |
1092 | transfer_commit_list(struct list_head *src, struct list_head *dst, |
1093 | struct nfs_commit_info *cinfo, int max) | |
8dd37758 | 1094 | { |
8dd37758 TM |
1095 | struct nfs_page *req, *tmp; |
1096 | int ret = 0; | |
1097 | ||
1098 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
1099 | if (!nfs_lock_request(req)) | |
1100 | continue; | |
53b8ee34 | 1101 | kref_get(&req->wb_kref); |
ea2cf228 | 1102 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 1103 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 1104 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
1105 | clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); |
1106 | nfs_list_add_request(req, dst); | |
1107 | ret++; | |
1763da12 | 1108 | if ((ret == max) && !cinfo->dreq) |
8dd37758 TM |
1109 | break; |
1110 | } | |
1763da12 FI |
1111 | return ret; |
1112 | } | |
1113 | ||
1114 | static int | |
1115 | filelayout_scan_ds_commit_list(struct pnfs_commit_bucket *bucket, | |
1116 | struct nfs_commit_info *cinfo, | |
1117 | int max) | |
1118 | { | |
1119 | struct list_head *src = &bucket->written; | |
1120 | struct list_head *dst = &bucket->committing; | |
1121 | int ret; | |
1122 | ||
1123 | ret = transfer_commit_list(src, dst, cinfo, max); | |
799ba8d5 | 1124 | if (ret) { |
ea2cf228 FI |
1125 | cinfo->ds->nwritten -= ret; |
1126 | cinfo->ds->ncommitting += ret; | |
799ba8d5 FI |
1127 | bucket->clseg = bucket->wlseg; |
1128 | if (list_empty(src)) | |
1129 | bucket->wlseg = NULL; | |
1130 | else | |
1131 | get_lseg(bucket->clseg); | |
1132 | } | |
8dd37758 TM |
1133 | return ret; |
1134 | } | |
1135 | ||
d6d6dc7c | 1136 | /* Move reqs from written to committing lists, returning count of number moved. |
ea2cf228 | 1137 | * Note called with cinfo->lock held. |
d6d6dc7c | 1138 | */ |
ea2cf228 FI |
1139 | static int filelayout_scan_commit_lists(struct nfs_commit_info *cinfo, |
1140 | int max) | |
d6d6dc7c | 1141 | { |
d6d6dc7c FI |
1142 | int i, rv = 0, cnt; |
1143 | ||
ea2cf228 FI |
1144 | for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) { |
1145 | cnt = filelayout_scan_ds_commit_list(&cinfo->ds->buckets[i], | |
1146 | cinfo, max); | |
d6d6dc7c FI |
1147 | max -= cnt; |
1148 | rv += cnt; | |
1149 | } | |
d6d6dc7c FI |
1150 | return rv; |
1151 | } | |
1152 | ||
1763da12 FI |
1153 | /* Pull everything off the committing lists and dump into @dst */ |
1154 | static void filelayout_recover_commit_reqs(struct list_head *dst, | |
1155 | struct nfs_commit_info *cinfo) | |
1156 | { | |
1157 | struct pnfs_commit_bucket *b; | |
1158 | int i; | |
1159 | ||
1160 | /* NOTE cinfo->lock is NOT held, relying on fact that this is | |
1161 | * only called on single thread per dreq. | |
1162 | * Can't take the lock because need to do put_lseg | |
1163 | */ | |
1164 | for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) { | |
1165 | if (transfer_commit_list(&b->written, dst, cinfo, 0)) { | |
1166 | BUG_ON(!list_empty(&b->written)); | |
1167 | put_lseg(b->wlseg); | |
1168 | b->wlseg = NULL; | |
1169 | } | |
1170 | } | |
1171 | cinfo->ds->nwritten = 0; | |
1172 | } | |
1173 | ||
9390f425 | 1174 | static unsigned int |
ea2cf228 | 1175 | alloc_ds_commits(struct nfs_commit_info *cinfo, struct list_head *list) |
e0c2b380 | 1176 | { |
ea2cf228 FI |
1177 | struct pnfs_ds_commit_info *fl_cinfo; |
1178 | struct pnfs_commit_bucket *bucket; | |
0b7c0153 | 1179 | struct nfs_commit_data *data; |
e0c2b380 | 1180 | int i, j; |
9390f425 | 1181 | unsigned int nreq = 0; |
e0c2b380 | 1182 | |
ea2cf228 | 1183 | fl_cinfo = cinfo->ds; |
799ba8d5 FI |
1184 | bucket = fl_cinfo->buckets; |
1185 | for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) { | |
1186 | if (list_empty(&bucket->committing)) | |
e0c2b380 FI |
1187 | continue; |
1188 | data = nfs_commitdata_alloc(); | |
1189 | if (!data) | |
9390f425 | 1190 | break; |
e0c2b380 | 1191 | data->ds_commit_index = i; |
799ba8d5 FI |
1192 | data->lseg = bucket->clseg; |
1193 | bucket->clseg = NULL; | |
e0c2b380 | 1194 | list_add(&data->pages, list); |
9390f425 | 1195 | nreq++; |
e0c2b380 | 1196 | } |
e0c2b380 | 1197 | |
9390f425 | 1198 | /* Clean up on error */ |
799ba8d5 FI |
1199 | for (j = i; j < fl_cinfo->nbuckets; j++, bucket++) { |
1200 | if (list_empty(&bucket->committing)) | |
e0c2b380 | 1201 | continue; |
ea2cf228 | 1202 | nfs_retry_commit(&bucket->committing, bucket->clseg, cinfo); |
799ba8d5 FI |
1203 | put_lseg(bucket->clseg); |
1204 | bucket->clseg = NULL; | |
e0c2b380 | 1205 | } |
e0c2b380 | 1206 | /* Caller will clean up entries put on list */ |
9390f425 | 1207 | return nreq; |
e0c2b380 FI |
1208 | } |
1209 | ||
1210 | /* This follows nfs_commit_list pretty closely */ | |
1211 | static int | |
1212 | filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, | |
ea2cf228 | 1213 | int how, struct nfs_commit_info *cinfo) |
e0c2b380 | 1214 | { |
0b7c0153 | 1215 | struct nfs_commit_data *data, *tmp; |
e0c2b380 | 1216 | LIST_HEAD(list); |
9390f425 | 1217 | unsigned int nreq = 0; |
e0c2b380 FI |
1218 | |
1219 | if (!list_empty(mds_pages)) { | |
1220 | data = nfs_commitdata_alloc(); | |
9390f425 TM |
1221 | if (data != NULL) { |
1222 | data->lseg = NULL; | |
1223 | list_add(&data->pages, &list); | |
1224 | nreq++; | |
1225 | } else | |
ea2cf228 | 1226 | nfs_retry_commit(mds_pages, NULL, cinfo); |
e0c2b380 FI |
1227 | } |
1228 | ||
ea2cf228 | 1229 | nreq += alloc_ds_commits(cinfo, &list); |
9390f425 TM |
1230 | |
1231 | if (nreq == 0) { | |
f453a54a | 1232 | cinfo->completion_ops->error_cleanup(NFS_I(inode)); |
9390f425 TM |
1233 | goto out; |
1234 | } | |
1235 | ||
ea2cf228 | 1236 | atomic_add(nreq, &cinfo->mds->rpcs_out); |
e0c2b380 FI |
1237 | |
1238 | list_for_each_entry_safe(data, tmp, &list, pages) { | |
1239 | list_del_init(&data->pages); | |
e0c2b380 | 1240 | if (!data->lseg) { |
f453a54a | 1241 | nfs_init_commit(data, mds_pages, NULL, cinfo); |
0b7c0153 | 1242 | nfs_initiate_commit(NFS_CLIENT(inode), data, |
9f0ec176 | 1243 | data->mds_ops, how, 0); |
e0c2b380 | 1244 | } else { |
ea2cf228 | 1245 | struct pnfs_commit_bucket *buckets; |
799ba8d5 | 1246 | |
ea2cf228 | 1247 | buckets = cinfo->ds->buckets; |
f453a54a | 1248 | nfs_init_commit(data, &buckets[data->ds_commit_index].committing, data->lseg, cinfo); |
e0c2b380 FI |
1249 | filelayout_initiate_commit(data, how); |
1250 | } | |
1251 | } | |
9390f425 | 1252 | out: |
ea2cf228 | 1253 | cinfo->ds->ncommitting = 0; |
9390f425 | 1254 | return PNFS_ATTEMPTED; |
e0c2b380 FI |
1255 | } |
1256 | ||
1775bc34 BH |
1257 | static void |
1258 | filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d) | |
1259 | { | |
1260 | nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node)); | |
1261 | } | |
1262 | ||
799ba8d5 FI |
1263 | static struct pnfs_layout_hdr * |
1264 | filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags) | |
1265 | { | |
1266 | struct nfs4_filelayout *flo; | |
1267 | ||
1268 | flo = kzalloc(sizeof(*flo), gfp_flags); | |
1269 | return &flo->generic_hdr; | |
1270 | } | |
1271 | ||
1272 | static void | |
1273 | filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
1274 | { | |
1275 | kfree(FILELAYOUT_FROM_HDR(lo)); | |
1276 | } | |
1277 | ||
ea2cf228 FI |
1278 | static struct pnfs_ds_commit_info * |
1279 | filelayout_get_ds_info(struct inode *inode) | |
1280 | { | |
df011748 FI |
1281 | struct pnfs_layout_hdr *layout = NFS_I(inode)->layout; |
1282 | ||
1283 | if (layout == NULL) | |
1284 | return NULL; | |
1285 | else | |
1286 | return &FILELAYOUT_FROM_HDR(layout)->commit_info; | |
ea2cf228 FI |
1287 | } |
1288 | ||
7ab672ce | 1289 | static struct pnfs_layoutdriver_type filelayout_type = { |
ea8eecdd CH |
1290 | .id = LAYOUT_NFSV4_1_FILES, |
1291 | .name = "LAYOUT_NFSV4_1_FILES", | |
1292 | .owner = THIS_MODULE, | |
799ba8d5 FI |
1293 | .alloc_layout_hdr = filelayout_alloc_layout_hdr, |
1294 | .free_layout_hdr = filelayout_free_layout_hdr, | |
ea8eecdd CH |
1295 | .alloc_lseg = filelayout_alloc_lseg, |
1296 | .free_lseg = filelayout_free_lseg, | |
1751c363 TM |
1297 | .pg_read_ops = &filelayout_pg_read_ops, |
1298 | .pg_write_ops = &filelayout_pg_write_ops, | |
ea2cf228 | 1299 | .get_ds_info = &filelayout_get_ds_info, |
8dd37758 TM |
1300 | .mark_request_commit = filelayout_mark_request_commit, |
1301 | .clear_request_commit = filelayout_clear_request_commit, | |
d6d6dc7c | 1302 | .scan_commit_lists = filelayout_scan_commit_lists, |
1763da12 | 1303 | .recover_commit_reqs = filelayout_recover_commit_reqs, |
e0c2b380 | 1304 | .commit_pagelist = filelayout_commit_pagelist, |
dc70d7b3 | 1305 | .read_pagelist = filelayout_read_pagelist, |
0382b744 | 1306 | .write_pagelist = filelayout_write_pagelist, |
1775bc34 | 1307 | .free_deviceid_node = filelayout_free_deveiceid_node, |
7ab672ce DH |
1308 | }; |
1309 | ||
1310 | static int __init nfs4filelayout_init(void) | |
1311 | { | |
1312 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n", | |
1313 | __func__); | |
1314 | return pnfs_register_layoutdriver(&filelayout_type); | |
1315 | } | |
1316 | ||
1317 | static void __exit nfs4filelayout_exit(void) | |
1318 | { | |
1319 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n", | |
1320 | __func__); | |
1321 | pnfs_unregister_layoutdriver(&filelayout_type); | |
1322 | } | |
1323 | ||
f85ef69c BF |
1324 | MODULE_ALIAS("nfs-layouttype4-1"); |
1325 | ||
7ab672ce DH |
1326 | module_init(nfs4filelayout_init); |
1327 | module_exit(nfs4filelayout_exit); |