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 AA |
35 | |
36 | #include "internal.h" | |
37 | #include "nfs4filelayout.h" | |
7ab672ce DH |
38 | |
39 | #define NFSDBG_FACILITY NFSDBG_PNFS_LD | |
40 | ||
41 | MODULE_LICENSE("GPL"); | |
42 | MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>"); | |
43 | MODULE_DESCRIPTION("The NFSv4 file layout driver"); | |
44 | ||
cbdabc7f AA |
45 | #define FILELAYOUT_POLL_RETRY_MAX (15*HZ) |
46 | ||
cfe7f412 FI |
47 | static loff_t |
48 | filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg, | |
49 | loff_t offset) | |
50 | { | |
51 | u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count; | |
52 | u64 tmp; | |
53 | ||
54 | offset -= flseg->pattern_offset; | |
55 | tmp = offset; | |
56 | do_div(tmp, stripe_width); | |
57 | ||
58 | return tmp * flseg->stripe_unit + do_div(offset, flseg->stripe_unit); | |
59 | } | |
60 | ||
61 | /* This function is used by the layout driver to calculate the | |
62 | * offset of the file on the dserver based on whether the | |
63 | * layout type is STRIPE_DENSE or STRIPE_SPARSE | |
64 | */ | |
65 | static loff_t | |
66 | filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset) | |
67 | { | |
68 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
69 | ||
70 | switch (flseg->stripe_type) { | |
71 | case STRIPE_SPARSE: | |
72 | return offset; | |
73 | ||
74 | case STRIPE_DENSE: | |
75 | return filelayout_get_dense_offset(flseg, offset); | |
76 | } | |
77 | ||
78 | BUG(); | |
79 | } | |
80 | ||
cbdabc7f AA |
81 | static int filelayout_async_handle_error(struct rpc_task *task, |
82 | struct nfs4_state *state, | |
83 | struct nfs_client *clp, | |
84 | int *reset) | |
85 | { | |
86 | if (task->tk_status >= 0) | |
87 | return 0; | |
88 | ||
89 | *reset = 0; | |
90 | ||
91 | switch (task->tk_status) { | |
92 | case -NFS4ERR_BADSESSION: | |
93 | case -NFS4ERR_BADSLOT: | |
94 | case -NFS4ERR_BAD_HIGH_SLOT: | |
95 | case -NFS4ERR_DEADSESSION: | |
96 | case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: | |
97 | case -NFS4ERR_SEQ_FALSE_RETRY: | |
98 | case -NFS4ERR_SEQ_MISORDERED: | |
99 | dprintk("%s ERROR %d, Reset session. Exchangeid " | |
100 | "flags 0x%x\n", __func__, task->tk_status, | |
101 | clp->cl_exchange_flags); | |
102 | nfs4_schedule_session_recovery(clp->cl_session); | |
103 | break; | |
104 | case -NFS4ERR_DELAY: | |
105 | case -NFS4ERR_GRACE: | |
106 | case -EKEYEXPIRED: | |
107 | rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX); | |
108 | break; | |
a8a4ae3a AA |
109 | case -NFS4ERR_RETRY_UNCACHED_REP: |
110 | break; | |
cbdabc7f AA |
111 | default: |
112 | dprintk("%s DS error. Retry through MDS %d\n", __func__, | |
113 | task->tk_status); | |
114 | *reset = 1; | |
115 | break; | |
116 | } | |
117 | task->tk_status = 0; | |
118 | return -EAGAIN; | |
119 | } | |
120 | ||
121 | /* NFS_PROTO call done callback routines */ | |
122 | ||
123 | static int filelayout_read_done_cb(struct rpc_task *task, | |
124 | struct nfs_read_data *data) | |
125 | { | |
cbdabc7f AA |
126 | int reset = 0; |
127 | ||
128 | dprintk("%s DS read\n", __func__); | |
129 | ||
130 | if (filelayout_async_handle_error(task, data->args.context->state, | |
131 | data->ds_clp, &reset) == -EAGAIN) { | |
132 | dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n", | |
133 | __func__, data->ds_clp, data->ds_clp->cl_session); | |
134 | if (reset) { | |
1b0ae068 | 135 | pnfs_set_lo_fail(data->lseg); |
cbdabc7f | 136 | nfs4_reset_read(task, data); |
cbdabc7f | 137 | } |
d00c5d43 | 138 | rpc_restart_call_prepare(task); |
cbdabc7f AA |
139 | return -EAGAIN; |
140 | } | |
141 | ||
142 | return 0; | |
143 | } | |
144 | ||
863a3c6c AA |
145 | /* |
146 | * We reference the rpc_cred of the first WRITE that triggers the need for | |
147 | * a LAYOUTCOMMIT, and use it to send the layoutcommit compound. | |
148 | * rfc5661 is not clear about which credential should be used. | |
149 | */ | |
150 | static void | |
151 | filelayout_set_layoutcommit(struct nfs_write_data *wdata) | |
152 | { | |
153 | if (FILELAYOUT_LSEG(wdata->lseg)->commit_through_mds || | |
154 | wdata->res.verf->committed == NFS_FILE_SYNC) | |
155 | return; | |
156 | ||
157 | pnfs_set_layoutcommit(wdata); | |
158 | dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino, | |
acff5880 | 159 | (unsigned long) NFS_I(wdata->inode)->layout->plh_lwb); |
863a3c6c AA |
160 | } |
161 | ||
dc70d7b3 AA |
162 | /* |
163 | * Call ops for the async read/write cases | |
164 | * In the case of dense layouts, the offset needs to be reset to its | |
165 | * original value. | |
166 | */ | |
167 | static void filelayout_read_prepare(struct rpc_task *task, void *data) | |
168 | { | |
169 | struct nfs_read_data *rdata = (struct nfs_read_data *)data; | |
170 | ||
cbdabc7f AA |
171 | rdata->read_done_cb = filelayout_read_done_cb; |
172 | ||
dc70d7b3 AA |
173 | if (nfs41_setup_sequence(rdata->ds_clp->cl_session, |
174 | &rdata->args.seq_args, &rdata->res.seq_res, | |
175 | 0, task)) | |
176 | return; | |
177 | ||
178 | rpc_call_start(task); | |
179 | } | |
180 | ||
181 | static void filelayout_read_call_done(struct rpc_task *task, void *data) | |
182 | { | |
183 | struct nfs_read_data *rdata = (struct nfs_read_data *)data; | |
184 | ||
185 | dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status); | |
186 | ||
187 | /* Note this may cause RPC to be resent */ | |
188 | rdata->mds_ops->rpc_call_done(task, data); | |
189 | } | |
190 | ||
191 | static void filelayout_read_release(void *data) | |
192 | { | |
193 | struct nfs_read_data *rdata = (struct nfs_read_data *)data; | |
194 | ||
195 | rdata->mds_ops->rpc_release(data); | |
196 | } | |
197 | ||
a69aef14 FI |
198 | static int filelayout_write_done_cb(struct rpc_task *task, |
199 | struct nfs_write_data *data) | |
200 | { | |
201 | int reset = 0; | |
202 | ||
203 | if (filelayout_async_handle_error(task, data->args.context->state, | |
204 | data->ds_clp, &reset) == -EAGAIN) { | |
a69aef14 FI |
205 | dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n", |
206 | __func__, data->ds_clp, data->ds_clp->cl_session); | |
207 | if (reset) { | |
1b0ae068 | 208 | pnfs_set_lo_fail(data->lseg); |
a69aef14 | 209 | nfs4_reset_write(task, data); |
d00c5d43 TM |
210 | } |
211 | rpc_restart_call_prepare(task); | |
a69aef14 FI |
212 | return -EAGAIN; |
213 | } | |
214 | ||
863a3c6c | 215 | filelayout_set_layoutcommit(data); |
a69aef14 FI |
216 | return 0; |
217 | } | |
218 | ||
e0c2b380 FI |
219 | /* Fake up some data that will cause nfs_commit_release to retry the writes. */ |
220 | static void prepare_to_resend_writes(struct nfs_write_data *data) | |
221 | { | |
222 | struct nfs_page *first = nfs_list_entry(data->pages.next); | |
223 | ||
224 | data->task.tk_status = 0; | |
225 | memcpy(data->verf.verifier, first->wb_verf.verifier, | |
226 | sizeof(first->wb_verf.verifier)); | |
227 | data->verf.verifier[0]++; /* ensure verifier mismatch */ | |
228 | } | |
229 | ||
230 | static int filelayout_commit_done_cb(struct rpc_task *task, | |
231 | struct nfs_write_data *data) | |
232 | { | |
233 | int reset = 0; | |
234 | ||
235 | if (filelayout_async_handle_error(task, data->args.context->state, | |
236 | data->ds_clp, &reset) == -EAGAIN) { | |
237 | dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n", | |
238 | __func__, data->ds_clp, data->ds_clp->cl_session); | |
239 | if (reset) { | |
240 | prepare_to_resend_writes(data); | |
1b0ae068 | 241 | pnfs_set_lo_fail(data->lseg); |
e0c2b380 | 242 | } else |
d00c5d43 | 243 | rpc_restart_call_prepare(task); |
e0c2b380 FI |
244 | return -EAGAIN; |
245 | } | |
246 | ||
247 | return 0; | |
248 | } | |
249 | ||
a69aef14 FI |
250 | static void filelayout_write_prepare(struct rpc_task *task, void *data) |
251 | { | |
252 | struct nfs_write_data *wdata = (struct nfs_write_data *)data; | |
253 | ||
254 | if (nfs41_setup_sequence(wdata->ds_clp->cl_session, | |
255 | &wdata->args.seq_args, &wdata->res.seq_res, | |
256 | 0, task)) | |
257 | return; | |
258 | ||
259 | rpc_call_start(task); | |
260 | } | |
261 | ||
262 | static void filelayout_write_call_done(struct rpc_task *task, void *data) | |
263 | { | |
264 | struct nfs_write_data *wdata = (struct nfs_write_data *)data; | |
265 | ||
266 | /* Note this may cause RPC to be resent */ | |
267 | wdata->mds_ops->rpc_call_done(task, data); | |
268 | } | |
269 | ||
270 | static void filelayout_write_release(void *data) | |
271 | { | |
272 | struct nfs_write_data *wdata = (struct nfs_write_data *)data; | |
273 | ||
274 | wdata->mds_ops->rpc_release(data); | |
275 | } | |
276 | ||
e0c2b380 FI |
277 | static void filelayout_commit_release(void *data) |
278 | { | |
279 | struct nfs_write_data *wdata = (struct nfs_write_data *)data; | |
280 | ||
281 | nfs_commit_release_pages(wdata); | |
282 | if (atomic_dec_and_test(&NFS_I(wdata->inode)->commits_outstanding)) | |
283 | nfs_commit_clear_lock(NFS_I(wdata->inode)); | |
284 | nfs_commitdata_release(wdata); | |
285 | } | |
286 | ||
dc70d7b3 AA |
287 | struct rpc_call_ops filelayout_read_call_ops = { |
288 | .rpc_call_prepare = filelayout_read_prepare, | |
289 | .rpc_call_done = filelayout_read_call_done, | |
290 | .rpc_release = filelayout_read_release, | |
291 | }; | |
292 | ||
a69aef14 FI |
293 | struct rpc_call_ops filelayout_write_call_ops = { |
294 | .rpc_call_prepare = filelayout_write_prepare, | |
295 | .rpc_call_done = filelayout_write_call_done, | |
296 | .rpc_release = filelayout_write_release, | |
297 | }; | |
298 | ||
e0c2b380 FI |
299 | struct rpc_call_ops filelayout_commit_call_ops = { |
300 | .rpc_call_prepare = filelayout_write_prepare, | |
301 | .rpc_call_done = filelayout_write_call_done, | |
302 | .rpc_release = filelayout_commit_release, | |
303 | }; | |
304 | ||
dc70d7b3 AA |
305 | static enum pnfs_try_status |
306 | filelayout_read_pagelist(struct nfs_read_data *data) | |
307 | { | |
308 | struct pnfs_layout_segment *lseg = data->lseg; | |
309 | struct nfs4_pnfs_ds *ds; | |
310 | loff_t offset = data->args.offset; | |
311 | u32 j, idx; | |
312 | struct nfs_fh *fh; | |
313 | int status; | |
314 | ||
315 | dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n", | |
316 | __func__, data->inode->i_ino, | |
317 | data->args.pgbase, (size_t)data->args.count, offset); | |
318 | ||
c47abcf8 AA |
319 | if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags)) |
320 | return PNFS_NOT_ATTEMPTED; | |
321 | ||
dc70d7b3 AA |
322 | /* Retrieve the correct rpc_client for the byte range */ |
323 | j = nfs4_fl_calc_j_index(lseg, offset); | |
324 | idx = nfs4_fl_calc_ds_index(lseg, j); | |
325 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
326 | if (!ds) { | |
568e8c49 AA |
327 | /* Either layout fh index faulty, or ds connect failed */ |
328 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
329 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
dc70d7b3 AA |
330 | return PNFS_NOT_ATTEMPTED; |
331 | } | |
c9895cb6 | 332 | dprintk("%s USE DS: %s\n", __func__, ds->ds_remotestr); |
dc70d7b3 AA |
333 | |
334 | /* No multipath support. Use first DS */ | |
335 | data->ds_clp = ds->ds_clp; | |
336 | fh = nfs4_fl_select_ds_fh(lseg, j); | |
337 | if (fh) | |
338 | data->args.fh = fh; | |
339 | ||
340 | data->args.offset = filelayout_get_dserver_offset(lseg, offset); | |
341 | data->mds_offset = offset; | |
342 | ||
343 | /* Perform an asynchronous read to ds */ | |
344 | status = nfs_initiate_read(data, ds->ds_clp->cl_rpcclient, | |
345 | &filelayout_read_call_ops); | |
346 | BUG_ON(status != 0); | |
347 | return PNFS_ATTEMPTED; | |
348 | } | |
349 | ||
a69aef14 | 350 | /* Perform async writes. */ |
0382b744 AA |
351 | static enum pnfs_try_status |
352 | filelayout_write_pagelist(struct nfs_write_data *data, int sync) | |
353 | { | |
a69aef14 FI |
354 | struct pnfs_layout_segment *lseg = data->lseg; |
355 | struct nfs4_pnfs_ds *ds; | |
356 | loff_t offset = data->args.offset; | |
357 | u32 j, idx; | |
358 | struct nfs_fh *fh; | |
359 | int status; | |
360 | ||
c47abcf8 AA |
361 | if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags)) |
362 | return PNFS_NOT_ATTEMPTED; | |
363 | ||
a69aef14 FI |
364 | /* Retrieve the correct rpc_client for the byte range */ |
365 | j = nfs4_fl_calc_j_index(lseg, offset); | |
366 | idx = nfs4_fl_calc_ds_index(lseg, j); | |
367 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
368 | if (!ds) { | |
369 | printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__); | |
370 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
371 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
372 | return PNFS_NOT_ATTEMPTED; | |
373 | } | |
c9895cb6 | 374 | dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s\n", __func__, |
a69aef14 | 375 | data->inode->i_ino, sync, (size_t) data->args.count, offset, |
c9895cb6 | 376 | ds->ds_remotestr); |
a69aef14 | 377 | |
a69aef14 FI |
378 | data->write_done_cb = filelayout_write_done_cb; |
379 | data->ds_clp = ds->ds_clp; | |
380 | fh = nfs4_fl_select_ds_fh(lseg, j); | |
381 | if (fh) | |
382 | data->args.fh = fh; | |
383 | /* | |
384 | * Get the file offset on the dserver. Set the write offset to | |
385 | * this offset and save the original offset. | |
386 | */ | |
387 | data->args.offset = filelayout_get_dserver_offset(lseg, offset); | |
a69aef14 FI |
388 | |
389 | /* Perform an asynchronous write */ | |
390 | status = nfs_initiate_write(data, ds->ds_clp->cl_rpcclient, | |
391 | &filelayout_write_call_ops, sync); | |
392 | BUG_ON(status != 0); | |
393 | return PNFS_ATTEMPTED; | |
0382b744 AA |
394 | } |
395 | ||
16b374ca AA |
396 | /* |
397 | * filelayout_check_layout() | |
398 | * | |
399 | * Make sure layout segment parameters are sane WRT the device. | |
400 | * At this point no generic layer initialization of the lseg has occurred, | |
401 | * and nothing has been added to the layout_hdr cache. | |
402 | * | |
403 | */ | |
404 | static int | |
405 | filelayout_check_layout(struct pnfs_layout_hdr *lo, | |
406 | struct nfs4_filelayout_segment *fl, | |
407 | struct nfs4_layoutget_res *lgr, | |
a75b9df9 TM |
408 | struct nfs4_deviceid *id, |
409 | gfp_t gfp_flags) | |
16b374ca | 410 | { |
a1eaecbc | 411 | struct nfs4_deviceid_node *d; |
16b374ca AA |
412 | struct nfs4_file_layout_dsaddr *dsaddr; |
413 | int status = -EINVAL; | |
b7edfaa1 | 414 | struct nfs_server *nfss = NFS_SERVER(lo->plh_inode); |
16b374ca AA |
415 | |
416 | dprintk("--> %s\n", __func__); | |
417 | ||
7c24d948 AA |
418 | /* FIXME: remove this check when layout segment support is added */ |
419 | if (lgr->range.offset != 0 || | |
420 | lgr->range.length != NFS4_MAX_UINT64) { | |
421 | dprintk("%s Only whole file layouts supported. Use MDS i/o\n", | |
422 | __func__); | |
423 | goto out; | |
424 | } | |
425 | ||
16b374ca | 426 | if (fl->pattern_offset > lgr->range.offset) { |
3b6445a6 | 427 | dprintk("%s pattern_offset %lld too large\n", |
16b374ca AA |
428 | __func__, fl->pattern_offset); |
429 | goto out; | |
430 | } | |
431 | ||
75247aff BH |
432 | if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) { |
433 | dprintk("%s Invalid stripe unit (%u)\n", | |
16b374ca AA |
434 | __func__, fl->stripe_unit); |
435 | goto out; | |
436 | } | |
437 | ||
438 | /* find and reference the deviceid */ | |
35c8bb54 BH |
439 | d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld, |
440 | NFS_SERVER(lo->plh_inode)->nfs_client, id); | |
a1eaecbc | 441 | if (d == NULL) { |
a75b9df9 | 442 | dsaddr = get_device_info(lo->plh_inode, id, gfp_flags); |
16b374ca AA |
443 | if (dsaddr == NULL) |
444 | goto out; | |
a1eaecbc BH |
445 | } else |
446 | dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node); | |
c47abcf8 AA |
447 | /* Found deviceid is being reaped */ |
448 | if (test_bit(NFS_DEVICEID_INVALID, &dsaddr->id_node.flags)) | |
449 | goto out_put; | |
450 | ||
16b374ca AA |
451 | fl->dsaddr = dsaddr; |
452 | ||
e414966b CL |
453 | if (fl->first_stripe_index >= dsaddr->stripe_count) { |
454 | dprintk("%s Bad first_stripe_index %u\n", | |
16b374ca AA |
455 | __func__, fl->first_stripe_index); |
456 | goto out_put; | |
457 | } | |
458 | ||
459 | if ((fl->stripe_type == STRIPE_SPARSE && | |
460 | fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) || | |
461 | (fl->stripe_type == STRIPE_DENSE && | |
462 | fl->num_fh != dsaddr->stripe_count)) { | |
463 | dprintk("%s num_fh %u not valid for given packing\n", | |
464 | __func__, fl->num_fh); | |
465 | goto out_put; | |
466 | } | |
467 | ||
468 | if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) { | |
469 | dprintk("%s Stripe unit (%u) not aligned with rsize %u " | |
470 | "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize, | |
471 | nfss->wsize); | |
472 | } | |
473 | ||
474 | status = 0; | |
475 | out: | |
476 | dprintk("--> %s returns %d\n", __func__, status); | |
477 | return status; | |
478 | out_put: | |
ea8eecdd | 479 | nfs4_fl_put_deviceid(dsaddr); |
16b374ca AA |
480 | goto out; |
481 | } | |
482 | ||
483 | static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl) | |
484 | { | |
485 | int i; | |
486 | ||
487 | for (i = 0; i < fl->num_fh; i++) { | |
488 | if (!fl->fh_array[i]) | |
489 | break; | |
490 | kfree(fl->fh_array[i]); | |
491 | } | |
492 | kfree(fl->fh_array); | |
493 | fl->fh_array = NULL; | |
494 | } | |
495 | ||
496 | static void | |
497 | _filelayout_free_lseg(struct nfs4_filelayout_segment *fl) | |
498 | { | |
499 | filelayout_free_fh_array(fl); | |
500 | kfree(fl); | |
501 | } | |
502 | ||
503 | static int | |
504 | filelayout_decode_layout(struct pnfs_layout_hdr *flo, | |
505 | struct nfs4_filelayout_segment *fl, | |
506 | struct nfs4_layoutget_res *lgr, | |
a75b9df9 TM |
507 | struct nfs4_deviceid *id, |
508 | gfp_t gfp_flags) | |
16b374ca | 509 | { |
35124a09 | 510 | struct xdr_stream stream; |
f7da7a12 | 511 | struct xdr_buf buf; |
35124a09 WAA |
512 | struct page *scratch; |
513 | __be32 *p; | |
16b374ca AA |
514 | uint32_t nfl_util; |
515 | int i; | |
516 | ||
517 | dprintk("%s: set_layout_map Begin\n", __func__); | |
518 | ||
a75b9df9 | 519 | scratch = alloc_page(gfp_flags); |
35124a09 WAA |
520 | if (!scratch) |
521 | return -ENOMEM; | |
522 | ||
f7da7a12 | 523 | xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len); |
35124a09 WAA |
524 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); |
525 | ||
526 | /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8), | |
527 | * num_fh (4) */ | |
528 | p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20); | |
529 | if (unlikely(!p)) | |
530 | goto out_err; | |
531 | ||
16b374ca AA |
532 | memcpy(id, p, sizeof(*id)); |
533 | p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE); | |
a1eaecbc | 534 | nfs4_print_deviceid(id); |
16b374ca AA |
535 | |
536 | nfl_util = be32_to_cpup(p++); | |
537 | if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS) | |
538 | fl->commit_through_mds = 1; | |
539 | if (nfl_util & NFL4_UFLG_DENSE) | |
540 | fl->stripe_type = STRIPE_DENSE; | |
541 | else | |
542 | fl->stripe_type = STRIPE_SPARSE; | |
543 | fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK; | |
544 | ||
545 | fl->first_stripe_index = be32_to_cpup(p++); | |
546 | p = xdr_decode_hyper(p, &fl->pattern_offset); | |
547 | fl->num_fh = be32_to_cpup(p++); | |
548 | ||
549 | dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n", | |
550 | __func__, nfl_util, fl->num_fh, fl->first_stripe_index, | |
551 | fl->pattern_offset); | |
552 | ||
cec765cf AA |
553 | /* Note that a zero value for num_fh is legal for STRIPE_SPARSE. |
554 | * Futher checking is done in filelayout_check_layout */ | |
e414966b | 555 | if (fl->num_fh > |
cec765cf | 556 | max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT)) |
35124a09 WAA |
557 | goto out_err; |
558 | ||
cec765cf AA |
559 | if (fl->num_fh > 0) { |
560 | fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *), | |
561 | gfp_flags); | |
562 | if (!fl->fh_array) | |
563 | goto out_err; | |
564 | } | |
16b374ca AA |
565 | |
566 | for (i = 0; i < fl->num_fh; i++) { | |
567 | /* Do we want to use a mempool here? */ | |
a75b9df9 | 568 | fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags); |
35124a09 WAA |
569 | if (!fl->fh_array[i]) |
570 | goto out_err_free; | |
571 | ||
572 | p = xdr_inline_decode(&stream, 4); | |
573 | if (unlikely(!p)) | |
574 | goto out_err_free; | |
16b374ca AA |
575 | fl->fh_array[i]->size = be32_to_cpup(p++); |
576 | if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) { | |
577 | printk(KERN_ERR "Too big fh %d received %d\n", | |
578 | i, fl->fh_array[i]->size); | |
35124a09 | 579 | goto out_err_free; |
16b374ca | 580 | } |
35124a09 WAA |
581 | |
582 | p = xdr_inline_decode(&stream, fl->fh_array[i]->size); | |
583 | if (unlikely(!p)) | |
584 | goto out_err_free; | |
16b374ca | 585 | memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size); |
16b374ca AA |
586 | dprintk("DEBUG: %s: fh len %d\n", __func__, |
587 | fl->fh_array[i]->size); | |
588 | } | |
589 | ||
35124a09 | 590 | __free_page(scratch); |
16b374ca | 591 | return 0; |
35124a09 WAA |
592 | |
593 | out_err_free: | |
594 | filelayout_free_fh_array(fl); | |
595 | out_err: | |
596 | __free_page(scratch); | |
597 | return -EIO; | |
16b374ca AA |
598 | } |
599 | ||
c879513e FI |
600 | static void |
601 | filelayout_free_lseg(struct pnfs_layout_segment *lseg) | |
602 | { | |
603 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); | |
604 | ||
605 | dprintk("--> %s\n", __func__); | |
606 | nfs4_fl_put_deviceid(fl->dsaddr); | |
425eb736 | 607 | kfree(fl->commit_buckets); |
c879513e FI |
608 | _filelayout_free_lseg(fl); |
609 | } | |
610 | ||
16b374ca AA |
611 | static struct pnfs_layout_segment * |
612 | filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid, | |
a75b9df9 TM |
613 | struct nfs4_layoutget_res *lgr, |
614 | gfp_t gfp_flags) | |
16b374ca AA |
615 | { |
616 | struct nfs4_filelayout_segment *fl; | |
617 | int rc; | |
618 | struct nfs4_deviceid id; | |
619 | ||
620 | dprintk("--> %s\n", __func__); | |
a75b9df9 | 621 | fl = kzalloc(sizeof(*fl), gfp_flags); |
16b374ca AA |
622 | if (!fl) |
623 | return NULL; | |
624 | ||
a75b9df9 TM |
625 | rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags); |
626 | if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) { | |
16b374ca AA |
627 | _filelayout_free_lseg(fl); |
628 | return NULL; | |
629 | } | |
425eb736 FI |
630 | |
631 | /* This assumes there is only one IOMODE_RW lseg. What | |
632 | * we really want to do is have a layout_hdr level | |
633 | * dictionary of <multipath_list4, fh> keys, each | |
634 | * associated with a struct list_head, populated by calls | |
635 | * to filelayout_write_pagelist(). | |
636 | * */ | |
637 | if ((!fl->commit_through_mds) && (lgr->range.iomode == IOMODE_RW)) { | |
638 | int i; | |
639 | int size = (fl->stripe_type == STRIPE_SPARSE) ? | |
640 | fl->dsaddr->ds_num : fl->dsaddr->stripe_count; | |
641 | ||
a75b9df9 | 642 | fl->commit_buckets = kcalloc(size, sizeof(struct list_head), gfp_flags); |
425eb736 FI |
643 | if (!fl->commit_buckets) { |
644 | filelayout_free_lseg(&fl->generic_hdr); | |
645 | return NULL; | |
646 | } | |
647 | fl->number_of_buckets = size; | |
648 | for (i = 0; i < size; i++) | |
649 | INIT_LIST_HEAD(&fl->commit_buckets[i]); | |
650 | } | |
16b374ca AA |
651 | return &fl->generic_hdr; |
652 | } | |
653 | ||
94ad1c80 FI |
654 | /* |
655 | * filelayout_pg_test(). Called by nfs_can_coalesce_requests() | |
656 | * | |
18ad0a9f BH |
657 | * return true : coalesce page |
658 | * return false : don't coalesce page | |
94ad1c80 | 659 | */ |
1751c363 | 660 | static bool |
94ad1c80 FI |
661 | filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
662 | struct nfs_page *req) | |
663 | { | |
664 | u64 p_stripe, r_stripe; | |
665 | u32 stripe_unit; | |
666 | ||
19345cb2 BH |
667 | if (!pnfs_generic_pg_test(pgio, prev, req) || |
668 | !nfs_generic_pg_test(pgio, prev, req)) | |
669 | return false; | |
89a58e32 | 670 | |
94ad1c80 FI |
671 | p_stripe = (u64)prev->wb_index << PAGE_CACHE_SHIFT; |
672 | r_stripe = (u64)req->wb_index << PAGE_CACHE_SHIFT; | |
673 | stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit; | |
674 | ||
675 | do_div(p_stripe, stripe_unit); | |
676 | do_div(r_stripe, stripe_unit); | |
677 | ||
678 | return (p_stripe == r_stripe); | |
679 | } | |
680 | ||
7c24d948 AA |
681 | void |
682 | filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio, | |
683 | struct nfs_page *req) | |
684 | { | |
685 | BUG_ON(pgio->pg_lseg != NULL); | |
686 | ||
687 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
688 | req->wb_context, | |
689 | 0, | |
690 | NFS4_MAX_UINT64, | |
691 | IOMODE_READ, | |
692 | GFP_KERNEL); | |
693 | /* If no lseg, fall back to read through mds */ | |
694 | if (pgio->pg_lseg == NULL) | |
1f945357 | 695 | nfs_pageio_reset_read_mds(pgio); |
7c24d948 AA |
696 | } |
697 | ||
698 | void | |
699 | filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio, | |
700 | struct nfs_page *req) | |
701 | { | |
702 | BUG_ON(pgio->pg_lseg != NULL); | |
703 | ||
704 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
705 | req->wb_context, | |
706 | 0, | |
707 | NFS4_MAX_UINT64, | |
708 | IOMODE_RW, | |
709 | GFP_NOFS); | |
710 | /* If no lseg, fall back to write through mds */ | |
711 | if (pgio->pg_lseg == NULL) | |
1f945357 | 712 | nfs_pageio_reset_write_mds(pgio); |
7c24d948 AA |
713 | } |
714 | ||
1751c363 | 715 | static const struct nfs_pageio_ops filelayout_pg_read_ops = { |
7c24d948 | 716 | .pg_init = filelayout_pg_init_read, |
1751c363 | 717 | .pg_test = filelayout_pg_test, |
493292dd | 718 | .pg_doio = pnfs_generic_pg_readpages, |
1751c363 TM |
719 | }; |
720 | ||
721 | static const struct nfs_pageio_ops filelayout_pg_write_ops = { | |
7c24d948 | 722 | .pg_init = filelayout_pg_init_write, |
1751c363 | 723 | .pg_test = filelayout_pg_test, |
dce81290 | 724 | .pg_doio = pnfs_generic_pg_writepages, |
1751c363 TM |
725 | }; |
726 | ||
e0c2b380 FI |
727 | static bool filelayout_mark_pnfs_commit(struct pnfs_layout_segment *lseg) |
728 | { | |
729 | return !FILELAYOUT_LSEG(lseg)->commit_through_mds; | |
730 | } | |
731 | ||
732 | static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j) | |
733 | { | |
734 | if (fl->stripe_type == STRIPE_SPARSE) | |
735 | return nfs4_fl_calc_ds_index(&fl->generic_hdr, j); | |
736 | else | |
737 | return j; | |
738 | } | |
739 | ||
740 | struct list_head *filelayout_choose_commit_list(struct nfs_page *req) | |
741 | { | |
742 | struct pnfs_layout_segment *lseg = req->wb_commit_lseg; | |
743 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); | |
744 | u32 i, j; | |
745 | struct list_head *list; | |
746 | ||
747 | /* Note that we are calling nfs4_fl_calc_j_index on each page | |
748 | * that ends up being committed to a data server. An attractive | |
749 | * alternative is to add a field to nfs_write_data and nfs_page | |
750 | * to store the value calculated in filelayout_write_pagelist | |
751 | * and just use that here. | |
752 | */ | |
753 | j = nfs4_fl_calc_j_index(lseg, | |
754 | (loff_t)req->wb_index << PAGE_CACHE_SHIFT); | |
755 | i = select_bucket_index(fl, j); | |
756 | list = &fl->commit_buckets[i]; | |
757 | if (list_empty(list)) { | |
758 | /* Non-empty buckets hold a reference on the lseg */ | |
759 | get_lseg(lseg); | |
760 | } | |
761 | return list; | |
762 | } | |
763 | ||
764 | static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i) | |
765 | { | |
766 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
767 | ||
768 | if (flseg->stripe_type == STRIPE_SPARSE) | |
769 | return i; | |
770 | else | |
771 | return nfs4_fl_calc_ds_index(lseg, i); | |
772 | } | |
773 | ||
774 | static struct nfs_fh * | |
775 | select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i) | |
776 | { | |
777 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
778 | ||
779 | if (flseg->stripe_type == STRIPE_SPARSE) { | |
780 | if (flseg->num_fh == 1) | |
781 | i = 0; | |
782 | else if (flseg->num_fh == 0) | |
783 | /* Use the MDS OPEN fh set in nfs_read_rpcsetup */ | |
784 | return NULL; | |
785 | } | |
786 | return flseg->fh_array[i]; | |
787 | } | |
788 | ||
789 | static int filelayout_initiate_commit(struct nfs_write_data *data, int how) | |
790 | { | |
791 | struct pnfs_layout_segment *lseg = data->lseg; | |
792 | struct nfs4_pnfs_ds *ds; | |
793 | u32 idx; | |
794 | struct nfs_fh *fh; | |
795 | ||
796 | idx = calc_ds_index_from_commit(lseg, data->ds_commit_index); | |
797 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
798 | if (!ds) { | |
799 | printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__); | |
800 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
801 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
802 | prepare_to_resend_writes(data); | |
803 | data->mds_ops->rpc_release(data); | |
804 | return -EAGAIN; | |
805 | } | |
806 | dprintk("%s ino %lu, how %d\n", __func__, data->inode->i_ino, how); | |
807 | data->write_done_cb = filelayout_commit_done_cb; | |
808 | data->ds_clp = ds->ds_clp; | |
809 | fh = select_ds_fh_from_commit(lseg, data->ds_commit_index); | |
810 | if (fh) | |
811 | data->args.fh = fh; | |
812 | return nfs_initiate_commit(data, ds->ds_clp->cl_rpcclient, | |
813 | &filelayout_commit_call_ops, how); | |
814 | } | |
815 | ||
816 | /* | |
817 | * This is only useful while we are using whole file layouts. | |
818 | */ | |
819 | static struct pnfs_layout_segment *find_only_write_lseg(struct inode *inode) | |
820 | { | |
821 | struct pnfs_layout_segment *lseg, *rv = NULL; | |
822 | ||
823 | spin_lock(&inode->i_lock); | |
824 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) | |
825 | if (lseg->pls_range.iomode == IOMODE_RW) | |
826 | rv = get_lseg(lseg); | |
827 | spin_unlock(&inode->i_lock); | |
828 | return rv; | |
829 | } | |
830 | ||
831 | static int alloc_ds_commits(struct inode *inode, struct list_head *list) | |
832 | { | |
833 | struct pnfs_layout_segment *lseg; | |
834 | struct nfs4_filelayout_segment *fl; | |
835 | struct nfs_write_data *data; | |
836 | int i, j; | |
837 | ||
838 | /* Won't need this when non-whole file layout segments are supported | |
839 | * instead we will use a pnfs_layout_hdr structure */ | |
840 | lseg = find_only_write_lseg(inode); | |
841 | if (!lseg) | |
842 | return 0; | |
843 | fl = FILELAYOUT_LSEG(lseg); | |
844 | for (i = 0; i < fl->number_of_buckets; i++) { | |
845 | if (list_empty(&fl->commit_buckets[i])) | |
846 | continue; | |
847 | data = nfs_commitdata_alloc(); | |
848 | if (!data) | |
849 | goto out_bad; | |
850 | data->ds_commit_index = i; | |
851 | data->lseg = lseg; | |
852 | list_add(&data->pages, list); | |
853 | } | |
854 | put_lseg(lseg); | |
855 | return 0; | |
856 | ||
857 | out_bad: | |
858 | for (j = i; j < fl->number_of_buckets; j++) { | |
859 | if (list_empty(&fl->commit_buckets[i])) | |
860 | continue; | |
861 | nfs_retry_commit(&fl->commit_buckets[i], lseg); | |
862 | put_lseg(lseg); /* associated with emptying bucket */ | |
863 | } | |
864 | put_lseg(lseg); | |
865 | /* Caller will clean up entries put on list */ | |
866 | return -ENOMEM; | |
867 | } | |
868 | ||
869 | /* This follows nfs_commit_list pretty closely */ | |
870 | static int | |
871 | filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, | |
872 | int how) | |
873 | { | |
874 | struct nfs_write_data *data, *tmp; | |
875 | LIST_HEAD(list); | |
876 | ||
877 | if (!list_empty(mds_pages)) { | |
878 | data = nfs_commitdata_alloc(); | |
879 | if (!data) | |
880 | goto out_bad; | |
881 | data->lseg = NULL; | |
882 | list_add(&data->pages, &list); | |
883 | } | |
884 | ||
885 | if (alloc_ds_commits(inode, &list)) | |
886 | goto out_bad; | |
887 | ||
888 | list_for_each_entry_safe(data, tmp, &list, pages) { | |
889 | list_del_init(&data->pages); | |
890 | atomic_inc(&NFS_I(inode)->commits_outstanding); | |
891 | if (!data->lseg) { | |
892 | nfs_init_commit(data, mds_pages, NULL); | |
893 | nfs_initiate_commit(data, NFS_CLIENT(inode), | |
894 | data->mds_ops, how); | |
895 | } else { | |
896 | nfs_init_commit(data, &FILELAYOUT_LSEG(data->lseg)->commit_buckets[data->ds_commit_index], data->lseg); | |
897 | filelayout_initiate_commit(data, how); | |
898 | } | |
899 | } | |
900 | return 0; | |
901 | out_bad: | |
902 | list_for_each_entry_safe(data, tmp, &list, pages) { | |
903 | nfs_retry_commit(&data->pages, data->lseg); | |
904 | list_del_init(&data->pages); | |
905 | nfs_commit_free(data); | |
906 | } | |
907 | nfs_retry_commit(mds_pages, NULL); | |
908 | nfs_commit_clear_lock(NFS_I(inode)); | |
909 | return -ENOMEM; | |
910 | } | |
911 | ||
1775bc34 BH |
912 | static void |
913 | filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d) | |
914 | { | |
915 | nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node)); | |
916 | } | |
917 | ||
7ab672ce | 918 | static struct pnfs_layoutdriver_type filelayout_type = { |
ea8eecdd CH |
919 | .id = LAYOUT_NFSV4_1_FILES, |
920 | .name = "LAYOUT_NFSV4_1_FILES", | |
921 | .owner = THIS_MODULE, | |
922 | .alloc_lseg = filelayout_alloc_lseg, | |
923 | .free_lseg = filelayout_free_lseg, | |
1751c363 TM |
924 | .pg_read_ops = &filelayout_pg_read_ops, |
925 | .pg_write_ops = &filelayout_pg_write_ops, | |
e0c2b380 FI |
926 | .mark_pnfs_commit = filelayout_mark_pnfs_commit, |
927 | .choose_commit_list = filelayout_choose_commit_list, | |
928 | .commit_pagelist = filelayout_commit_pagelist, | |
dc70d7b3 | 929 | .read_pagelist = filelayout_read_pagelist, |
0382b744 | 930 | .write_pagelist = filelayout_write_pagelist, |
1775bc34 | 931 | .free_deviceid_node = filelayout_free_deveiceid_node, |
7ab672ce DH |
932 | }; |
933 | ||
934 | static int __init nfs4filelayout_init(void) | |
935 | { | |
936 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n", | |
937 | __func__); | |
938 | return pnfs_register_layoutdriver(&filelayout_type); | |
939 | } | |
940 | ||
941 | static void __exit nfs4filelayout_exit(void) | |
942 | { | |
943 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n", | |
944 | __func__); | |
945 | pnfs_unregister_layoutdriver(&filelayout_type); | |
946 | } | |
947 | ||
f85ef69c BF |
948 | MODULE_ALIAS("nfs-layouttype4-1"); |
949 | ||
7ab672ce DH |
950 | module_init(nfs4filelayout_init); |
951 | module_exit(nfs4filelayout_exit); |