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> | |
16b374ca AA |
33 | |
34 | #include "internal.h" | |
35 | #include "nfs4filelayout.h" | |
7ab672ce DH |
36 | |
37 | #define NFSDBG_FACILITY NFSDBG_PNFS_LD | |
38 | ||
39 | MODULE_LICENSE("GPL"); | |
40 | MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>"); | |
41 | MODULE_DESCRIPTION("The NFSv4 file layout driver"); | |
42 | ||
cbdabc7f AA |
43 | #define FILELAYOUT_POLL_RETRY_MAX (15*HZ) |
44 | ||
cfe7f412 FI |
45 | static loff_t |
46 | filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg, | |
47 | loff_t offset) | |
48 | { | |
49 | u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count; | |
50 | u64 tmp; | |
51 | ||
52 | offset -= flseg->pattern_offset; | |
53 | tmp = offset; | |
54 | do_div(tmp, stripe_width); | |
55 | ||
56 | return tmp * flseg->stripe_unit + do_div(offset, flseg->stripe_unit); | |
57 | } | |
58 | ||
59 | /* This function is used by the layout driver to calculate the | |
60 | * offset of the file on the dserver based on whether the | |
61 | * layout type is STRIPE_DENSE or STRIPE_SPARSE | |
62 | */ | |
63 | static loff_t | |
64 | filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset) | |
65 | { | |
66 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
67 | ||
68 | switch (flseg->stripe_type) { | |
69 | case STRIPE_SPARSE: | |
70 | return offset; | |
71 | ||
72 | case STRIPE_DENSE: | |
73 | return filelayout_get_dense_offset(flseg, offset); | |
74 | } | |
75 | ||
76 | BUG(); | |
77 | } | |
78 | ||
cbdabc7f AA |
79 | /* For data server errors we don't recover from */ |
80 | static void | |
81 | filelayout_set_lo_fail(struct pnfs_layout_segment *lseg) | |
82 | { | |
83 | if (lseg->pls_range.iomode == IOMODE_RW) { | |
84 | dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__); | |
85 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
86 | } else { | |
87 | dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__); | |
88 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
89 | } | |
90 | } | |
91 | ||
92 | static int filelayout_async_handle_error(struct rpc_task *task, | |
93 | struct nfs4_state *state, | |
94 | struct nfs_client *clp, | |
95 | int *reset) | |
96 | { | |
97 | if (task->tk_status >= 0) | |
98 | return 0; | |
99 | ||
100 | *reset = 0; | |
101 | ||
102 | switch (task->tk_status) { | |
103 | case -NFS4ERR_BADSESSION: | |
104 | case -NFS4ERR_BADSLOT: | |
105 | case -NFS4ERR_BAD_HIGH_SLOT: | |
106 | case -NFS4ERR_DEADSESSION: | |
107 | case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: | |
108 | case -NFS4ERR_SEQ_FALSE_RETRY: | |
109 | case -NFS4ERR_SEQ_MISORDERED: | |
110 | dprintk("%s ERROR %d, Reset session. Exchangeid " | |
111 | "flags 0x%x\n", __func__, task->tk_status, | |
112 | clp->cl_exchange_flags); | |
113 | nfs4_schedule_session_recovery(clp->cl_session); | |
114 | break; | |
115 | case -NFS4ERR_DELAY: | |
116 | case -NFS4ERR_GRACE: | |
117 | case -EKEYEXPIRED: | |
118 | rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX); | |
119 | break; | |
a8a4ae3a AA |
120 | case -NFS4ERR_RETRY_UNCACHED_REP: |
121 | break; | |
cbdabc7f AA |
122 | default: |
123 | dprintk("%s DS error. Retry through MDS %d\n", __func__, | |
124 | task->tk_status); | |
125 | *reset = 1; | |
126 | break; | |
127 | } | |
128 | task->tk_status = 0; | |
129 | return -EAGAIN; | |
130 | } | |
131 | ||
132 | /* NFS_PROTO call done callback routines */ | |
133 | ||
134 | static int filelayout_read_done_cb(struct rpc_task *task, | |
135 | struct nfs_read_data *data) | |
136 | { | |
137 | struct nfs_client *clp = data->ds_clp; | |
138 | int reset = 0; | |
139 | ||
140 | dprintk("%s DS read\n", __func__); | |
141 | ||
142 | if (filelayout_async_handle_error(task, data->args.context->state, | |
143 | data->ds_clp, &reset) == -EAGAIN) { | |
144 | dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n", | |
145 | __func__, data->ds_clp, data->ds_clp->cl_session); | |
146 | if (reset) { | |
147 | filelayout_set_lo_fail(data->lseg); | |
148 | nfs4_reset_read(task, data); | |
149 | clp = NFS_SERVER(data->inode)->nfs_client; | |
150 | } | |
151 | nfs_restart_rpc(task, clp); | |
152 | return -EAGAIN; | |
153 | } | |
154 | ||
155 | return 0; | |
156 | } | |
157 | ||
863a3c6c AA |
158 | /* |
159 | * We reference the rpc_cred of the first WRITE that triggers the need for | |
160 | * a LAYOUTCOMMIT, and use it to send the layoutcommit compound. | |
161 | * rfc5661 is not clear about which credential should be used. | |
162 | */ | |
163 | static void | |
164 | filelayout_set_layoutcommit(struct nfs_write_data *wdata) | |
165 | { | |
166 | if (FILELAYOUT_LSEG(wdata->lseg)->commit_through_mds || | |
167 | wdata->res.verf->committed == NFS_FILE_SYNC) | |
168 | return; | |
169 | ||
170 | pnfs_set_layoutcommit(wdata); | |
171 | dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino, | |
172 | (unsigned long) wdata->lseg->pls_end_pos); | |
173 | } | |
174 | ||
dc70d7b3 AA |
175 | /* |
176 | * Call ops for the async read/write cases | |
177 | * In the case of dense layouts, the offset needs to be reset to its | |
178 | * original value. | |
179 | */ | |
180 | static void filelayout_read_prepare(struct rpc_task *task, void *data) | |
181 | { | |
182 | struct nfs_read_data *rdata = (struct nfs_read_data *)data; | |
183 | ||
cbdabc7f AA |
184 | rdata->read_done_cb = filelayout_read_done_cb; |
185 | ||
dc70d7b3 AA |
186 | if (nfs41_setup_sequence(rdata->ds_clp->cl_session, |
187 | &rdata->args.seq_args, &rdata->res.seq_res, | |
188 | 0, task)) | |
189 | return; | |
190 | ||
191 | rpc_call_start(task); | |
192 | } | |
193 | ||
194 | static void filelayout_read_call_done(struct rpc_task *task, void *data) | |
195 | { | |
196 | struct nfs_read_data *rdata = (struct nfs_read_data *)data; | |
197 | ||
198 | dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status); | |
199 | ||
200 | /* Note this may cause RPC to be resent */ | |
201 | rdata->mds_ops->rpc_call_done(task, data); | |
202 | } | |
203 | ||
204 | static void filelayout_read_release(void *data) | |
205 | { | |
206 | struct nfs_read_data *rdata = (struct nfs_read_data *)data; | |
207 | ||
208 | rdata->mds_ops->rpc_release(data); | |
209 | } | |
210 | ||
a69aef14 FI |
211 | static int filelayout_write_done_cb(struct rpc_task *task, |
212 | struct nfs_write_data *data) | |
213 | { | |
214 | int reset = 0; | |
215 | ||
216 | if (filelayout_async_handle_error(task, data->args.context->state, | |
217 | data->ds_clp, &reset) == -EAGAIN) { | |
218 | struct nfs_client *clp; | |
219 | ||
220 | dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n", | |
221 | __func__, data->ds_clp, data->ds_clp->cl_session); | |
222 | if (reset) { | |
223 | filelayout_set_lo_fail(data->lseg); | |
224 | nfs4_reset_write(task, data); | |
225 | clp = NFS_SERVER(data->inode)->nfs_client; | |
226 | } else | |
227 | clp = data->ds_clp; | |
228 | nfs_restart_rpc(task, clp); | |
229 | return -EAGAIN; | |
230 | } | |
231 | ||
863a3c6c | 232 | filelayout_set_layoutcommit(data); |
a69aef14 FI |
233 | return 0; |
234 | } | |
235 | ||
e0c2b380 FI |
236 | /* Fake up some data that will cause nfs_commit_release to retry the writes. */ |
237 | static void prepare_to_resend_writes(struct nfs_write_data *data) | |
238 | { | |
239 | struct nfs_page *first = nfs_list_entry(data->pages.next); | |
240 | ||
241 | data->task.tk_status = 0; | |
242 | memcpy(data->verf.verifier, first->wb_verf.verifier, | |
243 | sizeof(first->wb_verf.verifier)); | |
244 | data->verf.verifier[0]++; /* ensure verifier mismatch */ | |
245 | } | |
246 | ||
247 | static int filelayout_commit_done_cb(struct rpc_task *task, | |
248 | struct nfs_write_data *data) | |
249 | { | |
250 | int reset = 0; | |
251 | ||
252 | if (filelayout_async_handle_error(task, data->args.context->state, | |
253 | data->ds_clp, &reset) == -EAGAIN) { | |
254 | dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n", | |
255 | __func__, data->ds_clp, data->ds_clp->cl_session); | |
256 | if (reset) { | |
257 | prepare_to_resend_writes(data); | |
258 | filelayout_set_lo_fail(data->lseg); | |
259 | } else | |
260 | nfs_restart_rpc(task, data->ds_clp); | |
261 | return -EAGAIN; | |
262 | } | |
263 | ||
264 | return 0; | |
265 | } | |
266 | ||
a69aef14 FI |
267 | static void filelayout_write_prepare(struct rpc_task *task, void *data) |
268 | { | |
269 | struct nfs_write_data *wdata = (struct nfs_write_data *)data; | |
270 | ||
271 | if (nfs41_setup_sequence(wdata->ds_clp->cl_session, | |
272 | &wdata->args.seq_args, &wdata->res.seq_res, | |
273 | 0, task)) | |
274 | return; | |
275 | ||
276 | rpc_call_start(task); | |
277 | } | |
278 | ||
279 | static void filelayout_write_call_done(struct rpc_task *task, void *data) | |
280 | { | |
281 | struct nfs_write_data *wdata = (struct nfs_write_data *)data; | |
282 | ||
283 | /* Note this may cause RPC to be resent */ | |
284 | wdata->mds_ops->rpc_call_done(task, data); | |
285 | } | |
286 | ||
287 | static void filelayout_write_release(void *data) | |
288 | { | |
289 | struct nfs_write_data *wdata = (struct nfs_write_data *)data; | |
290 | ||
291 | wdata->mds_ops->rpc_release(data); | |
292 | } | |
293 | ||
e0c2b380 FI |
294 | static void filelayout_commit_release(void *data) |
295 | { | |
296 | struct nfs_write_data *wdata = (struct nfs_write_data *)data; | |
297 | ||
298 | nfs_commit_release_pages(wdata); | |
299 | if (atomic_dec_and_test(&NFS_I(wdata->inode)->commits_outstanding)) | |
300 | nfs_commit_clear_lock(NFS_I(wdata->inode)); | |
301 | nfs_commitdata_release(wdata); | |
302 | } | |
303 | ||
dc70d7b3 AA |
304 | struct rpc_call_ops filelayout_read_call_ops = { |
305 | .rpc_call_prepare = filelayout_read_prepare, | |
306 | .rpc_call_done = filelayout_read_call_done, | |
307 | .rpc_release = filelayout_read_release, | |
308 | }; | |
309 | ||
a69aef14 FI |
310 | struct rpc_call_ops filelayout_write_call_ops = { |
311 | .rpc_call_prepare = filelayout_write_prepare, | |
312 | .rpc_call_done = filelayout_write_call_done, | |
313 | .rpc_release = filelayout_write_release, | |
314 | }; | |
315 | ||
e0c2b380 FI |
316 | struct rpc_call_ops filelayout_commit_call_ops = { |
317 | .rpc_call_prepare = filelayout_write_prepare, | |
318 | .rpc_call_done = filelayout_write_call_done, | |
319 | .rpc_release = filelayout_commit_release, | |
320 | }; | |
321 | ||
dc70d7b3 AA |
322 | static enum pnfs_try_status |
323 | filelayout_read_pagelist(struct nfs_read_data *data) | |
324 | { | |
325 | struct pnfs_layout_segment *lseg = data->lseg; | |
326 | struct nfs4_pnfs_ds *ds; | |
327 | loff_t offset = data->args.offset; | |
328 | u32 j, idx; | |
329 | struct nfs_fh *fh; | |
330 | int status; | |
331 | ||
332 | dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n", | |
333 | __func__, data->inode->i_ino, | |
334 | data->args.pgbase, (size_t)data->args.count, offset); | |
335 | ||
336 | /* Retrieve the correct rpc_client for the byte range */ | |
337 | j = nfs4_fl_calc_j_index(lseg, offset); | |
338 | idx = nfs4_fl_calc_ds_index(lseg, j); | |
339 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
340 | if (!ds) { | |
568e8c49 AA |
341 | /* Either layout fh index faulty, or ds connect failed */ |
342 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
343 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
dc70d7b3 AA |
344 | return PNFS_NOT_ATTEMPTED; |
345 | } | |
346 | dprintk("%s USE DS:ip %x %hu\n", __func__, | |
347 | ntohl(ds->ds_ip_addr), ntohs(ds->ds_port)); | |
348 | ||
349 | /* No multipath support. Use first DS */ | |
350 | data->ds_clp = ds->ds_clp; | |
351 | fh = nfs4_fl_select_ds_fh(lseg, j); | |
352 | if (fh) | |
353 | data->args.fh = fh; | |
354 | ||
355 | data->args.offset = filelayout_get_dserver_offset(lseg, offset); | |
356 | data->mds_offset = offset; | |
357 | ||
358 | /* Perform an asynchronous read to ds */ | |
359 | status = nfs_initiate_read(data, ds->ds_clp->cl_rpcclient, | |
360 | &filelayout_read_call_ops); | |
361 | BUG_ON(status != 0); | |
362 | return PNFS_ATTEMPTED; | |
363 | } | |
364 | ||
a69aef14 | 365 | /* Perform async writes. */ |
0382b744 AA |
366 | static enum pnfs_try_status |
367 | filelayout_write_pagelist(struct nfs_write_data *data, int sync) | |
368 | { | |
a69aef14 FI |
369 | struct pnfs_layout_segment *lseg = data->lseg; |
370 | struct nfs4_pnfs_ds *ds; | |
371 | loff_t offset = data->args.offset; | |
372 | u32 j, idx; | |
373 | struct nfs_fh *fh; | |
374 | int status; | |
375 | ||
376 | /* Retrieve the correct rpc_client for the byte range */ | |
377 | j = nfs4_fl_calc_j_index(lseg, offset); | |
378 | idx = nfs4_fl_calc_ds_index(lseg, j); | |
379 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
380 | if (!ds) { | |
381 | printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__); | |
382 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
383 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
384 | return PNFS_NOT_ATTEMPTED; | |
385 | } | |
386 | dprintk("%s ino %lu sync %d req %Zu@%llu DS:%x:%hu\n", __func__, | |
387 | data->inode->i_ino, sync, (size_t) data->args.count, offset, | |
388 | ntohl(ds->ds_ip_addr), ntohs(ds->ds_port)); | |
389 | ||
a69aef14 FI |
390 | data->write_done_cb = filelayout_write_done_cb; |
391 | data->ds_clp = ds->ds_clp; | |
392 | fh = nfs4_fl_select_ds_fh(lseg, j); | |
393 | if (fh) | |
394 | data->args.fh = fh; | |
395 | /* | |
396 | * Get the file offset on the dserver. Set the write offset to | |
397 | * this offset and save the original offset. | |
398 | */ | |
399 | data->args.offset = filelayout_get_dserver_offset(lseg, offset); | |
400 | data->mds_offset = offset; | |
401 | ||
402 | /* Perform an asynchronous write */ | |
403 | status = nfs_initiate_write(data, ds->ds_clp->cl_rpcclient, | |
404 | &filelayout_write_call_ops, sync); | |
405 | BUG_ON(status != 0); | |
406 | return PNFS_ATTEMPTED; | |
0382b744 AA |
407 | } |
408 | ||
16b374ca AA |
409 | /* |
410 | * filelayout_check_layout() | |
411 | * | |
412 | * Make sure layout segment parameters are sane WRT the device. | |
413 | * At this point no generic layer initialization of the lseg has occurred, | |
414 | * and nothing has been added to the layout_hdr cache. | |
415 | * | |
416 | */ | |
417 | static int | |
418 | filelayout_check_layout(struct pnfs_layout_hdr *lo, | |
419 | struct nfs4_filelayout_segment *fl, | |
420 | struct nfs4_layoutget_res *lgr, | |
a75b9df9 TM |
421 | struct nfs4_deviceid *id, |
422 | gfp_t gfp_flags) | |
16b374ca | 423 | { |
a1eaecbc | 424 | struct nfs4_deviceid_node *d; |
16b374ca AA |
425 | struct nfs4_file_layout_dsaddr *dsaddr; |
426 | int status = -EINVAL; | |
b7edfaa1 | 427 | struct nfs_server *nfss = NFS_SERVER(lo->plh_inode); |
16b374ca AA |
428 | |
429 | dprintk("--> %s\n", __func__); | |
430 | ||
431 | if (fl->pattern_offset > lgr->range.offset) { | |
3b6445a6 | 432 | dprintk("%s pattern_offset %lld too large\n", |
16b374ca AA |
433 | __func__, fl->pattern_offset); |
434 | goto out; | |
435 | } | |
436 | ||
75247aff BH |
437 | if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) { |
438 | dprintk("%s Invalid stripe unit (%u)\n", | |
16b374ca AA |
439 | __func__, fl->stripe_unit); |
440 | goto out; | |
441 | } | |
442 | ||
443 | /* find and reference the deviceid */ | |
a1eaecbc BH |
444 | d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->nfs_client, id); |
445 | if (d == NULL) { | |
a75b9df9 | 446 | dsaddr = get_device_info(lo->plh_inode, id, gfp_flags); |
16b374ca AA |
447 | if (dsaddr == NULL) |
448 | goto out; | |
a1eaecbc BH |
449 | } else |
450 | dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node); | |
16b374ca AA |
451 | fl->dsaddr = dsaddr; |
452 | ||
453 | if (fl->first_stripe_index < 0 || | |
454 | fl->first_stripe_index >= dsaddr->stripe_count) { | |
455 | dprintk("%s Bad first_stripe_index %d\n", | |
456 | __func__, fl->first_stripe_index); | |
457 | goto out_put; | |
458 | } | |
459 | ||
460 | if ((fl->stripe_type == STRIPE_SPARSE && | |
461 | fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) || | |
462 | (fl->stripe_type == STRIPE_DENSE && | |
463 | fl->num_fh != dsaddr->stripe_count)) { | |
464 | dprintk("%s num_fh %u not valid for given packing\n", | |
465 | __func__, fl->num_fh); | |
466 | goto out_put; | |
467 | } | |
468 | ||
469 | if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) { | |
470 | dprintk("%s Stripe unit (%u) not aligned with rsize %u " | |
471 | "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize, | |
472 | nfss->wsize); | |
473 | } | |
474 | ||
475 | status = 0; | |
476 | out: | |
477 | dprintk("--> %s returns %d\n", __func__, status); | |
478 | return status; | |
479 | out_put: | |
ea8eecdd | 480 | nfs4_fl_put_deviceid(dsaddr); |
16b374ca AA |
481 | goto out; |
482 | } | |
483 | ||
484 | static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl) | |
485 | { | |
486 | int i; | |
487 | ||
488 | for (i = 0; i < fl->num_fh; i++) { | |
489 | if (!fl->fh_array[i]) | |
490 | break; | |
491 | kfree(fl->fh_array[i]); | |
492 | } | |
493 | kfree(fl->fh_array); | |
494 | fl->fh_array = NULL; | |
495 | } | |
496 | ||
497 | static void | |
498 | _filelayout_free_lseg(struct nfs4_filelayout_segment *fl) | |
499 | { | |
500 | filelayout_free_fh_array(fl); | |
501 | kfree(fl); | |
502 | } | |
503 | ||
504 | static int | |
505 | filelayout_decode_layout(struct pnfs_layout_hdr *flo, | |
506 | struct nfs4_filelayout_segment *fl, | |
507 | struct nfs4_layoutget_res *lgr, | |
a75b9df9 TM |
508 | struct nfs4_deviceid *id, |
509 | gfp_t gfp_flags) | |
16b374ca | 510 | { |
35124a09 WAA |
511 | struct xdr_stream stream; |
512 | struct xdr_buf buf = { | |
513 | .pages = lgr->layoutp->pages, | |
514 | .page_len = lgr->layoutp->len, | |
515 | .buflen = lgr->layoutp->len, | |
516 | .len = lgr->layoutp->len, | |
517 | }; | |
518 | struct page *scratch; | |
519 | __be32 *p; | |
16b374ca AA |
520 | uint32_t nfl_util; |
521 | int i; | |
522 | ||
523 | dprintk("%s: set_layout_map Begin\n", __func__); | |
524 | ||
a75b9df9 | 525 | scratch = alloc_page(gfp_flags); |
35124a09 WAA |
526 | if (!scratch) |
527 | return -ENOMEM; | |
528 | ||
529 | xdr_init_decode(&stream, &buf, NULL); | |
530 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); | |
531 | ||
532 | /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8), | |
533 | * num_fh (4) */ | |
534 | p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20); | |
535 | if (unlikely(!p)) | |
536 | goto out_err; | |
537 | ||
16b374ca AA |
538 | memcpy(id, p, sizeof(*id)); |
539 | p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE); | |
a1eaecbc | 540 | nfs4_print_deviceid(id); |
16b374ca AA |
541 | |
542 | nfl_util = be32_to_cpup(p++); | |
543 | if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS) | |
544 | fl->commit_through_mds = 1; | |
545 | if (nfl_util & NFL4_UFLG_DENSE) | |
546 | fl->stripe_type = STRIPE_DENSE; | |
547 | else | |
548 | fl->stripe_type = STRIPE_SPARSE; | |
549 | fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK; | |
550 | ||
551 | fl->first_stripe_index = be32_to_cpup(p++); | |
552 | p = xdr_decode_hyper(p, &fl->pattern_offset); | |
553 | fl->num_fh = be32_to_cpup(p++); | |
554 | ||
555 | dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n", | |
556 | __func__, nfl_util, fl->num_fh, fl->first_stripe_index, | |
557 | fl->pattern_offset); | |
558 | ||
35124a09 WAA |
559 | if (!fl->num_fh) |
560 | goto out_err; | |
561 | ||
16b374ca | 562 | fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *), |
a75b9df9 | 563 | gfp_flags); |
16b374ca | 564 | if (!fl->fh_array) |
35124a09 | 565 | goto out_err; |
16b374ca AA |
566 | |
567 | for (i = 0; i < fl->num_fh; i++) { | |
568 | /* Do we want to use a mempool here? */ | |
a75b9df9 | 569 | fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags); |
35124a09 WAA |
570 | if (!fl->fh_array[i]) |
571 | goto out_err_free; | |
572 | ||
573 | p = xdr_inline_decode(&stream, 4); | |
574 | if (unlikely(!p)) | |
575 | goto out_err_free; | |
16b374ca AA |
576 | fl->fh_array[i]->size = be32_to_cpup(p++); |
577 | if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) { | |
578 | printk(KERN_ERR "Too big fh %d received %d\n", | |
579 | i, fl->fh_array[i]->size); | |
35124a09 | 580 | goto out_err_free; |
16b374ca | 581 | } |
35124a09 WAA |
582 | |
583 | p = xdr_inline_decode(&stream, fl->fh_array[i]->size); | |
584 | if (unlikely(!p)) | |
585 | goto out_err_free; | |
16b374ca | 586 | memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size); |
16b374ca AA |
587 | dprintk("DEBUG: %s: fh len %d\n", __func__, |
588 | fl->fh_array[i]->size); | |
589 | } | |
590 | ||
35124a09 | 591 | __free_page(scratch); |
16b374ca | 592 | return 0; |
35124a09 WAA |
593 | |
594 | out_err_free: | |
595 | filelayout_free_fh_array(fl); | |
596 | out_err: | |
597 | __free_page(scratch); | |
598 | return -EIO; | |
16b374ca AA |
599 | } |
600 | ||
c879513e FI |
601 | static void |
602 | filelayout_free_lseg(struct pnfs_layout_segment *lseg) | |
603 | { | |
604 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); | |
605 | ||
606 | dprintk("--> %s\n", __func__); | |
607 | nfs4_fl_put_deviceid(fl->dsaddr); | |
425eb736 | 608 | kfree(fl->commit_buckets); |
c879513e FI |
609 | _filelayout_free_lseg(fl); |
610 | } | |
611 | ||
16b374ca AA |
612 | static struct pnfs_layout_segment * |
613 | filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid, | |
a75b9df9 TM |
614 | struct nfs4_layoutget_res *lgr, |
615 | gfp_t gfp_flags) | |
16b374ca AA |
616 | { |
617 | struct nfs4_filelayout_segment *fl; | |
618 | int rc; | |
619 | struct nfs4_deviceid id; | |
620 | ||
621 | dprintk("--> %s\n", __func__); | |
a75b9df9 | 622 | fl = kzalloc(sizeof(*fl), gfp_flags); |
16b374ca AA |
623 | if (!fl) |
624 | return NULL; | |
625 | ||
a75b9df9 TM |
626 | rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags); |
627 | if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) { | |
16b374ca AA |
628 | _filelayout_free_lseg(fl); |
629 | return NULL; | |
630 | } | |
425eb736 FI |
631 | |
632 | /* This assumes there is only one IOMODE_RW lseg. What | |
633 | * we really want to do is have a layout_hdr level | |
634 | * dictionary of <multipath_list4, fh> keys, each | |
635 | * associated with a struct list_head, populated by calls | |
636 | * to filelayout_write_pagelist(). | |
637 | * */ | |
638 | if ((!fl->commit_through_mds) && (lgr->range.iomode == IOMODE_RW)) { | |
639 | int i; | |
640 | int size = (fl->stripe_type == STRIPE_SPARSE) ? | |
641 | fl->dsaddr->ds_num : fl->dsaddr->stripe_count; | |
642 | ||
a75b9df9 | 643 | fl->commit_buckets = kcalloc(size, sizeof(struct list_head), gfp_flags); |
425eb736 FI |
644 | if (!fl->commit_buckets) { |
645 | filelayout_free_lseg(&fl->generic_hdr); | |
646 | return NULL; | |
647 | } | |
648 | fl->number_of_buckets = size; | |
649 | for (i = 0; i < size; i++) | |
650 | INIT_LIST_HEAD(&fl->commit_buckets[i]); | |
651 | } | |
16b374ca AA |
652 | return &fl->generic_hdr; |
653 | } | |
654 | ||
94ad1c80 FI |
655 | /* |
656 | * filelayout_pg_test(). Called by nfs_can_coalesce_requests() | |
657 | * | |
658 | * return 1 : coalesce page | |
659 | * return 0 : don't coalesce page | |
660 | */ | |
661 | int | |
662 | filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, | |
663 | struct nfs_page *req) | |
664 | { | |
665 | u64 p_stripe, r_stripe; | |
666 | u32 stripe_unit; | |
667 | ||
668 | if (!pgio->pg_lseg) | |
669 | return 1; | |
670 | p_stripe = (u64)prev->wb_index << PAGE_CACHE_SHIFT; | |
671 | r_stripe = (u64)req->wb_index << PAGE_CACHE_SHIFT; | |
672 | stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit; | |
673 | ||
674 | do_div(p_stripe, stripe_unit); | |
675 | do_div(r_stripe, stripe_unit); | |
676 | ||
677 | return (p_stripe == r_stripe); | |
678 | } | |
679 | ||
e0c2b380 FI |
680 | static bool filelayout_mark_pnfs_commit(struct pnfs_layout_segment *lseg) |
681 | { | |
682 | return !FILELAYOUT_LSEG(lseg)->commit_through_mds; | |
683 | } | |
684 | ||
685 | static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j) | |
686 | { | |
687 | if (fl->stripe_type == STRIPE_SPARSE) | |
688 | return nfs4_fl_calc_ds_index(&fl->generic_hdr, j); | |
689 | else | |
690 | return j; | |
691 | } | |
692 | ||
693 | struct list_head *filelayout_choose_commit_list(struct nfs_page *req) | |
694 | { | |
695 | struct pnfs_layout_segment *lseg = req->wb_commit_lseg; | |
696 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); | |
697 | u32 i, j; | |
698 | struct list_head *list; | |
699 | ||
700 | /* Note that we are calling nfs4_fl_calc_j_index on each page | |
701 | * that ends up being committed to a data server. An attractive | |
702 | * alternative is to add a field to nfs_write_data and nfs_page | |
703 | * to store the value calculated in filelayout_write_pagelist | |
704 | * and just use that here. | |
705 | */ | |
706 | j = nfs4_fl_calc_j_index(lseg, | |
707 | (loff_t)req->wb_index << PAGE_CACHE_SHIFT); | |
708 | i = select_bucket_index(fl, j); | |
709 | list = &fl->commit_buckets[i]; | |
710 | if (list_empty(list)) { | |
711 | /* Non-empty buckets hold a reference on the lseg */ | |
712 | get_lseg(lseg); | |
713 | } | |
714 | return list; | |
715 | } | |
716 | ||
717 | static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i) | |
718 | { | |
719 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
720 | ||
721 | if (flseg->stripe_type == STRIPE_SPARSE) | |
722 | return i; | |
723 | else | |
724 | return nfs4_fl_calc_ds_index(lseg, i); | |
725 | } | |
726 | ||
727 | static struct nfs_fh * | |
728 | select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i) | |
729 | { | |
730 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
731 | ||
732 | if (flseg->stripe_type == STRIPE_SPARSE) { | |
733 | if (flseg->num_fh == 1) | |
734 | i = 0; | |
735 | else if (flseg->num_fh == 0) | |
736 | /* Use the MDS OPEN fh set in nfs_read_rpcsetup */ | |
737 | return NULL; | |
738 | } | |
739 | return flseg->fh_array[i]; | |
740 | } | |
741 | ||
742 | static int filelayout_initiate_commit(struct nfs_write_data *data, int how) | |
743 | { | |
744 | struct pnfs_layout_segment *lseg = data->lseg; | |
745 | struct nfs4_pnfs_ds *ds; | |
746 | u32 idx; | |
747 | struct nfs_fh *fh; | |
748 | ||
749 | idx = calc_ds_index_from_commit(lseg, data->ds_commit_index); | |
750 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
751 | if (!ds) { | |
752 | printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__); | |
753 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
754 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
755 | prepare_to_resend_writes(data); | |
756 | data->mds_ops->rpc_release(data); | |
757 | return -EAGAIN; | |
758 | } | |
759 | dprintk("%s ino %lu, how %d\n", __func__, data->inode->i_ino, how); | |
760 | data->write_done_cb = filelayout_commit_done_cb; | |
761 | data->ds_clp = ds->ds_clp; | |
762 | fh = select_ds_fh_from_commit(lseg, data->ds_commit_index); | |
763 | if (fh) | |
764 | data->args.fh = fh; | |
765 | return nfs_initiate_commit(data, ds->ds_clp->cl_rpcclient, | |
766 | &filelayout_commit_call_ops, how); | |
767 | } | |
768 | ||
769 | /* | |
770 | * This is only useful while we are using whole file layouts. | |
771 | */ | |
772 | static struct pnfs_layout_segment *find_only_write_lseg(struct inode *inode) | |
773 | { | |
774 | struct pnfs_layout_segment *lseg, *rv = NULL; | |
775 | ||
776 | spin_lock(&inode->i_lock); | |
777 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) | |
778 | if (lseg->pls_range.iomode == IOMODE_RW) | |
779 | rv = get_lseg(lseg); | |
780 | spin_unlock(&inode->i_lock); | |
781 | return rv; | |
782 | } | |
783 | ||
784 | static int alloc_ds_commits(struct inode *inode, struct list_head *list) | |
785 | { | |
786 | struct pnfs_layout_segment *lseg; | |
787 | struct nfs4_filelayout_segment *fl; | |
788 | struct nfs_write_data *data; | |
789 | int i, j; | |
790 | ||
791 | /* Won't need this when non-whole file layout segments are supported | |
792 | * instead we will use a pnfs_layout_hdr structure */ | |
793 | lseg = find_only_write_lseg(inode); | |
794 | if (!lseg) | |
795 | return 0; | |
796 | fl = FILELAYOUT_LSEG(lseg); | |
797 | for (i = 0; i < fl->number_of_buckets; i++) { | |
798 | if (list_empty(&fl->commit_buckets[i])) | |
799 | continue; | |
800 | data = nfs_commitdata_alloc(); | |
801 | if (!data) | |
802 | goto out_bad; | |
803 | data->ds_commit_index = i; | |
804 | data->lseg = lseg; | |
805 | list_add(&data->pages, list); | |
806 | } | |
807 | put_lseg(lseg); | |
808 | return 0; | |
809 | ||
810 | out_bad: | |
811 | for (j = i; j < fl->number_of_buckets; j++) { | |
812 | if (list_empty(&fl->commit_buckets[i])) | |
813 | continue; | |
814 | nfs_retry_commit(&fl->commit_buckets[i], lseg); | |
815 | put_lseg(lseg); /* associated with emptying bucket */ | |
816 | } | |
817 | put_lseg(lseg); | |
818 | /* Caller will clean up entries put on list */ | |
819 | return -ENOMEM; | |
820 | } | |
821 | ||
822 | /* This follows nfs_commit_list pretty closely */ | |
823 | static int | |
824 | filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, | |
825 | int how) | |
826 | { | |
827 | struct nfs_write_data *data, *tmp; | |
828 | LIST_HEAD(list); | |
829 | ||
830 | if (!list_empty(mds_pages)) { | |
831 | data = nfs_commitdata_alloc(); | |
832 | if (!data) | |
833 | goto out_bad; | |
834 | data->lseg = NULL; | |
835 | list_add(&data->pages, &list); | |
836 | } | |
837 | ||
838 | if (alloc_ds_commits(inode, &list)) | |
839 | goto out_bad; | |
840 | ||
841 | list_for_each_entry_safe(data, tmp, &list, pages) { | |
842 | list_del_init(&data->pages); | |
843 | atomic_inc(&NFS_I(inode)->commits_outstanding); | |
844 | if (!data->lseg) { | |
845 | nfs_init_commit(data, mds_pages, NULL); | |
846 | nfs_initiate_commit(data, NFS_CLIENT(inode), | |
847 | data->mds_ops, how); | |
848 | } else { | |
849 | nfs_init_commit(data, &FILELAYOUT_LSEG(data->lseg)->commit_buckets[data->ds_commit_index], data->lseg); | |
850 | filelayout_initiate_commit(data, how); | |
851 | } | |
852 | } | |
853 | return 0; | |
854 | out_bad: | |
855 | list_for_each_entry_safe(data, tmp, &list, pages) { | |
856 | nfs_retry_commit(&data->pages, data->lseg); | |
857 | list_del_init(&data->pages); | |
858 | nfs_commit_free(data); | |
859 | } | |
860 | nfs_retry_commit(mds_pages, NULL); | |
861 | nfs_commit_clear_lock(NFS_I(inode)); | |
862 | return -ENOMEM; | |
863 | } | |
864 | ||
7ab672ce | 865 | static struct pnfs_layoutdriver_type filelayout_type = { |
ea8eecdd CH |
866 | .id = LAYOUT_NFSV4_1_FILES, |
867 | .name = "LAYOUT_NFSV4_1_FILES", | |
868 | .owner = THIS_MODULE, | |
869 | .alloc_lseg = filelayout_alloc_lseg, | |
870 | .free_lseg = filelayout_free_lseg, | |
94ad1c80 | 871 | .pg_test = filelayout_pg_test, |
e0c2b380 FI |
872 | .mark_pnfs_commit = filelayout_mark_pnfs_commit, |
873 | .choose_commit_list = filelayout_choose_commit_list, | |
874 | .commit_pagelist = filelayout_commit_pagelist, | |
dc70d7b3 | 875 | .read_pagelist = filelayout_read_pagelist, |
0382b744 | 876 | .write_pagelist = filelayout_write_pagelist, |
7ab672ce DH |
877 | }; |
878 | ||
879 | static int __init nfs4filelayout_init(void) | |
880 | { | |
881 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n", | |
882 | __func__); | |
883 | return pnfs_register_layoutdriver(&filelayout_type); | |
884 | } | |
885 | ||
886 | static void __exit nfs4filelayout_exit(void) | |
887 | { | |
888 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n", | |
889 | __func__); | |
890 | pnfs_unregister_layoutdriver(&filelayout_type); | |
891 | } | |
892 | ||
893 | module_init(nfs4filelayout_init); | |
894 | module_exit(nfs4filelayout_exit); |