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>
33#include <linux/nfs_page.h>
34#include <linux/module.h>
35#include <linux/backing-dev.h>
36
37#include <linux/sunrpc/metrics.h>
38
39#include "../nfs4session.h"
40#include "../internal.h"
41#include "../delegation.h"
42#include "filelayout.h"
43#include "../nfs4trace.h"
44
45#define NFSDBG_FACILITY NFSDBG_PNFS_LD
46
47MODULE_LICENSE("GPL");
48MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
49MODULE_DESCRIPTION("The NFSv4 file layout driver");
50
51#define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
52static const struct pnfs_commit_ops filelayout_commit_ops;
53
54static loff_t
55filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
56 loff_t offset)
57{
58 u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
59 u64 stripe_no;
60 u32 rem;
61
62 offset -= flseg->pattern_offset;
63 stripe_no = div_u64(dividend: offset, divisor: stripe_width);
64 div_u64_rem(dividend: offset, divisor: flseg->stripe_unit, remainder: &rem);
65
66 return stripe_no * flseg->stripe_unit + rem;
67}
68
69/* This function is used by the layout driver to calculate the
70 * offset of the file on the dserver based on whether the
71 * layout type is STRIPE_DENSE or STRIPE_SPARSE
72 */
73static loff_t
74filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
75{
76 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
77
78 switch (flseg->stripe_type) {
79 case STRIPE_SPARSE:
80 return offset;
81
82 case STRIPE_DENSE:
83 return filelayout_get_dense_offset(flseg, offset);
84 }
85
86 BUG();
87}
88
89static void filelayout_reset_write(struct nfs_pgio_header *hdr)
90{
91 struct rpc_task *task = &hdr->task;
92
93 if (!test_and_set_bit(nr: NFS_IOHDR_REDO, addr: &hdr->flags)) {
94 dprintk("%s Reset task %5u for i/o through MDS "
95 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
96 hdr->task.tk_pid,
97 hdr->inode->i_sb->s_id,
98 (unsigned long long)NFS_FILEID(hdr->inode),
99 hdr->args.count,
100 (unsigned long long)hdr->args.offset);
101
102 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
103 }
104}
105
106static void filelayout_reset_read(struct nfs_pgio_header *hdr)
107{
108 struct rpc_task *task = &hdr->task;
109
110 if (!test_and_set_bit(nr: NFS_IOHDR_REDO, addr: &hdr->flags)) {
111 dprintk("%s Reset task %5u for i/o through MDS "
112 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
113 hdr->task.tk_pid,
114 hdr->inode->i_sb->s_id,
115 (unsigned long long)NFS_FILEID(hdr->inode),
116 hdr->args.count,
117 (unsigned long long)hdr->args.offset);
118
119 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
120 }
121}
122
123static int filelayout_async_handle_error(struct rpc_task *task,
124 struct nfs4_state *state,
125 struct nfs_client *clp,
126 struct pnfs_layout_segment *lseg)
127{
128 struct pnfs_layout_hdr *lo = lseg->pls_layout;
129 struct inode *inode = lo->plh_inode;
130 struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
131 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
132
133 if (task->tk_status >= 0)
134 return 0;
135
136 switch (task->tk_status) {
137 /* DS session errors */
138 case -NFS4ERR_BADSESSION:
139 case -NFS4ERR_BADSLOT:
140 case -NFS4ERR_BAD_HIGH_SLOT:
141 case -NFS4ERR_DEADSESSION:
142 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
143 case -NFS4ERR_SEQ_FALSE_RETRY:
144 case -NFS4ERR_SEQ_MISORDERED:
145 dprintk("%s ERROR %d, Reset session. Exchangeid "
146 "flags 0x%x\n", __func__, task->tk_status,
147 clp->cl_exchange_flags);
148 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
149 break;
150 case -NFS4ERR_DELAY:
151 case -NFS4ERR_GRACE:
152 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
153 break;
154 case -NFS4ERR_RETRY_UNCACHED_REP:
155 break;
156 /* Invalidate Layout errors */
157 case -NFS4ERR_ACCESS:
158 case -NFS4ERR_PNFS_NO_LAYOUT:
159 case -ESTALE: /* mapped NFS4ERR_STALE */
160 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
161 case -EISDIR: /* mapped NFS4ERR_ISDIR */
162 case -NFS4ERR_FHEXPIRED:
163 case -NFS4ERR_WRONG_TYPE:
164 dprintk("%s Invalid layout error %d\n", __func__,
165 task->tk_status);
166 /*
167 * Destroy layout so new i/o will get a new layout.
168 * Layout will not be destroyed until all current lseg
169 * references are put. Mark layout as invalid to resend failed
170 * i/o and all i/o waiting on the slot table to the MDS until
171 * layout is destroyed and a new valid layout is obtained.
172 */
173 pnfs_destroy_layout(NFS_I(inode));
174 rpc_wake_up(&tbl->slot_tbl_waitq);
175 goto reset;
176 /* RPC connection errors */
177 case -ECONNREFUSED:
178 case -EHOSTDOWN:
179 case -EHOSTUNREACH:
180 case -ENETUNREACH:
181 case -EIO:
182 case -ETIMEDOUT:
183 case -EPIPE:
184 case -EPROTO:
185 case -ENODEV:
186 dprintk("%s DS connection error %d\n", __func__,
187 task->tk_status);
188 nfs4_mark_deviceid_unavailable(node: devid);
189 pnfs_error_mark_layout_for_return(inode, lseg);
190 pnfs_set_lo_fail(lseg);
191 rpc_wake_up(&tbl->slot_tbl_waitq);
192 fallthrough;
193 default:
194reset:
195 dprintk("%s Retry through MDS. Error %d\n", __func__,
196 task->tk_status);
197 return -NFS4ERR_RESET_TO_MDS;
198 }
199 task->tk_status = 0;
200 return -EAGAIN;
201}
202
203/* NFS_PROTO call done callback routines */
204
205static int filelayout_read_done_cb(struct rpc_task *task,
206 struct nfs_pgio_header *hdr)
207{
208 int err;
209
210 trace_nfs4_pnfs_read(hdr, error: task->tk_status);
211 err = filelayout_async_handle_error(task, state: hdr->args.context->state,
212 clp: hdr->ds_clp, lseg: hdr->lseg);
213
214 switch (err) {
215 case -NFS4ERR_RESET_TO_MDS:
216 filelayout_reset_read(hdr);
217 return task->tk_status;
218 case -EAGAIN:
219 rpc_restart_call_prepare(task);
220 return -EAGAIN;
221 }
222
223 return 0;
224}
225
226/*
227 * We reference the rpc_cred of the first WRITE that triggers the need for
228 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
229 * rfc5661 is not clear about which credential should be used.
230 */
231static void
232filelayout_set_layoutcommit(struct nfs_pgio_header *hdr)
233{
234 loff_t end_offs = 0;
235
236 if (FILELAYOUT_LSEG(lseg: hdr->lseg)->commit_through_mds ||
237 hdr->res.verf->committed == NFS_FILE_SYNC)
238 return;
239 if (hdr->res.verf->committed == NFS_DATA_SYNC)
240 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
241
242 /* Note: if the write is unstable, don't set end_offs until commit */
243 pnfs_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
244 dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
245 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
246}
247
248bool
249filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node)
250{
251 return filelayout_test_devid_invalid(node) ||
252 nfs4_test_deviceid_unavailable(node);
253}
254
255static bool
256filelayout_reset_to_mds(struct pnfs_layout_segment *lseg)
257{
258 struct nfs4_deviceid_node *node = FILELAYOUT_DEVID_NODE(lseg);
259
260 return filelayout_test_devid_unavailable(node);
261}
262
263/*
264 * Call ops for the async read/write cases
265 * In the case of dense layouts, the offset needs to be reset to its
266 * original value.
267 */
268static void filelayout_read_prepare(struct rpc_task *task, void *data)
269{
270 struct nfs_pgio_header *hdr = data;
271
272 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
273 rpc_exit(task, -EIO);
274 return;
275 }
276 if (filelayout_reset_to_mds(lseg: hdr->lseg)) {
277 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
278 filelayout_reset_read(hdr);
279 rpc_exit(task, 0);
280 return;
281 }
282 hdr->pgio_done_cb = filelayout_read_done_cb;
283
284 if (nfs4_setup_sequence(client: hdr->ds_clp,
285 args: &hdr->args.seq_args,
286 res: &hdr->res.seq_res,
287 task))
288 return;
289 if (nfs4_set_rw_stateid(stateid: &hdr->args.stateid, ctx: hdr->args.context,
290 l_ctx: hdr->args.lock_context, FMODE_READ) == -EIO)
291 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
292}
293
294static void filelayout_read_call_done(struct rpc_task *task, void *data)
295{
296 struct nfs_pgio_header *hdr = data;
297
298 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
299 task->tk_status == 0) {
300 nfs41_sequence_done(task, &hdr->res.seq_res);
301 return;
302 }
303
304 /* Note this may cause RPC to be resent */
305 hdr->mds_ops->rpc_call_done(task, data);
306}
307
308static void filelayout_read_count_stats(struct rpc_task *task, void *data)
309{
310 struct nfs_pgio_header *hdr = data;
311
312 rpc_count_iostats(task, NFS_SERVER(inode: hdr->inode)->client->cl_metrics);
313}
314
315static int filelayout_write_done_cb(struct rpc_task *task,
316 struct nfs_pgio_header *hdr)
317{
318 int err;
319
320 trace_nfs4_pnfs_write(hdr, error: task->tk_status);
321 err = filelayout_async_handle_error(task, state: hdr->args.context->state,
322 clp: hdr->ds_clp, lseg: hdr->lseg);
323
324 switch (err) {
325 case -NFS4ERR_RESET_TO_MDS:
326 filelayout_reset_write(hdr);
327 return task->tk_status;
328 case -EAGAIN:
329 rpc_restart_call_prepare(task);
330 return -EAGAIN;
331 }
332
333 filelayout_set_layoutcommit(hdr);
334
335 /* zero out the fattr */
336 hdr->fattr.valid = 0;
337 if (task->tk_status >= 0)
338 nfs_writeback_update_inode(hdr);
339
340 return 0;
341}
342
343static int filelayout_commit_done_cb(struct rpc_task *task,
344 struct nfs_commit_data *data)
345{
346 int err;
347
348 trace_nfs4_pnfs_commit_ds(data, error: task->tk_status);
349 err = filelayout_async_handle_error(task, NULL, clp: data->ds_clp,
350 lseg: data->lseg);
351
352 switch (err) {
353 case -NFS4ERR_RESET_TO_MDS:
354 pnfs_generic_prepare_to_resend_writes(data);
355 return -EAGAIN;
356 case -EAGAIN:
357 rpc_restart_call_prepare(task);
358 return -EAGAIN;
359 }
360
361 pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
362
363 return 0;
364}
365
366static void filelayout_write_prepare(struct rpc_task *task, void *data)
367{
368 struct nfs_pgio_header *hdr = data;
369
370 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
371 rpc_exit(task, -EIO);
372 return;
373 }
374 if (filelayout_reset_to_mds(lseg: hdr->lseg)) {
375 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
376 filelayout_reset_write(hdr);
377 rpc_exit(task, 0);
378 return;
379 }
380 if (nfs4_setup_sequence(client: hdr->ds_clp,
381 args: &hdr->args.seq_args,
382 res: &hdr->res.seq_res,
383 task))
384 return;
385 if (nfs4_set_rw_stateid(stateid: &hdr->args.stateid, ctx: hdr->args.context,
386 l_ctx: hdr->args.lock_context, FMODE_WRITE) == -EIO)
387 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
388}
389
390static void filelayout_write_call_done(struct rpc_task *task, void *data)
391{
392 struct nfs_pgio_header *hdr = data;
393
394 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
395 task->tk_status == 0) {
396 nfs41_sequence_done(task, &hdr->res.seq_res);
397 return;
398 }
399
400 /* Note this may cause RPC to be resent */
401 hdr->mds_ops->rpc_call_done(task, data);
402}
403
404static void filelayout_write_count_stats(struct rpc_task *task, void *data)
405{
406 struct nfs_pgio_header *hdr = data;
407
408 rpc_count_iostats(task, NFS_SERVER(inode: hdr->inode)->client->cl_metrics);
409}
410
411static void filelayout_commit_prepare(struct rpc_task *task, void *data)
412{
413 struct nfs_commit_data *wdata = data;
414
415 nfs4_setup_sequence(client: wdata->ds_clp,
416 args: &wdata->args.seq_args,
417 res: &wdata->res.seq_res,
418 task);
419}
420
421static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
422{
423 struct nfs_commit_data *cdata = data;
424
425 rpc_count_iostats(task, NFS_SERVER(inode: cdata->inode)->client->cl_metrics);
426}
427
428static const struct rpc_call_ops filelayout_read_call_ops = {
429 .rpc_call_prepare = filelayout_read_prepare,
430 .rpc_call_done = filelayout_read_call_done,
431 .rpc_count_stats = filelayout_read_count_stats,
432 .rpc_release = pnfs_generic_rw_release,
433};
434
435static const struct rpc_call_ops filelayout_write_call_ops = {
436 .rpc_call_prepare = filelayout_write_prepare,
437 .rpc_call_done = filelayout_write_call_done,
438 .rpc_count_stats = filelayout_write_count_stats,
439 .rpc_release = pnfs_generic_rw_release,
440};
441
442static const struct rpc_call_ops filelayout_commit_call_ops = {
443 .rpc_call_prepare = filelayout_commit_prepare,
444 .rpc_call_done = pnfs_generic_write_commit_done,
445 .rpc_count_stats = filelayout_commit_count_stats,
446 .rpc_release = pnfs_generic_commit_release,
447};
448
449static enum pnfs_try_status
450filelayout_read_pagelist(struct nfs_pgio_header *hdr)
451{
452 struct pnfs_layout_segment *lseg = hdr->lseg;
453 struct nfs4_pnfs_ds *ds;
454 struct rpc_clnt *ds_clnt;
455 loff_t offset = hdr->args.offset;
456 u32 j, idx;
457 struct nfs_fh *fh;
458
459 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
460 __func__, hdr->inode->i_ino,
461 hdr->args.pgbase, (size_t)hdr->args.count, offset);
462
463 /* Retrieve the correct rpc_client for the byte range */
464 j = nfs4_fl_calc_j_index(lseg, offset);
465 idx = nfs4_fl_calc_ds_index(lseg, j);
466 ds = nfs4_fl_prepare_ds(lseg, ds_idx: idx);
467 if (!ds)
468 return PNFS_NOT_ATTEMPTED;
469
470 ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
471 if (IS_ERR(ptr: ds_clnt))
472 return PNFS_NOT_ATTEMPTED;
473
474 dprintk("%s USE DS: %s cl_count %d\n", __func__,
475 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count));
476
477 /* No multipath support. Use first DS */
478 refcount_inc(r: &ds->ds_clp->cl_count);
479 hdr->ds_clp = ds->ds_clp;
480 hdr->ds_commit_idx = idx;
481 fh = nfs4_fl_select_ds_fh(lseg, j);
482 if (fh)
483 hdr->args.fh = fh;
484
485 hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
486 hdr->mds_offset = offset;
487
488 /* Perform an asynchronous read to ds */
489 nfs_initiate_pgio(clnt: ds_clnt, hdr, cred: hdr->cred,
490 rpc_ops: NFS_PROTO(inode: hdr->inode), call_ops: &filelayout_read_call_ops,
491 how: 0, RPC_TASK_SOFTCONN);
492 return PNFS_ATTEMPTED;
493}
494
495/* Perform async writes. */
496static enum pnfs_try_status
497filelayout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
498{
499 struct pnfs_layout_segment *lseg = hdr->lseg;
500 struct nfs4_pnfs_ds *ds;
501 struct rpc_clnt *ds_clnt;
502 loff_t offset = hdr->args.offset;
503 u32 j, idx;
504 struct nfs_fh *fh;
505
506 /* Retrieve the correct rpc_client for the byte range */
507 j = nfs4_fl_calc_j_index(lseg, offset);
508 idx = nfs4_fl_calc_ds_index(lseg, j);
509 ds = nfs4_fl_prepare_ds(lseg, ds_idx: idx);
510 if (!ds)
511 return PNFS_NOT_ATTEMPTED;
512
513 ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
514 if (IS_ERR(ptr: ds_clnt))
515 return PNFS_NOT_ATTEMPTED;
516
517 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d\n",
518 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
519 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count));
520
521 hdr->pgio_done_cb = filelayout_write_done_cb;
522 refcount_inc(r: &ds->ds_clp->cl_count);
523 hdr->ds_clp = ds->ds_clp;
524 hdr->ds_commit_idx = idx;
525 fh = nfs4_fl_select_ds_fh(lseg, j);
526 if (fh)
527 hdr->args.fh = fh;
528 hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
529
530 /* Perform an asynchronous write */
531 nfs_initiate_pgio(clnt: ds_clnt, hdr, cred: hdr->cred,
532 rpc_ops: NFS_PROTO(inode: hdr->inode), call_ops: &filelayout_write_call_ops,
533 how: sync, RPC_TASK_SOFTCONN);
534 return PNFS_ATTEMPTED;
535}
536
537static int
538filelayout_check_deviceid(struct pnfs_layout_hdr *lo,
539 struct nfs4_filelayout_segment *fl,
540 gfp_t gfp_flags)
541{
542 struct nfs4_deviceid_node *d;
543 struct nfs4_file_layout_dsaddr *dsaddr;
544 int status = -EINVAL;
545
546 /* Is the deviceid already set? If so, we're good. */
547 if (fl->dsaddr != NULL)
548 return 0;
549
550 /* find and reference the deviceid */
551 d = nfs4_find_get_deviceid(server: NFS_SERVER(inode: lo->plh_inode), id: &fl->deviceid,
552 cred: lo->plh_lc_cred, gfp_mask: gfp_flags);
553 if (d == NULL)
554 goto out;
555
556 dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
557 /* Found deviceid is unavailable */
558 if (filelayout_test_devid_unavailable(node: &dsaddr->id_node))
559 goto out_put;
560
561 if (fl->first_stripe_index >= dsaddr->stripe_count) {
562 dprintk("%s Bad first_stripe_index %u\n",
563 __func__, fl->first_stripe_index);
564 goto out_put;
565 }
566
567 if ((fl->stripe_type == STRIPE_SPARSE &&
568 fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
569 (fl->stripe_type == STRIPE_DENSE &&
570 fl->num_fh != dsaddr->stripe_count)) {
571 dprintk("%s num_fh %u not valid for given packing\n",
572 __func__, fl->num_fh);
573 goto out_put;
574 }
575 status = 0;
576
577 /*
578 * Atomic compare and xchange to ensure we don't scribble
579 * over a non-NULL pointer.
580 */
581 if (cmpxchg(&fl->dsaddr, NULL, dsaddr) != NULL)
582 goto out_put;
583out:
584 return status;
585out_put:
586 nfs4_fl_put_deviceid(dsaddr);
587 goto out;
588}
589
590/*
591 * filelayout_check_layout()
592 *
593 * Make sure layout segment parameters are sane WRT the device.
594 * At this point no generic layer initialization of the lseg has occurred,
595 * and nothing has been added to the layout_hdr cache.
596 *
597 */
598static int
599filelayout_check_layout(struct pnfs_layout_hdr *lo,
600 struct nfs4_filelayout_segment *fl,
601 struct nfs4_layoutget_res *lgr,
602 gfp_t gfp_flags)
603{
604 int status = -EINVAL;
605
606 dprintk("--> %s\n", __func__);
607
608 /* FIXME: remove this check when layout segment support is added */
609 if (lgr->range.offset != 0 ||
610 lgr->range.length != NFS4_MAX_UINT64) {
611 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
612 __func__);
613 goto out;
614 }
615
616 if (fl->pattern_offset > lgr->range.offset) {
617 dprintk("%s pattern_offset %lld too large\n",
618 __func__, fl->pattern_offset);
619 goto out;
620 }
621
622 if (!fl->stripe_unit) {
623 dprintk("%s Invalid stripe unit (%u)\n",
624 __func__, fl->stripe_unit);
625 goto out;
626 }
627
628 status = 0;
629out:
630 dprintk("--> %s returns %d\n", __func__, status);
631 return status;
632}
633
634static void _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
635{
636 int i;
637
638 if (fl->fh_array) {
639 for (i = 0; i < fl->num_fh; i++) {
640 if (!fl->fh_array[i])
641 break;
642 kfree(objp: fl->fh_array[i]);
643 }
644 kfree(objp: fl->fh_array);
645 }
646 kfree(objp: fl);
647}
648
649static int
650filelayout_decode_layout(struct pnfs_layout_hdr *flo,
651 struct nfs4_filelayout_segment *fl,
652 struct nfs4_layoutget_res *lgr,
653 gfp_t gfp_flags)
654{
655 struct xdr_stream stream;
656 struct xdr_buf buf;
657 struct page *scratch;
658 __be32 *p;
659 uint32_t nfl_util;
660 int i;
661
662 dprintk("%s: set_layout_map Begin\n", __func__);
663
664 scratch = alloc_page(gfp_flags);
665 if (!scratch)
666 return -ENOMEM;
667
668 xdr_init_decode_pages(xdr: &stream, buf: &buf, pages: lgr->layoutp->pages, len: lgr->layoutp->len);
669 xdr_set_scratch_page(xdr: &stream, page: scratch);
670
671 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
672 * num_fh (4) */
673 p = xdr_inline_decode(xdr: &stream, NFS4_DEVICEID4_SIZE + 20);
674 if (unlikely(!p))
675 goto out_err;
676
677 memcpy(&fl->deviceid, p, sizeof(fl->deviceid));
678 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
679 nfs4_print_deviceid(dev_id: &fl->deviceid);
680
681 nfl_util = be32_to_cpup(p: p++);
682 if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
683 fl->commit_through_mds = 1;
684 if (nfl_util & NFL4_UFLG_DENSE)
685 fl->stripe_type = STRIPE_DENSE;
686 else
687 fl->stripe_type = STRIPE_SPARSE;
688 fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
689
690 fl->first_stripe_index = be32_to_cpup(p: p++);
691 p = xdr_decode_hyper(p, valp: &fl->pattern_offset);
692 fl->num_fh = be32_to_cpup(p: p++);
693
694 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
695 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
696 fl->pattern_offset);
697
698 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
699 * Futher checking is done in filelayout_check_layout */
700 if (fl->num_fh >
701 max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
702 goto out_err;
703
704 if (fl->num_fh > 0) {
705 fl->fh_array = kcalloc(n: fl->num_fh, size: sizeof(fl->fh_array[0]),
706 flags: gfp_flags);
707 if (!fl->fh_array)
708 goto out_err;
709 }
710
711 for (i = 0; i < fl->num_fh; i++) {
712 /* Do we want to use a mempool here? */
713 fl->fh_array[i] = kmalloc(size: sizeof(struct nfs_fh), flags: gfp_flags);
714 if (!fl->fh_array[i])
715 goto out_err;
716
717 p = xdr_inline_decode(xdr: &stream, nbytes: 4);
718 if (unlikely(!p))
719 goto out_err;
720 fl->fh_array[i]->size = be32_to_cpup(p: p++);
721 if (fl->fh_array[i]->size > NFS_MAXFHSIZE) {
722 printk(KERN_ERR "NFS: Too big fh %d received %d\n",
723 i, fl->fh_array[i]->size);
724 goto out_err;
725 }
726
727 p = xdr_inline_decode(xdr: &stream, nbytes: fl->fh_array[i]->size);
728 if (unlikely(!p))
729 goto out_err;
730 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
731 dprintk("DEBUG: %s: fh len %d\n", __func__,
732 fl->fh_array[i]->size);
733 }
734
735 __free_page(scratch);
736 return 0;
737
738out_err:
739 __free_page(scratch);
740 return -EIO;
741}
742
743static void
744filelayout_free_lseg(struct pnfs_layout_segment *lseg)
745{
746 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
747
748 dprintk("--> %s\n", __func__);
749 if (fl->dsaddr != NULL)
750 nfs4_fl_put_deviceid(dsaddr: fl->dsaddr);
751 /* This assumes a single RW lseg */
752 if (lseg->pls_range.iomode == IOMODE_RW) {
753 struct nfs4_filelayout *flo;
754 struct inode *inode;
755
756 flo = FILELAYOUT_FROM_HDR(lo: lseg->pls_layout);
757 inode = flo->generic_hdr.plh_inode;
758 spin_lock(lock: &inode->i_lock);
759 pnfs_generic_ds_cinfo_release_lseg(fl_cinfo: &flo->commit_info, lseg);
760 spin_unlock(lock: &inode->i_lock);
761 }
762 _filelayout_free_lseg(fl);
763}
764
765static struct pnfs_layout_segment *
766filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
767 struct nfs4_layoutget_res *lgr,
768 gfp_t gfp_flags)
769{
770 struct nfs4_filelayout_segment *fl;
771 int rc;
772
773 dprintk("--> %s\n", __func__);
774 fl = kzalloc(size: sizeof(*fl), flags: gfp_flags);
775 if (!fl)
776 return NULL;
777
778 rc = filelayout_decode_layout(flo: layoutid, fl, lgr, gfp_flags);
779 if (rc != 0 || filelayout_check_layout(lo: layoutid, fl, lgr, gfp_flags)) {
780 _filelayout_free_lseg(fl);
781 return NULL;
782 }
783 return &fl->generic_hdr;
784}
785
786static bool
787filelayout_lseg_is_striped(const struct nfs4_filelayout_segment *flseg)
788{
789 return flseg->num_fh > 1;
790}
791
792/*
793 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
794 *
795 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
796 * of bytes (maximum @req->wb_bytes) that can be coalesced.
797 */
798static size_t
799filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
800 struct nfs_page *req)
801{
802 unsigned int size;
803 u64 p_stripe, r_stripe;
804 u32 stripe_offset;
805 u64 segment_offset = pgio->pg_lseg->pls_range.offset;
806 u32 stripe_unit = FILELAYOUT_LSEG(lseg: pgio->pg_lseg)->stripe_unit;
807
808 /* calls nfs_generic_pg_test */
809 size = pnfs_generic_pg_test(pgio, prev, req);
810 if (!size)
811 return 0;
812 else if (!filelayout_lseg_is_striped(flseg: FILELAYOUT_LSEG(lseg: pgio->pg_lseg)))
813 return size;
814
815 /* see if req and prev are in the same stripe */
816 if (prev) {
817 p_stripe = (u64)req_offset(req: prev) - segment_offset;
818 r_stripe = (u64)req_offset(req) - segment_offset;
819 do_div(p_stripe, stripe_unit);
820 do_div(r_stripe, stripe_unit);
821
822 if (p_stripe != r_stripe)
823 return 0;
824 }
825
826 /* calculate remaining bytes in the current stripe */
827 div_u64_rem(dividend: (u64)req_offset(req) - segment_offset,
828 divisor: stripe_unit,
829 remainder: &stripe_offset);
830 WARN_ON_ONCE(stripe_offset > stripe_unit);
831 if (stripe_offset >= stripe_unit)
832 return 0;
833 return min(stripe_unit - (unsigned int)stripe_offset, size);
834}
835
836static struct pnfs_layout_segment *
837fl_pnfs_update_layout(struct inode *ino,
838 struct nfs_open_context *ctx,
839 loff_t pos,
840 u64 count,
841 enum pnfs_iomode iomode,
842 bool strict_iomode,
843 gfp_t gfp_flags)
844{
845 struct pnfs_layout_segment *lseg = NULL;
846 struct pnfs_layout_hdr *lo;
847 struct nfs4_filelayout_segment *fl;
848 int status;
849
850 lseg = pnfs_update_layout(ino, ctx, pos, count, iomode, strict_iomode,
851 gfp_flags);
852 if (IS_ERR(ptr: lseg)) {
853 /* Fall back to MDS on recoverable errors */
854 if (!nfs_error_is_fatal_on_server(err: PTR_ERR(ptr: lseg)))
855 lseg = NULL;
856 goto out;
857 } else if (!lseg)
858 goto out;
859
860 lo = NFS_I(inode: ino)->layout;
861 fl = FILELAYOUT_LSEG(lseg);
862
863 status = filelayout_check_deviceid(lo, fl, gfp_flags);
864 if (status) {
865 pnfs_error_mark_layout_for_return(inode: ino, lseg);
866 pnfs_set_lo_fail(lseg);
867 pnfs_put_lseg(lseg);
868 lseg = NULL;
869 }
870out:
871 return lseg;
872}
873
874static void
875filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
876 struct nfs_page *req)
877{
878 pnfs_generic_pg_check_layout(pgio);
879 if (!pgio->pg_lseg) {
880 pgio->pg_lseg = fl_pnfs_update_layout(ino: pgio->pg_inode,
881 ctx: nfs_req_openctx(req),
882 pos: 0,
883 NFS4_MAX_UINT64,
884 iomode: IOMODE_READ,
885 strict_iomode: false,
886 GFP_KERNEL);
887 if (IS_ERR(ptr: pgio->pg_lseg)) {
888 pgio->pg_error = PTR_ERR(ptr: pgio->pg_lseg);
889 pgio->pg_lseg = NULL;
890 return;
891 }
892 }
893 /* If no lseg, fall back to read through mds */
894 if (pgio->pg_lseg == NULL)
895 nfs_pageio_reset_read_mds(pgio);
896}
897
898static void
899filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
900 struct nfs_page *req)
901{
902 pnfs_generic_pg_check_layout(pgio);
903 if (!pgio->pg_lseg) {
904 pgio->pg_lseg = fl_pnfs_update_layout(ino: pgio->pg_inode,
905 ctx: nfs_req_openctx(req),
906 pos: 0,
907 NFS4_MAX_UINT64,
908 iomode: IOMODE_RW,
909 strict_iomode: false,
910 GFP_NOFS);
911 if (IS_ERR(ptr: pgio->pg_lseg)) {
912 pgio->pg_error = PTR_ERR(ptr: pgio->pg_lseg);
913 pgio->pg_lseg = NULL;
914 return;
915 }
916 }
917
918 /* If no lseg, fall back to write through mds */
919 if (pgio->pg_lseg == NULL)
920 nfs_pageio_reset_write_mds(pgio);
921}
922
923static const struct nfs_pageio_ops filelayout_pg_read_ops = {
924 .pg_init = filelayout_pg_init_read,
925 .pg_test = filelayout_pg_test,
926 .pg_doio = pnfs_generic_pg_readpages,
927 .pg_cleanup = pnfs_generic_pg_cleanup,
928};
929
930static const struct nfs_pageio_ops filelayout_pg_write_ops = {
931 .pg_init = filelayout_pg_init_write,
932 .pg_test = filelayout_pg_test,
933 .pg_doio = pnfs_generic_pg_writepages,
934 .pg_cleanup = pnfs_generic_pg_cleanup,
935};
936
937static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
938{
939 if (fl->stripe_type == STRIPE_SPARSE)
940 return nfs4_fl_calc_ds_index(lseg: &fl->generic_hdr, j);
941 else
942 return j;
943}
944
945static void
946filelayout_mark_request_commit(struct nfs_page *req,
947 struct pnfs_layout_segment *lseg,
948 struct nfs_commit_info *cinfo,
949 u32 ds_commit_idx)
950
951{
952 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
953 u32 i, j;
954
955 if (fl->commit_through_mds) {
956 nfs_request_add_commit_list(req, cinfo);
957 } else {
958 /* Note that we are calling nfs4_fl_calc_j_index on each page
959 * that ends up being committed to a data server. An attractive
960 * alternative is to add a field to nfs_write_data and nfs_page
961 * to store the value calculated in filelayout_write_pagelist
962 * and just use that here.
963 */
964 j = nfs4_fl_calc_j_index(lseg, offset: req_offset(req));
965 i = select_bucket_index(fl, j);
966 pnfs_layout_mark_request_commit(req, lseg, cinfo, ds_commit_idx: i);
967 }
968}
969
970static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
971{
972 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
973
974 if (flseg->stripe_type == STRIPE_SPARSE)
975 return i;
976 else
977 return nfs4_fl_calc_ds_index(lseg, j: i);
978}
979
980static struct nfs_fh *
981select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
982{
983 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
984
985 if (flseg->stripe_type == STRIPE_SPARSE) {
986 if (flseg->num_fh == 1)
987 i = 0;
988 else if (flseg->num_fh == 0)
989 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
990 return NULL;
991 }
992 return flseg->fh_array[i];
993}
994
995static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
996{
997 struct pnfs_layout_segment *lseg = data->lseg;
998 struct nfs4_pnfs_ds *ds;
999 struct rpc_clnt *ds_clnt;
1000 u32 idx;
1001 struct nfs_fh *fh;
1002
1003 idx = calc_ds_index_from_commit(lseg, i: data->ds_commit_index);
1004 ds = nfs4_fl_prepare_ds(lseg, ds_idx: idx);
1005 if (!ds)
1006 goto out_err;
1007
1008 ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, data->inode);
1009 if (IS_ERR(ptr: ds_clnt))
1010 goto out_err;
1011
1012 dprintk("%s ino %lu, how %d cl_count %d\n", __func__,
1013 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count));
1014 data->commit_done_cb = filelayout_commit_done_cb;
1015 refcount_inc(r: &ds->ds_clp->cl_count);
1016 data->ds_clp = ds->ds_clp;
1017 fh = select_ds_fh_from_commit(lseg, i: data->ds_commit_index);
1018 if (fh)
1019 data->args.fh = fh;
1020 return nfs_initiate_commit(clnt: ds_clnt, data, nfs_ops: NFS_PROTO(inode: data->inode),
1021 call_ops: &filelayout_commit_call_ops, how,
1022 RPC_TASK_SOFTCONN);
1023out_err:
1024 pnfs_generic_prepare_to_resend_writes(data);
1025 pnfs_generic_commit_release(calldata: data);
1026 return -EAGAIN;
1027}
1028
1029static int
1030filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1031 int how, struct nfs_commit_info *cinfo)
1032{
1033 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1034 initiate_commit: filelayout_initiate_commit);
1035}
1036
1037static struct nfs4_deviceid_node *
1038filelayout_alloc_deviceid_node(struct nfs_server *server,
1039 struct pnfs_device *pdev, gfp_t gfp_flags)
1040{
1041 struct nfs4_file_layout_dsaddr *dsaddr;
1042
1043 dsaddr = nfs4_fl_alloc_deviceid_node(server, pdev, gfp_flags);
1044 if (!dsaddr)
1045 return NULL;
1046 return &dsaddr->id_node;
1047}
1048
1049static void
1050filelayout_free_deviceid_node(struct nfs4_deviceid_node *d)
1051{
1052 nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1053}
1054
1055static struct pnfs_layout_hdr *
1056filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
1057{
1058 struct nfs4_filelayout *flo;
1059
1060 flo = kzalloc(size: sizeof(*flo), flags: gfp_flags);
1061 if (flo == NULL)
1062 return NULL;
1063 pnfs_init_ds_commit_info(fl_cinfo: &flo->commit_info);
1064 flo->commit_info.ops = &filelayout_commit_ops;
1065 return &flo->generic_hdr;
1066}
1067
1068static void
1069filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
1070{
1071 kfree_rcu(FILELAYOUT_FROM_HDR(lo), generic_hdr.plh_rcu);
1072}
1073
1074static struct pnfs_ds_commit_info *
1075filelayout_get_ds_info(struct inode *inode)
1076{
1077 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1078
1079 if (layout == NULL)
1080 return NULL;
1081 else
1082 return &FILELAYOUT_FROM_HDR(lo: layout)->commit_info;
1083}
1084
1085static void
1086filelayout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1087 struct pnfs_layout_segment *lseg)
1088{
1089 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
1090 struct inode *inode = lseg->pls_layout->plh_inode;
1091 struct pnfs_commit_array *array, *new;
1092 unsigned int size = (fl->stripe_type == STRIPE_SPARSE) ?
1093 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
1094
1095 new = pnfs_alloc_commit_array(n: size, gfp_flags: nfs_io_gfp_mask());
1096 if (new) {
1097 spin_lock(lock: &inode->i_lock);
1098 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1099 spin_unlock(lock: &inode->i_lock);
1100 if (array != new)
1101 pnfs_free_commit_array(p: new);
1102 }
1103}
1104
1105static void
1106filelayout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1107 struct inode *inode)
1108{
1109 spin_lock(lock: &inode->i_lock);
1110 pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1111 spin_unlock(lock: &inode->i_lock);
1112}
1113
1114static const struct pnfs_commit_ops filelayout_commit_ops = {
1115 .setup_ds_info = filelayout_setup_ds_info,
1116 .release_ds_info = filelayout_release_ds_info,
1117 .mark_request_commit = filelayout_mark_request_commit,
1118 .clear_request_commit = pnfs_generic_clear_request_commit,
1119 .scan_commit_lists = pnfs_generic_scan_commit_lists,
1120 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
1121 .search_commit_reqs = pnfs_generic_search_commit_reqs,
1122 .commit_pagelist = filelayout_commit_pagelist,
1123};
1124
1125static struct pnfs_layoutdriver_type filelayout_type = {
1126 .id = LAYOUT_NFSV4_1_FILES,
1127 .name = "LAYOUT_NFSV4_1_FILES",
1128 .owner = THIS_MODULE,
1129 .flags = PNFS_LAYOUTGET_ON_OPEN,
1130 .max_layoutget_response = 4096, /* 1 page or so... */
1131 .alloc_layout_hdr = filelayout_alloc_layout_hdr,
1132 .free_layout_hdr = filelayout_free_layout_hdr,
1133 .alloc_lseg = filelayout_alloc_lseg,
1134 .free_lseg = filelayout_free_lseg,
1135 .pg_read_ops = &filelayout_pg_read_ops,
1136 .pg_write_ops = &filelayout_pg_write_ops,
1137 .get_ds_info = &filelayout_get_ds_info,
1138 .read_pagelist = filelayout_read_pagelist,
1139 .write_pagelist = filelayout_write_pagelist,
1140 .alloc_deviceid_node = filelayout_alloc_deviceid_node,
1141 .free_deviceid_node = filelayout_free_deviceid_node,
1142 .sync = pnfs_nfs_generic_sync,
1143};
1144
1145static int __init nfs4filelayout_init(void)
1146{
1147 printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1148 __func__);
1149 return pnfs_register_layoutdriver(&filelayout_type);
1150}
1151
1152static void __exit nfs4filelayout_exit(void)
1153{
1154 printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1155 __func__);
1156 pnfs_unregister_layoutdriver(&filelayout_type);
1157}
1158
1159MODULE_ALIAS("nfs-layouttype4-1");
1160
1161module_init(nfs4filelayout_init);
1162module_exit(nfs4filelayout_exit);
1163

source code of linux/fs/nfs/filelayout/filelayout.c