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.
7 * The Regents of the University of Michigan
10 * Dean Hildebrand <dhildebz@umich.edu>
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.
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
32 #include <linux/nfs_fs.h>
33 #include <linux/nfs_page.h>
34 #include <linux/module.h>
35 #include <linux/backing-dev.h>
37 #include <linux/sunrpc/metrics.h>
39 #include "../nfs4session.h"
40 #include "../internal.h"
41 #include "../delegation.h"
42 #include "filelayout.h"
43 #include "../nfs4trace.h"
45 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
47 MODULE_LICENSE("GPL");
48 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
49 MODULE_DESCRIPTION("The NFSv4 file layout driver");
51 #define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
54 filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
57 u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
61 offset -= flseg->pattern_offset;
62 stripe_no = div_u64(offset, stripe_width);
63 div_u64_rem(offset, flseg->stripe_unit, &rem);
65 return stripe_no * flseg->stripe_unit + rem;
68 /* This function is used by the layout driver to calculate the
69 * offset of the file on the dserver based on whether the
70 * layout type is STRIPE_DENSE or STRIPE_SPARSE
73 filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
75 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
77 switch (flseg->stripe_type) {
82 return filelayout_get_dense_offset(flseg, offset);
88 static void filelayout_reset_write(struct nfs_pgio_header *hdr)
90 struct rpc_task *task = &hdr->task;
92 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
93 dprintk("%s Reset task %5u for i/o through MDS "
94 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
96 hdr->inode->i_sb->s_id,
97 (unsigned long long)NFS_FILEID(hdr->inode),
99 (unsigned long long)hdr->args.offset);
101 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
105 static void filelayout_reset_read(struct nfs_pgio_header *hdr)
107 struct rpc_task *task = &hdr->task;
109 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
110 dprintk("%s Reset task %5u for i/o through MDS "
111 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
113 hdr->inode->i_sb->s_id,
114 (unsigned long long)NFS_FILEID(hdr->inode),
116 (unsigned long long)hdr->args.offset);
118 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
122 static int filelayout_async_handle_error(struct rpc_task *task,
123 struct nfs4_state *state,
124 struct nfs_client *clp,
125 struct pnfs_layout_segment *lseg)
127 struct pnfs_layout_hdr *lo = lseg->pls_layout;
128 struct inode *inode = lo->plh_inode;
129 struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
130 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
132 if (task->tk_status >= 0)
135 switch (task->tk_status) {
136 /* DS session errors */
137 case -NFS4ERR_BADSESSION:
138 case -NFS4ERR_BADSLOT:
139 case -NFS4ERR_BAD_HIGH_SLOT:
140 case -NFS4ERR_DEADSESSION:
141 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
142 case -NFS4ERR_SEQ_FALSE_RETRY:
143 case -NFS4ERR_SEQ_MISORDERED:
144 dprintk("%s ERROR %d, Reset session. Exchangeid "
145 "flags 0x%x\n", __func__, task->tk_status,
146 clp->cl_exchange_flags);
147 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
151 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
153 case -NFS4ERR_RETRY_UNCACHED_REP:
155 /* Invalidate Layout errors */
156 case -NFS4ERR_ACCESS:
157 case -NFS4ERR_PNFS_NO_LAYOUT:
158 case -ESTALE: /* mapped NFS4ERR_STALE */
159 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
160 case -EISDIR: /* mapped NFS4ERR_ISDIR */
161 case -NFS4ERR_FHEXPIRED:
162 case -NFS4ERR_WRONG_TYPE:
163 dprintk("%s Invalid layout error %d\n", __func__,
166 * Destroy layout so new i/o will get a new layout.
167 * Layout will not be destroyed until all current lseg
168 * references are put. Mark layout as invalid to resend failed
169 * i/o and all i/o waiting on the slot table to the MDS until
170 * layout is destroyed and a new valid layout is obtained.
172 pnfs_destroy_layout(NFS_I(inode));
173 rpc_wake_up(&tbl->slot_tbl_waitq);
175 /* RPC connection errors */
183 dprintk("%s DS connection error %d\n", __func__,
185 nfs4_mark_deviceid_unavailable(devid);
186 pnfs_error_mark_layout_for_return(inode, lseg);
187 pnfs_set_lo_fail(lseg);
188 rpc_wake_up(&tbl->slot_tbl_waitq);
192 dprintk("%s Retry through MDS. Error %d\n", __func__,
194 return -NFS4ERR_RESET_TO_MDS;
200 /* NFS_PROTO call done callback routines */
202 static int filelayout_read_done_cb(struct rpc_task *task,
203 struct nfs_pgio_header *hdr)
207 trace_nfs4_pnfs_read(hdr, task->tk_status);
208 err = filelayout_async_handle_error(task, hdr->args.context->state,
209 hdr->ds_clp, hdr->lseg);
212 case -NFS4ERR_RESET_TO_MDS:
213 filelayout_reset_read(hdr);
214 return task->tk_status;
216 rpc_restart_call_prepare(task);
224 * We reference the rpc_cred of the first WRITE that triggers the need for
225 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
226 * rfc5661 is not clear about which credential should be used.
229 filelayout_set_layoutcommit(struct nfs_pgio_header *hdr)
233 if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds ||
234 hdr->res.verf->committed == NFS_FILE_SYNC)
236 if (hdr->res.verf->committed == NFS_DATA_SYNC)
237 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
239 /* Note: if the write is unstable, don't set end_offs until commit */
240 pnfs_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
241 dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
242 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
246 filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node)
248 return filelayout_test_devid_invalid(node) ||
249 nfs4_test_deviceid_unavailable(node);
253 filelayout_reset_to_mds(struct pnfs_layout_segment *lseg)
255 struct nfs4_deviceid_node *node = FILELAYOUT_DEVID_NODE(lseg);
257 return filelayout_test_devid_unavailable(node);
261 * Call ops for the async read/write cases
262 * In the case of dense layouts, the offset needs to be reset to its
265 static void filelayout_read_prepare(struct rpc_task *task, void *data)
267 struct nfs_pgio_header *hdr = data;
269 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
270 rpc_exit(task, -EIO);
273 if (filelayout_reset_to_mds(hdr->lseg)) {
274 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
275 filelayout_reset_read(hdr);
279 hdr->pgio_done_cb = filelayout_read_done_cb;
281 if (nfs4_setup_sequence(hdr->ds_clp,
286 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
287 hdr->args.lock_context, FMODE_READ) == -EIO)
288 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
291 static void filelayout_read_call_done(struct rpc_task *task, void *data)
293 struct nfs_pgio_header *hdr = data;
295 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
297 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
298 task->tk_status == 0) {
299 nfs41_sequence_done(task, &hdr->res.seq_res);
303 /* Note this may cause RPC to be resent */
304 hdr->mds_ops->rpc_call_done(task, data);
307 static void filelayout_read_count_stats(struct rpc_task *task, void *data)
309 struct nfs_pgio_header *hdr = data;
311 rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
314 static int filelayout_write_done_cb(struct rpc_task *task,
315 struct nfs_pgio_header *hdr)
319 trace_nfs4_pnfs_write(hdr, task->tk_status);
320 err = filelayout_async_handle_error(task, hdr->args.context->state,
321 hdr->ds_clp, hdr->lseg);
324 case -NFS4ERR_RESET_TO_MDS:
325 filelayout_reset_write(hdr);
326 return task->tk_status;
328 rpc_restart_call_prepare(task);
332 filelayout_set_layoutcommit(hdr);
334 /* zero out the fattr */
335 hdr->fattr.valid = 0;
336 if (task->tk_status >= 0)
337 nfs_writeback_update_inode(hdr);
342 static int filelayout_commit_done_cb(struct rpc_task *task,
343 struct nfs_commit_data *data)
347 trace_nfs4_pnfs_commit_ds(data, task->tk_status);
348 err = filelayout_async_handle_error(task, NULL, data->ds_clp,
352 case -NFS4ERR_RESET_TO_MDS:
353 pnfs_generic_prepare_to_resend_writes(data);
356 rpc_restart_call_prepare(task);
360 pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
365 static void filelayout_write_prepare(struct rpc_task *task, void *data)
367 struct nfs_pgio_header *hdr = data;
369 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
370 rpc_exit(task, -EIO);
373 if (filelayout_reset_to_mds(hdr->lseg)) {
374 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
375 filelayout_reset_write(hdr);
379 if (nfs4_setup_sequence(hdr->ds_clp,
384 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
385 hdr->args.lock_context, FMODE_WRITE) == -EIO)
386 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
389 static void filelayout_write_call_done(struct rpc_task *task, void *data)
391 struct nfs_pgio_header *hdr = data;
393 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
394 task->tk_status == 0) {
395 nfs41_sequence_done(task, &hdr->res.seq_res);
399 /* Note this may cause RPC to be resent */
400 hdr->mds_ops->rpc_call_done(task, data);
403 static void filelayout_write_count_stats(struct rpc_task *task, void *data)
405 struct nfs_pgio_header *hdr = data;
407 rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
410 static void filelayout_commit_prepare(struct rpc_task *task, void *data)
412 struct nfs_commit_data *wdata = data;
414 nfs4_setup_sequence(wdata->ds_clp,
415 &wdata->args.seq_args,
420 static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
422 struct nfs_commit_data *cdata = data;
424 rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics);
427 static const struct rpc_call_ops filelayout_read_call_ops = {
428 .rpc_call_prepare = filelayout_read_prepare,
429 .rpc_call_done = filelayout_read_call_done,
430 .rpc_count_stats = filelayout_read_count_stats,
431 .rpc_release = pnfs_generic_rw_release,
434 static const struct rpc_call_ops filelayout_write_call_ops = {
435 .rpc_call_prepare = filelayout_write_prepare,
436 .rpc_call_done = filelayout_write_call_done,
437 .rpc_count_stats = filelayout_write_count_stats,
438 .rpc_release = pnfs_generic_rw_release,
441 static const struct rpc_call_ops filelayout_commit_call_ops = {
442 .rpc_call_prepare = filelayout_commit_prepare,
443 .rpc_call_done = pnfs_generic_write_commit_done,
444 .rpc_count_stats = filelayout_commit_count_stats,
445 .rpc_release = pnfs_generic_commit_release,
448 static enum pnfs_try_status
449 filelayout_read_pagelist(struct nfs_pgio_header *hdr)
451 struct pnfs_layout_segment *lseg = hdr->lseg;
452 struct nfs4_pnfs_ds *ds;
453 struct rpc_clnt *ds_clnt;
454 loff_t offset = hdr->args.offset;
458 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
459 __func__, hdr->inode->i_ino,
460 hdr->args.pgbase, (size_t)hdr->args.count, offset);
462 /* Retrieve the correct rpc_client for the byte range */
463 j = nfs4_fl_calc_j_index(lseg, offset);
464 idx = nfs4_fl_calc_ds_index(lseg, j);
465 ds = nfs4_fl_prepare_ds(lseg, idx);
467 return PNFS_NOT_ATTEMPTED;
469 ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
471 return PNFS_NOT_ATTEMPTED;
473 dprintk("%s USE DS: %s cl_count %d\n", __func__,
474 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count));
476 /* No multipath support. Use first DS */
477 refcount_inc(&ds->ds_clp->cl_count);
478 hdr->ds_clp = ds->ds_clp;
479 hdr->ds_commit_idx = idx;
480 fh = nfs4_fl_select_ds_fh(lseg, j);
484 hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
485 hdr->mds_offset = offset;
487 /* Perform an asynchronous read to ds */
488 nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
489 NFS_PROTO(hdr->inode), &filelayout_read_call_ops,
490 0, RPC_TASK_SOFTCONN);
491 return PNFS_ATTEMPTED;
494 /* Perform async writes. */
495 static enum pnfs_try_status
496 filelayout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
498 struct pnfs_layout_segment *lseg = hdr->lseg;
499 struct nfs4_pnfs_ds *ds;
500 struct rpc_clnt *ds_clnt;
501 loff_t offset = hdr->args.offset;
505 /* Retrieve the correct rpc_client for the byte range */
506 j = nfs4_fl_calc_j_index(lseg, offset);
507 idx = nfs4_fl_calc_ds_index(lseg, j);
508 ds = nfs4_fl_prepare_ds(lseg, idx);
510 return PNFS_NOT_ATTEMPTED;
512 ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
514 return PNFS_NOT_ATTEMPTED;
516 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d\n",
517 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
518 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count));
520 hdr->pgio_done_cb = filelayout_write_done_cb;
521 refcount_inc(&ds->ds_clp->cl_count);
522 hdr->ds_clp = ds->ds_clp;
523 hdr->ds_commit_idx = idx;
524 fh = nfs4_fl_select_ds_fh(lseg, j);
527 hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
529 /* Perform an asynchronous write */
530 nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
531 NFS_PROTO(hdr->inode), &filelayout_write_call_ops,
532 sync, RPC_TASK_SOFTCONN);
533 return PNFS_ATTEMPTED;
537 filelayout_check_deviceid(struct pnfs_layout_hdr *lo,
538 struct nfs4_filelayout_segment *fl,
541 struct nfs4_deviceid_node *d;
542 struct nfs4_file_layout_dsaddr *dsaddr;
543 int status = -EINVAL;
545 /* Is the deviceid already set? If so, we're good. */
546 if (fl->dsaddr != NULL)
549 /* find and reference the deviceid */
550 d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), &fl->deviceid,
551 lo->plh_lc_cred, gfp_flags);
555 dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
556 /* Found deviceid is unavailable */
557 if (filelayout_test_devid_unavailable(&dsaddr->id_node))
560 if (fl->first_stripe_index >= dsaddr->stripe_count) {
561 dprintk("%s Bad first_stripe_index %u\n",
562 __func__, fl->first_stripe_index);
566 if ((fl->stripe_type == STRIPE_SPARSE &&
567 fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
568 (fl->stripe_type == STRIPE_DENSE &&
569 fl->num_fh != dsaddr->stripe_count)) {
570 dprintk("%s num_fh %u not valid for given packing\n",
571 __func__, fl->num_fh);
577 * Atomic compare and xchange to ensure we don't scribble
578 * over a non-NULL pointer.
580 if (cmpxchg(&fl->dsaddr, NULL, dsaddr) != NULL)
585 nfs4_fl_put_deviceid(dsaddr);
590 * filelayout_check_layout()
592 * Make sure layout segment parameters are sane WRT the device.
593 * At this point no generic layer initialization of the lseg has occurred,
594 * and nothing has been added to the layout_hdr cache.
598 filelayout_check_layout(struct pnfs_layout_hdr *lo,
599 struct nfs4_filelayout_segment *fl,
600 struct nfs4_layoutget_res *lgr,
603 int status = -EINVAL;
605 dprintk("--> %s\n", __func__);
607 /* FIXME: remove this check when layout segment support is added */
608 if (lgr->range.offset != 0 ||
609 lgr->range.length != NFS4_MAX_UINT64) {
610 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
615 if (fl->pattern_offset > lgr->range.offset) {
616 dprintk("%s pattern_offset %lld too large\n",
617 __func__, fl->pattern_offset);
621 if (!fl->stripe_unit) {
622 dprintk("%s Invalid stripe unit (%u)\n",
623 __func__, fl->stripe_unit);
629 dprintk("--> %s returns %d\n", __func__, status);
633 static void _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
638 for (i = 0; i < fl->num_fh; i++) {
639 if (!fl->fh_array[i])
641 kfree(fl->fh_array[i]);
649 filelayout_decode_layout(struct pnfs_layout_hdr *flo,
650 struct nfs4_filelayout_segment *fl,
651 struct nfs4_layoutget_res *lgr,
654 struct xdr_stream stream;
656 struct page *scratch;
661 dprintk("%s: set_layout_map Begin\n", __func__);
663 scratch = alloc_page(gfp_flags);
667 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
668 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
670 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
672 p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
676 memcpy(&fl->deviceid, p, sizeof(fl->deviceid));
677 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
678 nfs4_print_deviceid(&fl->deviceid);
680 nfl_util = be32_to_cpup(p++);
681 if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
682 fl->commit_through_mds = 1;
683 if (nfl_util & NFL4_UFLG_DENSE)
684 fl->stripe_type = STRIPE_DENSE;
686 fl->stripe_type = STRIPE_SPARSE;
687 fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
689 fl->first_stripe_index = be32_to_cpup(p++);
690 p = xdr_decode_hyper(p, &fl->pattern_offset);
691 fl->num_fh = be32_to_cpup(p++);
693 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
694 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
697 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
698 * Futher checking is done in filelayout_check_layout */
700 max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
703 if (fl->num_fh > 0) {
704 fl->fh_array = kcalloc(fl->num_fh, sizeof(fl->fh_array[0]),
710 for (i = 0; i < fl->num_fh; i++) {
711 /* Do we want to use a mempool here? */
712 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
713 if (!fl->fh_array[i])
716 p = xdr_inline_decode(&stream, 4);
719 fl->fh_array[i]->size = be32_to_cpup(p++);
720 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
721 printk(KERN_ERR "NFS: Too big fh %d received %d\n",
722 i, fl->fh_array[i]->size);
726 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
729 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
730 dprintk("DEBUG: %s: fh len %d\n", __func__,
731 fl->fh_array[i]->size);
734 __free_page(scratch);
738 __free_page(scratch);
743 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
745 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
747 dprintk("--> %s\n", __func__);
748 if (fl->dsaddr != NULL)
749 nfs4_fl_put_deviceid(fl->dsaddr);
750 /* This assumes a single RW lseg */
751 if (lseg->pls_range.iomode == IOMODE_RW) {
752 struct nfs4_filelayout *flo;
755 flo = FILELAYOUT_FROM_HDR(lseg->pls_layout);
756 inode = flo->generic_hdr.plh_inode;
757 spin_lock(&inode->i_lock);
758 flo->commit_info.nbuckets = 0;
759 kfree(flo->commit_info.buckets);
760 flo->commit_info.buckets = NULL;
761 pnfs_generic_ds_cinfo_release_lseg(&flo->commit_info, lseg);
762 spin_unlock(&inode->i_lock);
764 _filelayout_free_lseg(fl);
768 filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg,
769 struct nfs_commit_info *cinfo,
772 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
773 struct pnfs_commit_bucket *buckets;
776 if (fl->commit_through_mds)
779 size = (fl->stripe_type == STRIPE_SPARSE) ?
780 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
782 if (cinfo->ds->nbuckets >= size) {
783 /* This assumes there is only one IOMODE_RW lseg. What
784 * we really want to do is have a layout_hdr level
785 * dictionary of <multipath_list4, fh> keys, each
786 * associated with a struct list_head, populated by calls
787 * to filelayout_write_pagelist().
792 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
796 for (i = 0; i < size; i++) {
797 INIT_LIST_HEAD(&buckets[i].written);
798 INIT_LIST_HEAD(&buckets[i].committing);
799 /* mark direct verifier as unset */
800 buckets[i].direct_verf.committed = NFS_INVALID_STABLE_HOW;
803 spin_lock(&cinfo->inode->i_lock);
804 if (cinfo->ds->nbuckets >= size)
806 for (i = 0; i < cinfo->ds->nbuckets; i++) {
807 list_splice(&cinfo->ds->buckets[i].written,
808 &buckets[i].written);
809 list_splice(&cinfo->ds->buckets[i].committing,
810 &buckets[i].committing);
811 buckets[i].direct_verf.committed =
812 cinfo->ds->buckets[i].direct_verf.committed;
813 buckets[i].wlseg = cinfo->ds->buckets[i].wlseg;
814 buckets[i].clseg = cinfo->ds->buckets[i].clseg;
816 swap(cinfo->ds->buckets, buckets);
817 cinfo->ds->nbuckets = size;
819 spin_unlock(&cinfo->inode->i_lock);
824 static struct pnfs_layout_segment *
825 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
826 struct nfs4_layoutget_res *lgr,
829 struct nfs4_filelayout_segment *fl;
832 dprintk("--> %s\n", __func__);
833 fl = kzalloc(sizeof(*fl), gfp_flags);
837 rc = filelayout_decode_layout(layoutid, fl, lgr, gfp_flags);
838 if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, gfp_flags)) {
839 _filelayout_free_lseg(fl);
842 return &fl->generic_hdr;
846 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
848 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
849 * of bytes (maximum @req->wb_bytes) that can be coalesced.
852 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
853 struct nfs_page *req)
856 u64 p_stripe, r_stripe;
858 u64 segment_offset = pgio->pg_lseg->pls_range.offset;
859 u32 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
861 /* calls nfs_generic_pg_test */
862 size = pnfs_generic_pg_test(pgio, prev, req);
866 /* see if req and prev are in the same stripe */
868 p_stripe = (u64)req_offset(prev) - segment_offset;
869 r_stripe = (u64)req_offset(req) - segment_offset;
870 do_div(p_stripe, stripe_unit);
871 do_div(r_stripe, stripe_unit);
873 if (p_stripe != r_stripe)
877 /* calculate remaining bytes in the current stripe */
878 div_u64_rem((u64)req_offset(req) - segment_offset,
881 WARN_ON_ONCE(stripe_offset > stripe_unit);
882 if (stripe_offset >= stripe_unit)
884 return min(stripe_unit - (unsigned int)stripe_offset, size);
887 static struct pnfs_layout_segment *
888 fl_pnfs_update_layout(struct inode *ino,
889 struct nfs_open_context *ctx,
892 enum pnfs_iomode iomode,
896 struct pnfs_layout_segment *lseg = NULL;
897 struct pnfs_layout_hdr *lo;
898 struct nfs4_filelayout_segment *fl;
901 lseg = pnfs_update_layout(ino, ctx, pos, count, iomode, strict_iomode,
903 if (IS_ERR_OR_NULL(lseg))
906 lo = NFS_I(ino)->layout;
907 fl = FILELAYOUT_LSEG(lseg);
909 status = filelayout_check_deviceid(lo, fl, gfp_flags);
919 filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
920 struct nfs_page *req)
922 pnfs_generic_pg_check_layout(pgio);
923 if (!pgio->pg_lseg) {
924 pgio->pg_lseg = fl_pnfs_update_layout(pgio->pg_inode,
925 nfs_req_openctx(req),
931 if (IS_ERR(pgio->pg_lseg)) {
932 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
933 pgio->pg_lseg = NULL;
937 /* If no lseg, fall back to read through mds */
938 if (pgio->pg_lseg == NULL)
939 nfs_pageio_reset_read_mds(pgio);
943 filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
944 struct nfs_page *req)
946 struct nfs_commit_info cinfo;
949 pnfs_generic_pg_check_layout(pgio);
950 if (!pgio->pg_lseg) {
951 pgio->pg_lseg = fl_pnfs_update_layout(pgio->pg_inode,
952 nfs_req_openctx(req),
958 if (IS_ERR(pgio->pg_lseg)) {
959 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
960 pgio->pg_lseg = NULL;
965 /* If no lseg, fall back to write through mds */
966 if (pgio->pg_lseg == NULL)
968 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
969 status = filelayout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
971 pnfs_put_lseg(pgio->pg_lseg);
972 pgio->pg_lseg = NULL;
977 nfs_pageio_reset_write_mds(pgio);
980 static const struct nfs_pageio_ops filelayout_pg_read_ops = {
981 .pg_init = filelayout_pg_init_read,
982 .pg_test = filelayout_pg_test,
983 .pg_doio = pnfs_generic_pg_readpages,
984 .pg_cleanup = pnfs_generic_pg_cleanup,
987 static const struct nfs_pageio_ops filelayout_pg_write_ops = {
988 .pg_init = filelayout_pg_init_write,
989 .pg_test = filelayout_pg_test,
990 .pg_doio = pnfs_generic_pg_writepages,
991 .pg_cleanup = pnfs_generic_pg_cleanup,
994 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
996 if (fl->stripe_type == STRIPE_SPARSE)
997 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
1003 filelayout_mark_request_commit(struct nfs_page *req,
1004 struct pnfs_layout_segment *lseg,
1005 struct nfs_commit_info *cinfo,
1009 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
1012 if (fl->commit_through_mds) {
1013 nfs_request_add_commit_list(req, cinfo);
1015 /* Note that we are calling nfs4_fl_calc_j_index on each page
1016 * that ends up being committed to a data server. An attractive
1017 * alternative is to add a field to nfs_write_data and nfs_page
1018 * to store the value calculated in filelayout_write_pagelist
1019 * and just use that here.
1021 j = nfs4_fl_calc_j_index(lseg, req_offset(req));
1022 i = select_bucket_index(fl, j);
1023 pnfs_layout_mark_request_commit(req, lseg, cinfo, i);
1027 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1029 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1031 if (flseg->stripe_type == STRIPE_SPARSE)
1034 return nfs4_fl_calc_ds_index(lseg, i);
1037 static struct nfs_fh *
1038 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1040 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1042 if (flseg->stripe_type == STRIPE_SPARSE) {
1043 if (flseg->num_fh == 1)
1045 else if (flseg->num_fh == 0)
1046 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
1049 return flseg->fh_array[i];
1052 static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
1054 struct pnfs_layout_segment *lseg = data->lseg;
1055 struct nfs4_pnfs_ds *ds;
1056 struct rpc_clnt *ds_clnt;
1060 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1061 ds = nfs4_fl_prepare_ds(lseg, idx);
1065 ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, data->inode);
1066 if (IS_ERR(ds_clnt))
1069 dprintk("%s ino %lu, how %d cl_count %d\n", __func__,
1070 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count));
1071 data->commit_done_cb = filelayout_commit_done_cb;
1072 refcount_inc(&ds->ds_clp->cl_count);
1073 data->ds_clp = ds->ds_clp;
1074 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1077 return nfs_initiate_commit(ds_clnt, data, NFS_PROTO(data->inode),
1078 &filelayout_commit_call_ops, how,
1081 pnfs_generic_prepare_to_resend_writes(data);
1082 pnfs_generic_commit_release(data);
1086 /* filelayout_search_commit_reqs - Search lists in @cinfo for the head reqest
1088 * @cinfo - commit info for current inode
1089 * @page - page to search for matching head request
1091 * Returns a the head request if one is found, otherwise returns NULL.
1093 static struct nfs_page *
1094 filelayout_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page)
1096 struct nfs_page *freq, *t;
1097 struct pnfs_commit_bucket *b;
1100 /* Linearly search the commit lists for each bucket until a matching
1101 * request is found */
1102 for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
1103 list_for_each_entry_safe(freq, t, &b->written, wb_list) {
1104 if (freq->wb_page == page)
1105 return freq->wb_head;
1107 list_for_each_entry_safe(freq, t, &b->committing, wb_list) {
1108 if (freq->wb_page == page)
1109 return freq->wb_head;
1117 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1118 int how, struct nfs_commit_info *cinfo)
1120 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1121 filelayout_initiate_commit);
1124 static struct nfs4_deviceid_node *
1125 filelayout_alloc_deviceid_node(struct nfs_server *server,
1126 struct pnfs_device *pdev, gfp_t gfp_flags)
1128 struct nfs4_file_layout_dsaddr *dsaddr;
1130 dsaddr = nfs4_fl_alloc_deviceid_node(server, pdev, gfp_flags);
1133 return &dsaddr->id_node;
1137 filelayout_free_deviceid_node(struct nfs4_deviceid_node *d)
1139 nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1142 static struct pnfs_layout_hdr *
1143 filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
1145 struct nfs4_filelayout *flo;
1147 flo = kzalloc(sizeof(*flo), gfp_flags);
1150 pnfs_init_ds_commit_info(&flo->commit_info);
1151 return &flo->generic_hdr;
1155 filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
1157 kfree_rcu(FILELAYOUT_FROM_HDR(lo), generic_hdr.plh_rcu);
1160 static struct pnfs_ds_commit_info *
1161 filelayout_get_ds_info(struct inode *inode)
1163 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1168 return &FILELAYOUT_FROM_HDR(layout)->commit_info;
1172 filelayout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1173 struct inode *inode)
1175 spin_lock(&inode->i_lock);
1176 pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1177 spin_unlock(&inode->i_lock);
1181 static struct pnfs_layoutdriver_type filelayout_type = {
1182 .id = LAYOUT_NFSV4_1_FILES,
1183 .name = "LAYOUT_NFSV4_1_FILES",
1184 .owner = THIS_MODULE,
1185 .flags = PNFS_LAYOUTGET_ON_OPEN,
1186 .max_layoutget_response = 4096, /* 1 page or so... */
1187 .alloc_layout_hdr = filelayout_alloc_layout_hdr,
1188 .free_layout_hdr = filelayout_free_layout_hdr,
1189 .alloc_lseg = filelayout_alloc_lseg,
1190 .free_lseg = filelayout_free_lseg,
1191 .pg_read_ops = &filelayout_pg_read_ops,
1192 .pg_write_ops = &filelayout_pg_write_ops,
1193 .get_ds_info = &filelayout_get_ds_info,
1194 .release_ds_info = filelayout_release_ds_info,
1195 .mark_request_commit = filelayout_mark_request_commit,
1196 .clear_request_commit = pnfs_generic_clear_request_commit,
1197 .scan_commit_lists = pnfs_generic_scan_commit_lists,
1198 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
1199 .search_commit_reqs = filelayout_search_commit_reqs,
1200 .commit_pagelist = filelayout_commit_pagelist,
1201 .read_pagelist = filelayout_read_pagelist,
1202 .write_pagelist = filelayout_write_pagelist,
1203 .alloc_deviceid_node = filelayout_alloc_deviceid_node,
1204 .free_deviceid_node = filelayout_free_deviceid_node,
1205 .sync = pnfs_nfs_generic_sync,
1208 static int __init nfs4filelayout_init(void)
1210 printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1212 return pnfs_register_layoutdriver(&filelayout_type);
1215 static void __exit nfs4filelayout_exit(void)
1217 printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1219 pnfs_unregister_layoutdriver(&filelayout_type);
1222 MODULE_ALIAS("nfs-layouttype4-1");
1224 module_init(nfs4filelayout_init);
1225 module_exit(nfs4filelayout_exit);