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>
36 #include "nfs4filelayout.h"
38 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
40 MODULE_LICENSE("GPL");
41 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
42 MODULE_DESCRIPTION("The NFSv4 file layout driver");
44 #define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
47 filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
50 u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
53 offset -= flseg->pattern_offset;
55 do_div(tmp, stripe_width);
57 return tmp * flseg->stripe_unit + do_div(offset, flseg->stripe_unit);
60 /* This function is used by the layout driver to calculate the
61 * offset of the file on the dserver based on whether the
62 * layout type is STRIPE_DENSE or STRIPE_SPARSE
65 filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
67 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
69 switch (flseg->stripe_type) {
74 return filelayout_get_dense_offset(flseg, offset);
80 /* For data server errors we don't recover from */
82 filelayout_set_lo_fail(struct pnfs_layout_segment *lseg)
84 if (lseg->pls_range.iomode == IOMODE_RW) {
85 dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__);
86 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
88 dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__);
89 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
93 static int filelayout_async_handle_error(struct rpc_task *task,
94 struct nfs4_state *state,
95 struct nfs_client *clp,
98 if (task->tk_status >= 0)
103 switch (task->tk_status) {
104 case -NFS4ERR_BADSESSION:
105 case -NFS4ERR_BADSLOT:
106 case -NFS4ERR_BAD_HIGH_SLOT:
107 case -NFS4ERR_DEADSESSION:
108 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
109 case -NFS4ERR_SEQ_FALSE_RETRY:
110 case -NFS4ERR_SEQ_MISORDERED:
111 dprintk("%s ERROR %d, Reset session. Exchangeid "
112 "flags 0x%x\n", __func__, task->tk_status,
113 clp->cl_exchange_flags);
114 nfs4_schedule_session_recovery(clp->cl_session);
119 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
121 case -NFS4ERR_RETRY_UNCACHED_REP:
124 dprintk("%s DS error. Retry through MDS %d\n", __func__,
133 /* NFS_PROTO call done callback routines */
135 static int filelayout_read_done_cb(struct rpc_task *task,
136 struct nfs_read_data *data)
138 struct nfs_client *clp = data->ds_clp;
141 dprintk("%s DS read\n", __func__);
143 if (filelayout_async_handle_error(task, data->args.context->state,
144 data->ds_clp, &reset) == -EAGAIN) {
145 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
146 __func__, data->ds_clp, data->ds_clp->cl_session);
148 filelayout_set_lo_fail(data->lseg);
149 nfs4_reset_read(task, data);
150 clp = NFS_SERVER(data->inode)->nfs_client;
152 nfs_restart_rpc(task, clp);
160 * We reference the rpc_cred of the first WRITE that triggers the need for
161 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
162 * rfc5661 is not clear about which credential should be used.
165 filelayout_set_layoutcommit(struct nfs_write_data *wdata)
167 if (FILELAYOUT_LSEG(wdata->lseg)->commit_through_mds ||
168 wdata->res.verf->committed == NFS_FILE_SYNC)
171 pnfs_set_layoutcommit(wdata);
172 dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino,
173 (unsigned long) NFS_I(wdata->inode)->layout->plh_lwb);
177 * Call ops for the async read/write cases
178 * In the case of dense layouts, the offset needs to be reset to its
181 static void filelayout_read_prepare(struct rpc_task *task, void *data)
183 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
185 rdata->read_done_cb = filelayout_read_done_cb;
187 if (nfs41_setup_sequence(rdata->ds_clp->cl_session,
188 &rdata->args.seq_args, &rdata->res.seq_res,
192 rpc_call_start(task);
195 static void filelayout_read_call_done(struct rpc_task *task, void *data)
197 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
199 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
201 /* Note this may cause RPC to be resent */
202 rdata->mds_ops->rpc_call_done(task, data);
205 static void filelayout_read_release(void *data)
207 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
209 rdata->mds_ops->rpc_release(data);
212 static int filelayout_write_done_cb(struct rpc_task *task,
213 struct nfs_write_data *data)
217 if (filelayout_async_handle_error(task, data->args.context->state,
218 data->ds_clp, &reset) == -EAGAIN) {
219 struct nfs_client *clp;
221 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
222 __func__, data->ds_clp, data->ds_clp->cl_session);
224 filelayout_set_lo_fail(data->lseg);
225 nfs4_reset_write(task, data);
226 clp = NFS_SERVER(data->inode)->nfs_client;
229 nfs_restart_rpc(task, clp);
233 filelayout_set_layoutcommit(data);
237 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
238 static void prepare_to_resend_writes(struct nfs_write_data *data)
240 struct nfs_page *first = nfs_list_entry(data->pages.next);
242 data->task.tk_status = 0;
243 memcpy(data->verf.verifier, first->wb_verf.verifier,
244 sizeof(first->wb_verf.verifier));
245 data->verf.verifier[0]++; /* ensure verifier mismatch */
248 static int filelayout_commit_done_cb(struct rpc_task *task,
249 struct nfs_write_data *data)
253 if (filelayout_async_handle_error(task, data->args.context->state,
254 data->ds_clp, &reset) == -EAGAIN) {
255 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
256 __func__, data->ds_clp, data->ds_clp->cl_session);
258 prepare_to_resend_writes(data);
259 filelayout_set_lo_fail(data->lseg);
261 nfs_restart_rpc(task, data->ds_clp);
268 static void filelayout_write_prepare(struct rpc_task *task, void *data)
270 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
272 if (nfs41_setup_sequence(wdata->ds_clp->cl_session,
273 &wdata->args.seq_args, &wdata->res.seq_res,
277 rpc_call_start(task);
280 static void filelayout_write_call_done(struct rpc_task *task, void *data)
282 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
284 /* Note this may cause RPC to be resent */
285 wdata->mds_ops->rpc_call_done(task, data);
288 static void filelayout_write_release(void *data)
290 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
292 wdata->mds_ops->rpc_release(data);
295 static void filelayout_commit_release(void *data)
297 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
299 nfs_commit_release_pages(wdata);
300 if (atomic_dec_and_test(&NFS_I(wdata->inode)->commits_outstanding))
301 nfs_commit_clear_lock(NFS_I(wdata->inode));
302 nfs_commitdata_release(wdata);
305 struct rpc_call_ops filelayout_read_call_ops = {
306 .rpc_call_prepare = filelayout_read_prepare,
307 .rpc_call_done = filelayout_read_call_done,
308 .rpc_release = filelayout_read_release,
311 struct rpc_call_ops filelayout_write_call_ops = {
312 .rpc_call_prepare = filelayout_write_prepare,
313 .rpc_call_done = filelayout_write_call_done,
314 .rpc_release = filelayout_write_release,
317 struct rpc_call_ops filelayout_commit_call_ops = {
318 .rpc_call_prepare = filelayout_write_prepare,
319 .rpc_call_done = filelayout_write_call_done,
320 .rpc_release = filelayout_commit_release,
323 static enum pnfs_try_status
324 filelayout_read_pagelist(struct nfs_read_data *data)
326 struct pnfs_layout_segment *lseg = data->lseg;
327 struct nfs4_pnfs_ds *ds;
328 loff_t offset = data->args.offset;
333 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
334 __func__, data->inode->i_ino,
335 data->args.pgbase, (size_t)data->args.count, offset);
337 /* Retrieve the correct rpc_client for the byte range */
338 j = nfs4_fl_calc_j_index(lseg, offset);
339 idx = nfs4_fl_calc_ds_index(lseg, j);
340 ds = nfs4_fl_prepare_ds(lseg, idx);
342 /* Either layout fh index faulty, or ds connect failed */
343 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
344 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
345 return PNFS_NOT_ATTEMPTED;
347 dprintk("%s USE DS:ip %x %hu\n", __func__,
348 ntohl(ds->ds_ip_addr), ntohs(ds->ds_port));
350 /* No multipath support. Use first DS */
351 data->ds_clp = ds->ds_clp;
352 fh = nfs4_fl_select_ds_fh(lseg, j);
356 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
357 data->mds_offset = offset;
359 /* Perform an asynchronous read to ds */
360 status = nfs_initiate_read(data, ds->ds_clp->cl_rpcclient,
361 &filelayout_read_call_ops);
363 return PNFS_ATTEMPTED;
366 /* Perform async writes. */
367 static enum pnfs_try_status
368 filelayout_write_pagelist(struct nfs_write_data *data, int sync)
370 struct pnfs_layout_segment *lseg = data->lseg;
371 struct nfs4_pnfs_ds *ds;
372 loff_t offset = data->args.offset;
377 /* Retrieve the correct rpc_client for the byte range */
378 j = nfs4_fl_calc_j_index(lseg, offset);
379 idx = nfs4_fl_calc_ds_index(lseg, j);
380 ds = nfs4_fl_prepare_ds(lseg, idx);
382 printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__);
383 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
384 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
385 return PNFS_NOT_ATTEMPTED;
387 dprintk("%s ino %lu sync %d req %Zu@%llu DS:%x:%hu\n", __func__,
388 data->inode->i_ino, sync, (size_t) data->args.count, offset,
389 ntohl(ds->ds_ip_addr), ntohs(ds->ds_port));
391 data->write_done_cb = filelayout_write_done_cb;
392 data->ds_clp = ds->ds_clp;
393 fh = nfs4_fl_select_ds_fh(lseg, j);
397 * Get the file offset on the dserver. Set the write offset to
398 * this offset and save the original offset.
400 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
402 /* Perform an asynchronous write */
403 status = nfs_initiate_write(data, ds->ds_clp->cl_rpcclient,
404 &filelayout_write_call_ops, sync);
406 return PNFS_ATTEMPTED;
410 * filelayout_check_layout()
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.
418 filelayout_check_layout(struct pnfs_layout_hdr *lo,
419 struct nfs4_filelayout_segment *fl,
420 struct nfs4_layoutget_res *lgr,
421 struct nfs4_deviceid *id,
424 struct nfs4_deviceid_node *d;
425 struct nfs4_file_layout_dsaddr *dsaddr;
426 int status = -EINVAL;
427 struct nfs_server *nfss = NFS_SERVER(lo->plh_inode);
429 dprintk("--> %s\n", __func__);
431 /* FIXME: remove this check when layout segment support is added */
432 if (lgr->range.offset != 0 ||
433 lgr->range.length != NFS4_MAX_UINT64) {
434 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
439 if (fl->pattern_offset > lgr->range.offset) {
440 dprintk("%s pattern_offset %lld too large\n",
441 __func__, fl->pattern_offset);
445 if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) {
446 dprintk("%s Invalid stripe unit (%u)\n",
447 __func__, fl->stripe_unit);
451 /* find and reference the deviceid */
452 d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld,
453 NFS_SERVER(lo->plh_inode)->nfs_client, id);
455 dsaddr = get_device_info(lo->plh_inode, id, gfp_flags);
459 dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
462 if (fl->first_stripe_index < 0 ||
463 fl->first_stripe_index >= dsaddr->stripe_count) {
464 dprintk("%s Bad first_stripe_index %d\n",
465 __func__, fl->first_stripe_index);
469 if ((fl->stripe_type == STRIPE_SPARSE &&
470 fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
471 (fl->stripe_type == STRIPE_DENSE &&
472 fl->num_fh != dsaddr->stripe_count)) {
473 dprintk("%s num_fh %u not valid for given packing\n",
474 __func__, fl->num_fh);
478 if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
479 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
480 "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
486 dprintk("--> %s returns %d\n", __func__, status);
489 nfs4_fl_put_deviceid(dsaddr);
493 static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
497 for (i = 0; i < fl->num_fh; i++) {
498 if (!fl->fh_array[i])
500 kfree(fl->fh_array[i]);
507 _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
509 filelayout_free_fh_array(fl);
514 filelayout_decode_layout(struct pnfs_layout_hdr *flo,
515 struct nfs4_filelayout_segment *fl,
516 struct nfs4_layoutget_res *lgr,
517 struct nfs4_deviceid *id,
520 struct xdr_stream stream;
522 struct page *scratch;
527 dprintk("%s: set_layout_map Begin\n", __func__);
529 scratch = alloc_page(gfp_flags);
533 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
534 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
536 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
538 p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
542 memcpy(id, p, sizeof(*id));
543 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
544 nfs4_print_deviceid(id);
546 nfl_util = be32_to_cpup(p++);
547 if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
548 fl->commit_through_mds = 1;
549 if (nfl_util & NFL4_UFLG_DENSE)
550 fl->stripe_type = STRIPE_DENSE;
552 fl->stripe_type = STRIPE_SPARSE;
553 fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
555 fl->first_stripe_index = be32_to_cpup(p++);
556 p = xdr_decode_hyper(p, &fl->pattern_offset);
557 fl->num_fh = be32_to_cpup(p++);
559 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
560 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
563 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
564 * Futher checking is done in filelayout_check_layout */
565 if (fl->num_fh < 0 || fl->num_fh >
566 max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
569 if (fl->num_fh > 0) {
570 fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *),
576 for (i = 0; i < fl->num_fh; i++) {
577 /* Do we want to use a mempool here? */
578 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
579 if (!fl->fh_array[i])
582 p = xdr_inline_decode(&stream, 4);
585 fl->fh_array[i]->size = be32_to_cpup(p++);
586 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
587 printk(KERN_ERR "Too big fh %d received %d\n",
588 i, fl->fh_array[i]->size);
592 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
595 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
596 dprintk("DEBUG: %s: fh len %d\n", __func__,
597 fl->fh_array[i]->size);
600 __free_page(scratch);
604 filelayout_free_fh_array(fl);
606 __free_page(scratch);
611 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
613 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
615 dprintk("--> %s\n", __func__);
616 nfs4_fl_put_deviceid(fl->dsaddr);
617 kfree(fl->commit_buckets);
618 _filelayout_free_lseg(fl);
621 static struct pnfs_layout_segment *
622 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
623 struct nfs4_layoutget_res *lgr,
626 struct nfs4_filelayout_segment *fl;
628 struct nfs4_deviceid id;
630 dprintk("--> %s\n", __func__);
631 fl = kzalloc(sizeof(*fl), gfp_flags);
635 rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
636 if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
637 _filelayout_free_lseg(fl);
641 /* This assumes there is only one IOMODE_RW lseg. What
642 * we really want to do is have a layout_hdr level
643 * dictionary of <multipath_list4, fh> keys, each
644 * associated with a struct list_head, populated by calls
645 * to filelayout_write_pagelist().
647 if ((!fl->commit_through_mds) && (lgr->range.iomode == IOMODE_RW)) {
649 int size = (fl->stripe_type == STRIPE_SPARSE) ?
650 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
652 fl->commit_buckets = kcalloc(size, sizeof(struct list_head), gfp_flags);
653 if (!fl->commit_buckets) {
654 filelayout_free_lseg(&fl->generic_hdr);
657 fl->number_of_buckets = size;
658 for (i = 0; i < size; i++)
659 INIT_LIST_HEAD(&fl->commit_buckets[i]);
661 return &fl->generic_hdr;
665 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
667 * return true : coalesce page
668 * return false : don't coalesce page
671 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
672 struct nfs_page *req)
674 u64 p_stripe, r_stripe;
677 if (!pnfs_generic_pg_test(pgio, prev, req) ||
678 !nfs_generic_pg_test(pgio, prev, req))
683 p_stripe = (u64)prev->wb_index << PAGE_CACHE_SHIFT;
684 r_stripe = (u64)req->wb_index << PAGE_CACHE_SHIFT;
685 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
687 do_div(p_stripe, stripe_unit);
688 do_div(r_stripe, stripe_unit);
690 return (p_stripe == r_stripe);
693 static bool filelayout_mark_pnfs_commit(struct pnfs_layout_segment *lseg)
695 return !FILELAYOUT_LSEG(lseg)->commit_through_mds;
698 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
700 if (fl->stripe_type == STRIPE_SPARSE)
701 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
706 struct list_head *filelayout_choose_commit_list(struct nfs_page *req)
708 struct pnfs_layout_segment *lseg = req->wb_commit_lseg;
709 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
711 struct list_head *list;
713 /* Note that we are calling nfs4_fl_calc_j_index on each page
714 * that ends up being committed to a data server. An attractive
715 * alternative is to add a field to nfs_write_data and nfs_page
716 * to store the value calculated in filelayout_write_pagelist
717 * and just use that here.
719 j = nfs4_fl_calc_j_index(lseg,
720 (loff_t)req->wb_index << PAGE_CACHE_SHIFT);
721 i = select_bucket_index(fl, j);
722 list = &fl->commit_buckets[i];
723 if (list_empty(list)) {
724 /* Non-empty buckets hold a reference on the lseg */
730 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
732 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
734 if (flseg->stripe_type == STRIPE_SPARSE)
737 return nfs4_fl_calc_ds_index(lseg, i);
740 static struct nfs_fh *
741 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
743 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
745 if (flseg->stripe_type == STRIPE_SPARSE) {
746 if (flseg->num_fh == 1)
748 else if (flseg->num_fh == 0)
749 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
752 return flseg->fh_array[i];
755 static int filelayout_initiate_commit(struct nfs_write_data *data, int how)
757 struct pnfs_layout_segment *lseg = data->lseg;
758 struct nfs4_pnfs_ds *ds;
762 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
763 ds = nfs4_fl_prepare_ds(lseg, idx);
765 printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__);
766 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
767 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
768 prepare_to_resend_writes(data);
769 data->mds_ops->rpc_release(data);
772 dprintk("%s ino %lu, how %d\n", __func__, data->inode->i_ino, how);
773 data->write_done_cb = filelayout_commit_done_cb;
774 data->ds_clp = ds->ds_clp;
775 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
778 return nfs_initiate_commit(data, ds->ds_clp->cl_rpcclient,
779 &filelayout_commit_call_ops, how);
783 * This is only useful while we are using whole file layouts.
785 static struct pnfs_layout_segment *find_only_write_lseg(struct inode *inode)
787 struct pnfs_layout_segment *lseg, *rv = NULL;
789 spin_lock(&inode->i_lock);
790 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
791 if (lseg->pls_range.iomode == IOMODE_RW)
793 spin_unlock(&inode->i_lock);
797 static int alloc_ds_commits(struct inode *inode, struct list_head *list)
799 struct pnfs_layout_segment *lseg;
800 struct nfs4_filelayout_segment *fl;
801 struct nfs_write_data *data;
804 /* Won't need this when non-whole file layout segments are supported
805 * instead we will use a pnfs_layout_hdr structure */
806 lseg = find_only_write_lseg(inode);
809 fl = FILELAYOUT_LSEG(lseg);
810 for (i = 0; i < fl->number_of_buckets; i++) {
811 if (list_empty(&fl->commit_buckets[i]))
813 data = nfs_commitdata_alloc();
816 data->ds_commit_index = i;
818 list_add(&data->pages, list);
824 for (j = i; j < fl->number_of_buckets; j++) {
825 if (list_empty(&fl->commit_buckets[i]))
827 nfs_retry_commit(&fl->commit_buckets[i], lseg);
828 put_lseg(lseg); /* associated with emptying bucket */
831 /* Caller will clean up entries put on list */
835 /* This follows nfs_commit_list pretty closely */
837 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
840 struct nfs_write_data *data, *tmp;
843 if (!list_empty(mds_pages)) {
844 data = nfs_commitdata_alloc();
848 list_add(&data->pages, &list);
851 if (alloc_ds_commits(inode, &list))
854 list_for_each_entry_safe(data, tmp, &list, pages) {
855 list_del_init(&data->pages);
856 atomic_inc(&NFS_I(inode)->commits_outstanding);
858 nfs_init_commit(data, mds_pages, NULL);
859 nfs_initiate_commit(data, NFS_CLIENT(inode),
862 nfs_init_commit(data, &FILELAYOUT_LSEG(data->lseg)->commit_buckets[data->ds_commit_index], data->lseg);
863 filelayout_initiate_commit(data, how);
868 list_for_each_entry_safe(data, tmp, &list, pages) {
869 nfs_retry_commit(&data->pages, data->lseg);
870 list_del_init(&data->pages);
871 nfs_commit_free(data);
873 nfs_retry_commit(mds_pages, NULL);
874 nfs_commit_clear_lock(NFS_I(inode));
879 filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
881 nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
884 static struct pnfs_layoutdriver_type filelayout_type = {
885 .id = LAYOUT_NFSV4_1_FILES,
886 .name = "LAYOUT_NFSV4_1_FILES",
887 .owner = THIS_MODULE,
888 .alloc_lseg = filelayout_alloc_lseg,
889 .free_lseg = filelayout_free_lseg,
890 .pg_test = filelayout_pg_test,
891 .mark_pnfs_commit = filelayout_mark_pnfs_commit,
892 .choose_commit_list = filelayout_choose_commit_list,
893 .commit_pagelist = filelayout_commit_pagelist,
894 .read_pagelist = filelayout_read_pagelist,
895 .write_pagelist = filelayout_write_pagelist,
896 .free_deviceid_node = filelayout_free_deveiceid_node,
899 static int __init nfs4filelayout_init(void)
901 printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
903 return pnfs_register_layoutdriver(&filelayout_type);
906 static void __exit nfs4filelayout_exit(void)
908 printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
910 pnfs_unregister_layoutdriver(&filelayout_type);
913 module_init(nfs4filelayout_init);
914 module_exit(nfs4filelayout_exit);