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) wdata->lseg->pls_end_pos);
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 if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags))
338 return PNFS_NOT_ATTEMPTED;
340 /* Retrieve the correct rpc_client for the byte range */
341 j = nfs4_fl_calc_j_index(lseg, offset);
342 idx = nfs4_fl_calc_ds_index(lseg, j);
343 ds = nfs4_fl_prepare_ds(lseg, idx);
345 /* Either layout fh index faulty, or ds connect failed */
346 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
347 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
348 return PNFS_NOT_ATTEMPTED;
350 dprintk("%s USE DS: %s\n", __func__, ds->ds_remotestr);
352 /* No multipath support. Use first DS */
353 data->ds_clp = ds->ds_clp;
354 fh = nfs4_fl_select_ds_fh(lseg, j);
358 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
359 data->mds_offset = offset;
361 /* Perform an asynchronous read to ds */
362 status = nfs_initiate_read(data, ds->ds_clp->cl_rpcclient,
363 &filelayout_read_call_ops);
365 return PNFS_ATTEMPTED;
368 /* Perform async writes. */
369 static enum pnfs_try_status
370 filelayout_write_pagelist(struct nfs_write_data *data, int sync)
372 struct pnfs_layout_segment *lseg = data->lseg;
373 struct nfs4_pnfs_ds *ds;
374 loff_t offset = data->args.offset;
379 if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags))
380 return PNFS_NOT_ATTEMPTED;
382 /* Retrieve the correct rpc_client for the byte range */
383 j = nfs4_fl_calc_j_index(lseg, offset);
384 idx = nfs4_fl_calc_ds_index(lseg, j);
385 ds = nfs4_fl_prepare_ds(lseg, idx);
387 printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__);
388 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
389 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
390 return PNFS_NOT_ATTEMPTED;
392 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s\n", __func__,
393 data->inode->i_ino, sync, (size_t) data->args.count, offset,
396 data->write_done_cb = filelayout_write_done_cb;
397 data->ds_clp = ds->ds_clp;
398 fh = nfs4_fl_select_ds_fh(lseg, j);
402 * Get the file offset on the dserver. Set the write offset to
403 * this offset and save the original offset.
405 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
407 /* Perform an asynchronous write */
408 status = nfs_initiate_write(data, ds->ds_clp->cl_rpcclient,
409 &filelayout_write_call_ops, sync);
411 return PNFS_ATTEMPTED;
415 * filelayout_check_layout()
417 * Make sure layout segment parameters are sane WRT the device.
418 * At this point no generic layer initialization of the lseg has occurred,
419 * and nothing has been added to the layout_hdr cache.
423 filelayout_check_layout(struct pnfs_layout_hdr *lo,
424 struct nfs4_filelayout_segment *fl,
425 struct nfs4_layoutget_res *lgr,
426 struct nfs4_deviceid *id,
429 struct nfs4_deviceid_node *d;
430 struct nfs4_file_layout_dsaddr *dsaddr;
431 int status = -EINVAL;
432 struct nfs_server *nfss = NFS_SERVER(lo->plh_inode);
434 dprintk("--> %s\n", __func__);
436 /* FIXME: remove this check when layout segment support is added */
437 if (lgr->range.offset != 0 ||
438 lgr->range.length != NFS4_MAX_UINT64) {
439 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
444 if (fl->pattern_offset > lgr->range.offset) {
445 dprintk("%s pattern_offset %lld too large\n",
446 __func__, fl->pattern_offset);
450 if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) {
451 dprintk("%s Invalid stripe unit (%u)\n",
452 __func__, fl->stripe_unit);
456 /* find and reference the deviceid */
457 d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld,
458 NFS_SERVER(lo->plh_inode)->nfs_client, id);
460 dsaddr = get_device_info(lo->plh_inode, id, gfp_flags);
464 dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
465 /* Found deviceid is being reaped */
466 if (test_bit(NFS_DEVICEID_INVALID, &dsaddr->id_node.flags))
471 if (fl->first_stripe_index < 0 ||
472 fl->first_stripe_index >= dsaddr->stripe_count) {
473 dprintk("%s Bad first_stripe_index %d\n",
474 __func__, fl->first_stripe_index);
478 if ((fl->stripe_type == STRIPE_SPARSE &&
479 fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
480 (fl->stripe_type == STRIPE_DENSE &&
481 fl->num_fh != dsaddr->stripe_count)) {
482 dprintk("%s num_fh %u not valid for given packing\n",
483 __func__, fl->num_fh);
487 if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
488 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
489 "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
495 dprintk("--> %s returns %d\n", __func__, status);
498 nfs4_fl_put_deviceid(dsaddr);
502 static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
506 for (i = 0; i < fl->num_fh; i++) {
507 if (!fl->fh_array[i])
509 kfree(fl->fh_array[i]);
516 _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
518 filelayout_free_fh_array(fl);
523 filelayout_decode_layout(struct pnfs_layout_hdr *flo,
524 struct nfs4_filelayout_segment *fl,
525 struct nfs4_layoutget_res *lgr,
526 struct nfs4_deviceid *id,
529 struct xdr_stream stream;
531 struct page *scratch;
536 dprintk("%s: set_layout_map Begin\n", __func__);
538 scratch = alloc_page(gfp_flags);
542 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
543 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
545 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
547 p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
551 memcpy(id, p, sizeof(*id));
552 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
553 nfs4_print_deviceid(id);
555 nfl_util = be32_to_cpup(p++);
556 if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
557 fl->commit_through_mds = 1;
558 if (nfl_util & NFL4_UFLG_DENSE)
559 fl->stripe_type = STRIPE_DENSE;
561 fl->stripe_type = STRIPE_SPARSE;
562 fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
564 fl->first_stripe_index = be32_to_cpup(p++);
565 p = xdr_decode_hyper(p, &fl->pattern_offset);
566 fl->num_fh = be32_to_cpup(p++);
568 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
569 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
572 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
573 * Futher checking is done in filelayout_check_layout */
574 if (fl->num_fh < 0 || fl->num_fh >
575 max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
578 if (fl->num_fh > 0) {
579 fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *),
585 for (i = 0; i < fl->num_fh; i++) {
586 /* Do we want to use a mempool here? */
587 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
588 if (!fl->fh_array[i])
591 p = xdr_inline_decode(&stream, 4);
594 fl->fh_array[i]->size = be32_to_cpup(p++);
595 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
596 printk(KERN_ERR "Too big fh %d received %d\n",
597 i, fl->fh_array[i]->size);
601 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
604 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
605 dprintk("DEBUG: %s: fh len %d\n", __func__,
606 fl->fh_array[i]->size);
609 __free_page(scratch);
613 filelayout_free_fh_array(fl);
615 __free_page(scratch);
620 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
622 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
624 dprintk("--> %s\n", __func__);
625 nfs4_fl_put_deviceid(fl->dsaddr);
626 kfree(fl->commit_buckets);
627 _filelayout_free_lseg(fl);
630 static struct pnfs_layout_segment *
631 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
632 struct nfs4_layoutget_res *lgr,
635 struct nfs4_filelayout_segment *fl;
637 struct nfs4_deviceid id;
639 dprintk("--> %s\n", __func__);
640 fl = kzalloc(sizeof(*fl), gfp_flags);
644 rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
645 if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
646 _filelayout_free_lseg(fl);
650 /* This assumes there is only one IOMODE_RW lseg. What
651 * we really want to do is have a layout_hdr level
652 * dictionary of <multipath_list4, fh> keys, each
653 * associated with a struct list_head, populated by calls
654 * to filelayout_write_pagelist().
656 if ((!fl->commit_through_mds) && (lgr->range.iomode == IOMODE_RW)) {
658 int size = (fl->stripe_type == STRIPE_SPARSE) ?
659 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
661 fl->commit_buckets = kcalloc(size, sizeof(struct list_head), gfp_flags);
662 if (!fl->commit_buckets) {
663 filelayout_free_lseg(&fl->generic_hdr);
666 fl->number_of_buckets = size;
667 for (i = 0; i < size; i++)
668 INIT_LIST_HEAD(&fl->commit_buckets[i]);
670 return &fl->generic_hdr;
674 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
676 * return true : coalesce page
677 * return false : don't coalesce page
680 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
681 struct nfs_page *req)
683 u64 p_stripe, r_stripe;
686 if (!pnfs_generic_pg_test(pgio, prev, req) ||
687 !nfs_generic_pg_test(pgio, prev, req))
690 p_stripe = (u64)prev->wb_index << PAGE_CACHE_SHIFT;
691 r_stripe = (u64)req->wb_index << PAGE_CACHE_SHIFT;
692 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
694 do_div(p_stripe, stripe_unit);
695 do_div(r_stripe, stripe_unit);
697 return (p_stripe == r_stripe);
701 filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
702 struct nfs_page *req)
704 BUG_ON(pgio->pg_lseg != NULL);
706 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
712 /* If no lseg, fall back to read through mds */
713 if (pgio->pg_lseg == NULL)
714 nfs_pageio_init_read_mds(pgio, pgio->pg_inode);
718 filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
719 struct nfs_page *req)
721 BUG_ON(pgio->pg_lseg != NULL);
723 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
729 /* If no lseg, fall back to write through mds */
730 if (pgio->pg_lseg == NULL)
731 nfs_pageio_init_write_mds(pgio, pgio->pg_inode,
735 static const struct nfs_pageio_ops filelayout_pg_read_ops = {
736 .pg_init = filelayout_pg_init_read,
737 .pg_test = filelayout_pg_test,
738 .pg_doio = pnfs_generic_pg_readpages,
741 static const struct nfs_pageio_ops filelayout_pg_write_ops = {
742 .pg_init = filelayout_pg_init_write,
743 .pg_test = filelayout_pg_test,
744 .pg_doio = pnfs_generic_pg_writepages,
747 static bool filelayout_mark_pnfs_commit(struct pnfs_layout_segment *lseg)
749 return !FILELAYOUT_LSEG(lseg)->commit_through_mds;
752 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
754 if (fl->stripe_type == STRIPE_SPARSE)
755 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
760 struct list_head *filelayout_choose_commit_list(struct nfs_page *req)
762 struct pnfs_layout_segment *lseg = req->wb_commit_lseg;
763 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
765 struct list_head *list;
767 /* Note that we are calling nfs4_fl_calc_j_index on each page
768 * that ends up being committed to a data server. An attractive
769 * alternative is to add a field to nfs_write_data and nfs_page
770 * to store the value calculated in filelayout_write_pagelist
771 * and just use that here.
773 j = nfs4_fl_calc_j_index(lseg,
774 (loff_t)req->wb_index << PAGE_CACHE_SHIFT);
775 i = select_bucket_index(fl, j);
776 list = &fl->commit_buckets[i];
777 if (list_empty(list)) {
778 /* Non-empty buckets hold a reference on the lseg */
784 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
786 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
788 if (flseg->stripe_type == STRIPE_SPARSE)
791 return nfs4_fl_calc_ds_index(lseg, i);
794 static struct nfs_fh *
795 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
797 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
799 if (flseg->stripe_type == STRIPE_SPARSE) {
800 if (flseg->num_fh == 1)
802 else if (flseg->num_fh == 0)
803 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
806 return flseg->fh_array[i];
809 static int filelayout_initiate_commit(struct nfs_write_data *data, int how)
811 struct pnfs_layout_segment *lseg = data->lseg;
812 struct nfs4_pnfs_ds *ds;
816 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
817 ds = nfs4_fl_prepare_ds(lseg, idx);
819 printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__);
820 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
821 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
822 prepare_to_resend_writes(data);
823 data->mds_ops->rpc_release(data);
826 dprintk("%s ino %lu, how %d\n", __func__, data->inode->i_ino, how);
827 data->write_done_cb = filelayout_commit_done_cb;
828 data->ds_clp = ds->ds_clp;
829 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
832 return nfs_initiate_commit(data, ds->ds_clp->cl_rpcclient,
833 &filelayout_commit_call_ops, how);
837 * This is only useful while we are using whole file layouts.
839 static struct pnfs_layout_segment *find_only_write_lseg(struct inode *inode)
841 struct pnfs_layout_segment *lseg, *rv = NULL;
843 spin_lock(&inode->i_lock);
844 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
845 if (lseg->pls_range.iomode == IOMODE_RW)
847 spin_unlock(&inode->i_lock);
851 static int alloc_ds_commits(struct inode *inode, struct list_head *list)
853 struct pnfs_layout_segment *lseg;
854 struct nfs4_filelayout_segment *fl;
855 struct nfs_write_data *data;
858 /* Won't need this when non-whole file layout segments are supported
859 * instead we will use a pnfs_layout_hdr structure */
860 lseg = find_only_write_lseg(inode);
863 fl = FILELAYOUT_LSEG(lseg);
864 for (i = 0; i < fl->number_of_buckets; i++) {
865 if (list_empty(&fl->commit_buckets[i]))
867 data = nfs_commitdata_alloc();
870 data->ds_commit_index = i;
872 list_add(&data->pages, list);
878 for (j = i; j < fl->number_of_buckets; j++) {
879 if (list_empty(&fl->commit_buckets[i]))
881 nfs_retry_commit(&fl->commit_buckets[i], lseg);
882 put_lseg(lseg); /* associated with emptying bucket */
885 /* Caller will clean up entries put on list */
889 /* This follows nfs_commit_list pretty closely */
891 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
894 struct nfs_write_data *data, *tmp;
897 if (!list_empty(mds_pages)) {
898 data = nfs_commitdata_alloc();
902 list_add(&data->pages, &list);
905 if (alloc_ds_commits(inode, &list))
908 list_for_each_entry_safe(data, tmp, &list, pages) {
909 list_del_init(&data->pages);
910 atomic_inc(&NFS_I(inode)->commits_outstanding);
912 nfs_init_commit(data, mds_pages, NULL);
913 nfs_initiate_commit(data, NFS_CLIENT(inode),
916 nfs_init_commit(data, &FILELAYOUT_LSEG(data->lseg)->commit_buckets[data->ds_commit_index], data->lseg);
917 filelayout_initiate_commit(data, how);
922 list_for_each_entry_safe(data, tmp, &list, pages) {
923 nfs_retry_commit(&data->pages, data->lseg);
924 list_del_init(&data->pages);
925 nfs_commit_free(data);
927 nfs_retry_commit(mds_pages, NULL);
928 nfs_commit_clear_lock(NFS_I(inode));
933 filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
935 nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
938 static struct pnfs_layoutdriver_type filelayout_type = {
939 .id = LAYOUT_NFSV4_1_FILES,
940 .name = "LAYOUT_NFSV4_1_FILES",
941 .owner = THIS_MODULE,
942 .alloc_lseg = filelayout_alloc_lseg,
943 .free_lseg = filelayout_free_lseg,
944 .pg_read_ops = &filelayout_pg_read_ops,
945 .pg_write_ops = &filelayout_pg_write_ops,
946 .mark_pnfs_commit = filelayout_mark_pnfs_commit,
947 .choose_commit_list = filelayout_choose_commit_list,
948 .commit_pagelist = filelayout_commit_pagelist,
949 .read_pagelist = filelayout_read_pagelist,
950 .write_pagelist = filelayout_write_pagelist,
951 .free_deviceid_node = filelayout_free_deveiceid_node,
954 static int __init nfs4filelayout_init(void)
956 printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
958 return pnfs_register_layoutdriver(&filelayout_type);
961 static void __exit nfs4filelayout_exit(void)
963 printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
965 pnfs_unregister_layoutdriver(&filelayout_type);
968 module_init(nfs4filelayout_init);
969 module_exit(nfs4filelayout_exit);