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>
36 #include <linux/sunrpc/metrics.h>
39 #include "delegation.h"
40 #include "nfs4filelayout.h"
42 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
44 MODULE_LICENSE("GPL");
45 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
46 MODULE_DESCRIPTION("The NFSv4 file layout driver");
48 #define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
51 filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
54 u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
58 offset -= flseg->pattern_offset;
59 stripe_no = div_u64(offset, stripe_width);
60 div_u64_rem(offset, flseg->stripe_unit, &rem);
62 return stripe_no * flseg->stripe_unit + rem;
65 /* This function is used by the layout driver to calculate the
66 * offset of the file on the dserver based on whether the
67 * layout type is STRIPE_DENSE or STRIPE_SPARSE
70 filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
72 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
74 switch (flseg->stripe_type) {
79 return filelayout_get_dense_offset(flseg, offset);
85 static int filelayout_async_handle_error(struct rpc_task *task,
86 struct nfs4_state *state,
87 struct nfs_client *clp,
90 struct nfs_server *mds_server = NFS_SERVER(state->inode);
91 struct nfs_client *mds_client = mds_server->nfs_client;
93 if (task->tk_status >= 0)
97 switch (task->tk_status) {
98 /* MDS state errors */
99 case -NFS4ERR_DELEG_REVOKED:
100 case -NFS4ERR_ADMIN_REVOKED:
101 case -NFS4ERR_BAD_STATEID:
102 nfs_remove_bad_delegation(state->inode);
103 case -NFS4ERR_OPENMODE:
104 nfs4_schedule_stateid_recovery(mds_server, state);
105 goto wait_on_recovery;
106 case -NFS4ERR_EXPIRED:
107 nfs4_schedule_stateid_recovery(mds_server, state);
108 nfs4_schedule_lease_recovery(mds_client);
109 goto wait_on_recovery;
110 /* DS session errors */
111 case -NFS4ERR_BADSESSION:
112 case -NFS4ERR_BADSLOT:
113 case -NFS4ERR_BAD_HIGH_SLOT:
114 case -NFS4ERR_DEADSESSION:
115 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
116 case -NFS4ERR_SEQ_FALSE_RETRY:
117 case -NFS4ERR_SEQ_MISORDERED:
118 dprintk("%s ERROR %d, Reset session. Exchangeid "
119 "flags 0x%x\n", __func__, task->tk_status,
120 clp->cl_exchange_flags);
121 nfs4_schedule_session_recovery(clp->cl_session);
126 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
128 case -NFS4ERR_RETRY_UNCACHED_REP:
131 dprintk("%s DS error. Retry through MDS %d\n", __func__,
140 rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
141 if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
142 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
146 /* NFS_PROTO call done callback routines */
148 static int filelayout_read_done_cb(struct rpc_task *task,
149 struct nfs_read_data *data)
153 dprintk("%s DS read\n", __func__);
155 if (filelayout_async_handle_error(task, data->args.context->state,
156 data->ds_clp, &reset) == -EAGAIN) {
157 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
158 __func__, data->ds_clp, data->ds_clp->cl_session);
160 pnfs_set_lo_fail(data->lseg);
161 nfs4_reset_read(task, data);
163 rpc_restart_call_prepare(task);
171 * We reference the rpc_cred of the first WRITE that triggers the need for
172 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
173 * rfc5661 is not clear about which credential should be used.
176 filelayout_set_layoutcommit(struct nfs_write_data *wdata)
178 if (FILELAYOUT_LSEG(wdata->lseg)->commit_through_mds ||
179 wdata->res.verf->committed == NFS_FILE_SYNC)
182 pnfs_set_layoutcommit(wdata);
183 dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino,
184 (unsigned long) NFS_I(wdata->inode)->layout->plh_lwb);
188 * Call ops for the async read/write cases
189 * In the case of dense layouts, the offset needs to be reset to its
192 static void filelayout_read_prepare(struct rpc_task *task, void *data)
194 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
196 rdata->read_done_cb = filelayout_read_done_cb;
198 if (nfs41_setup_sequence(rdata->ds_clp->cl_session,
199 &rdata->args.seq_args, &rdata->res.seq_res,
203 rpc_call_start(task);
206 static void filelayout_read_call_done(struct rpc_task *task, void *data)
208 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
210 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
212 /* Note this may cause RPC to be resent */
213 rdata->mds_ops->rpc_call_done(task, data);
216 static void filelayout_read_count_stats(struct rpc_task *task, void *data)
218 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
220 rpc_count_iostats(task, NFS_SERVER(rdata->inode)->client->cl_metrics);
223 static void filelayout_read_release(void *data)
225 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
227 put_lseg(rdata->lseg);
228 rdata->mds_ops->rpc_release(data);
231 static int filelayout_write_done_cb(struct rpc_task *task,
232 struct nfs_write_data *data)
236 if (filelayout_async_handle_error(task, data->args.context->state,
237 data->ds_clp, &reset) == -EAGAIN) {
238 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
239 __func__, data->ds_clp, data->ds_clp->cl_session);
241 pnfs_set_lo_fail(data->lseg);
242 nfs4_reset_write(task, data);
244 rpc_restart_call_prepare(task);
248 filelayout_set_layoutcommit(data);
252 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
253 static void prepare_to_resend_writes(struct nfs_write_data *data)
255 struct nfs_page *first = nfs_list_entry(data->pages.next);
257 data->task.tk_status = 0;
258 memcpy(data->verf.verifier, first->wb_verf.verifier,
259 sizeof(first->wb_verf.verifier));
260 data->verf.verifier[0]++; /* ensure verifier mismatch */
263 static int filelayout_commit_done_cb(struct rpc_task *task,
264 struct nfs_write_data *data)
268 if (filelayout_async_handle_error(task, data->args.context->state,
269 data->ds_clp, &reset) == -EAGAIN) {
270 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
271 __func__, data->ds_clp, data->ds_clp->cl_session);
273 prepare_to_resend_writes(data);
274 pnfs_set_lo_fail(data->lseg);
276 rpc_restart_call_prepare(task);
283 static void filelayout_write_prepare(struct rpc_task *task, void *data)
285 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
287 if (nfs41_setup_sequence(wdata->ds_clp->cl_session,
288 &wdata->args.seq_args, &wdata->res.seq_res,
292 rpc_call_start(task);
295 static void filelayout_write_call_done(struct rpc_task *task, void *data)
297 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
299 /* Note this may cause RPC to be resent */
300 wdata->mds_ops->rpc_call_done(task, data);
303 static void filelayout_write_count_stats(struct rpc_task *task, void *data)
305 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
307 rpc_count_iostats(task, NFS_SERVER(wdata->inode)->client->cl_metrics);
310 static void filelayout_write_release(void *data)
312 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
314 put_lseg(wdata->lseg);
315 wdata->mds_ops->rpc_release(data);
318 static void filelayout_commit_release(void *data)
320 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
322 nfs_commit_release_pages(wdata);
323 if (atomic_dec_and_test(&NFS_I(wdata->inode)->commits_outstanding))
324 nfs_commit_clear_lock(NFS_I(wdata->inode));
325 put_lseg(wdata->lseg);
326 nfs_commitdata_release(wdata);
329 static const struct rpc_call_ops filelayout_read_call_ops = {
330 .rpc_call_prepare = filelayout_read_prepare,
331 .rpc_call_done = filelayout_read_call_done,
332 .rpc_count_stats = filelayout_read_count_stats,
333 .rpc_release = filelayout_read_release,
336 static const struct rpc_call_ops filelayout_write_call_ops = {
337 .rpc_call_prepare = filelayout_write_prepare,
338 .rpc_call_done = filelayout_write_call_done,
339 .rpc_count_stats = filelayout_write_count_stats,
340 .rpc_release = filelayout_write_release,
343 static const struct rpc_call_ops filelayout_commit_call_ops = {
344 .rpc_call_prepare = filelayout_write_prepare,
345 .rpc_call_done = filelayout_write_call_done,
346 .rpc_count_stats = filelayout_write_count_stats,
347 .rpc_release = filelayout_commit_release,
350 static enum pnfs_try_status
351 filelayout_read_pagelist(struct nfs_read_data *data)
353 struct pnfs_layout_segment *lseg = data->lseg;
354 struct nfs4_pnfs_ds *ds;
355 loff_t offset = data->args.offset;
360 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
361 __func__, data->inode->i_ino,
362 data->args.pgbase, (size_t)data->args.count, offset);
364 if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags))
365 return PNFS_NOT_ATTEMPTED;
367 /* Retrieve the correct rpc_client for the byte range */
368 j = nfs4_fl_calc_j_index(lseg, offset);
369 idx = nfs4_fl_calc_ds_index(lseg, j);
370 ds = nfs4_fl_prepare_ds(lseg, idx);
372 /* Either layout fh index faulty, or ds connect failed */
373 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
374 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
375 return PNFS_NOT_ATTEMPTED;
377 dprintk("%s USE DS: %s\n", __func__, ds->ds_remotestr);
379 /* No multipath support. Use first DS */
380 data->ds_clp = ds->ds_clp;
381 fh = nfs4_fl_select_ds_fh(lseg, j);
385 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
386 data->mds_offset = offset;
388 /* Perform an asynchronous read to ds */
389 status = nfs_initiate_read(data, ds->ds_clp->cl_rpcclient,
390 &filelayout_read_call_ops);
392 return PNFS_ATTEMPTED;
395 /* Perform async writes. */
396 static enum pnfs_try_status
397 filelayout_write_pagelist(struct nfs_write_data *data, int sync)
399 struct pnfs_layout_segment *lseg = data->lseg;
400 struct nfs4_pnfs_ds *ds;
401 loff_t offset = data->args.offset;
406 if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags))
407 return PNFS_NOT_ATTEMPTED;
409 /* Retrieve the correct rpc_client for the byte range */
410 j = nfs4_fl_calc_j_index(lseg, offset);
411 idx = nfs4_fl_calc_ds_index(lseg, j);
412 ds = nfs4_fl_prepare_ds(lseg, idx);
414 printk(KERN_ERR "NFS: %s: prepare_ds failed, use MDS\n",
416 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
417 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
418 return PNFS_NOT_ATTEMPTED;
420 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s\n", __func__,
421 data->inode->i_ino, sync, (size_t) data->args.count, offset,
424 data->write_done_cb = filelayout_write_done_cb;
425 data->ds_clp = ds->ds_clp;
426 fh = nfs4_fl_select_ds_fh(lseg, j);
430 * Get the file offset on the dserver. Set the write offset to
431 * this offset and save the original offset.
433 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
435 /* Perform an asynchronous write */
436 status = nfs_initiate_write(data, ds->ds_clp->cl_rpcclient,
437 &filelayout_write_call_ops, sync);
439 return PNFS_ATTEMPTED;
443 * filelayout_check_layout()
445 * Make sure layout segment parameters are sane WRT the device.
446 * At this point no generic layer initialization of the lseg has occurred,
447 * and nothing has been added to the layout_hdr cache.
451 filelayout_check_layout(struct pnfs_layout_hdr *lo,
452 struct nfs4_filelayout_segment *fl,
453 struct nfs4_layoutget_res *lgr,
454 struct nfs4_deviceid *id,
457 struct nfs4_deviceid_node *d;
458 struct nfs4_file_layout_dsaddr *dsaddr;
459 int status = -EINVAL;
460 struct nfs_server *nfss = NFS_SERVER(lo->plh_inode);
462 dprintk("--> %s\n", __func__);
464 /* FIXME: remove this check when layout segment support is added */
465 if (lgr->range.offset != 0 ||
466 lgr->range.length != NFS4_MAX_UINT64) {
467 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
472 if (fl->pattern_offset > lgr->range.offset) {
473 dprintk("%s pattern_offset %lld too large\n",
474 __func__, fl->pattern_offset);
478 if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) {
479 dprintk("%s Invalid stripe unit (%u)\n",
480 __func__, fl->stripe_unit);
484 /* find and reference the deviceid */
485 d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld,
486 NFS_SERVER(lo->plh_inode)->nfs_client, id);
488 dsaddr = get_device_info(lo->plh_inode, id, gfp_flags);
492 dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
493 /* Found deviceid is being reaped */
494 if (test_bit(NFS_DEVICEID_INVALID, &dsaddr->id_node.flags))
499 if (fl->first_stripe_index >= dsaddr->stripe_count) {
500 dprintk("%s Bad first_stripe_index %u\n",
501 __func__, fl->first_stripe_index);
505 if ((fl->stripe_type == STRIPE_SPARSE &&
506 fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
507 (fl->stripe_type == STRIPE_DENSE &&
508 fl->num_fh != dsaddr->stripe_count)) {
509 dprintk("%s num_fh %u not valid for given packing\n",
510 __func__, fl->num_fh);
514 if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
515 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
516 "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
522 dprintk("--> %s returns %d\n", __func__, status);
525 nfs4_fl_put_deviceid(dsaddr);
529 static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
533 for (i = 0; i < fl->num_fh; i++) {
534 if (!fl->fh_array[i])
536 kfree(fl->fh_array[i]);
543 _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
545 filelayout_free_fh_array(fl);
550 filelayout_decode_layout(struct pnfs_layout_hdr *flo,
551 struct nfs4_filelayout_segment *fl,
552 struct nfs4_layoutget_res *lgr,
553 struct nfs4_deviceid *id,
556 struct xdr_stream stream;
558 struct page *scratch;
563 dprintk("%s: set_layout_map Begin\n", __func__);
565 scratch = alloc_page(gfp_flags);
569 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
570 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
572 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
574 p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
578 memcpy(id, p, sizeof(*id));
579 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
580 nfs4_print_deviceid(id);
582 nfl_util = be32_to_cpup(p++);
583 if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
584 fl->commit_through_mds = 1;
585 if (nfl_util & NFL4_UFLG_DENSE)
586 fl->stripe_type = STRIPE_DENSE;
588 fl->stripe_type = STRIPE_SPARSE;
589 fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
591 fl->first_stripe_index = be32_to_cpup(p++);
592 p = xdr_decode_hyper(p, &fl->pattern_offset);
593 fl->num_fh = be32_to_cpup(p++);
595 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
596 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
599 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
600 * Futher checking is done in filelayout_check_layout */
602 max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
605 if (fl->num_fh > 0) {
606 fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *),
612 for (i = 0; i < fl->num_fh; i++) {
613 /* Do we want to use a mempool here? */
614 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
615 if (!fl->fh_array[i])
618 p = xdr_inline_decode(&stream, 4);
621 fl->fh_array[i]->size = be32_to_cpup(p++);
622 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
623 printk(KERN_ERR "NFS: Too big fh %d received %d\n",
624 i, fl->fh_array[i]->size);
628 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
631 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
632 dprintk("DEBUG: %s: fh len %d\n", __func__,
633 fl->fh_array[i]->size);
636 __free_page(scratch);
640 filelayout_free_fh_array(fl);
642 __free_page(scratch);
647 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
649 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
651 dprintk("--> %s\n", __func__);
652 nfs4_fl_put_deviceid(fl->dsaddr);
653 kfree(fl->commit_buckets);
654 _filelayout_free_lseg(fl);
657 static struct pnfs_layout_segment *
658 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
659 struct nfs4_layoutget_res *lgr,
662 struct nfs4_filelayout_segment *fl;
664 struct nfs4_deviceid id;
666 dprintk("--> %s\n", __func__);
667 fl = kzalloc(sizeof(*fl), gfp_flags);
671 rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
672 if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
673 _filelayout_free_lseg(fl);
677 /* This assumes there is only one IOMODE_RW lseg. What
678 * we really want to do is have a layout_hdr level
679 * dictionary of <multipath_list4, fh> keys, each
680 * associated with a struct list_head, populated by calls
681 * to filelayout_write_pagelist().
683 if ((!fl->commit_through_mds) && (lgr->range.iomode == IOMODE_RW)) {
685 int size = (fl->stripe_type == STRIPE_SPARSE) ?
686 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
688 fl->commit_buckets = kcalloc(size, sizeof(struct nfs4_fl_commit_bucket), gfp_flags);
689 if (!fl->commit_buckets) {
690 filelayout_free_lseg(&fl->generic_hdr);
693 fl->number_of_buckets = size;
694 for (i = 0; i < size; i++) {
695 INIT_LIST_HEAD(&fl->commit_buckets[i].written);
696 INIT_LIST_HEAD(&fl->commit_buckets[i].committing);
699 return &fl->generic_hdr;
703 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
705 * return true : coalesce page
706 * return false : don't coalesce page
709 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
710 struct nfs_page *req)
712 u64 p_stripe, r_stripe;
715 if (!pnfs_generic_pg_test(pgio, prev, req) ||
716 !nfs_generic_pg_test(pgio, prev, req))
719 p_stripe = (u64)prev->wb_index << PAGE_CACHE_SHIFT;
720 r_stripe = (u64)req->wb_index << PAGE_CACHE_SHIFT;
721 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
723 do_div(p_stripe, stripe_unit);
724 do_div(r_stripe, stripe_unit);
726 return (p_stripe == r_stripe);
730 filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
731 struct nfs_page *req)
733 BUG_ON(pgio->pg_lseg != NULL);
735 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
741 /* If no lseg, fall back to read through mds */
742 if (pgio->pg_lseg == NULL)
743 nfs_pageio_reset_read_mds(pgio);
747 filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
748 struct nfs_page *req)
750 BUG_ON(pgio->pg_lseg != NULL);
752 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
758 /* If no lseg, fall back to write through mds */
759 if (pgio->pg_lseg == NULL)
760 nfs_pageio_reset_write_mds(pgio);
763 static const struct nfs_pageio_ops filelayout_pg_read_ops = {
764 .pg_init = filelayout_pg_init_read,
765 .pg_test = filelayout_pg_test,
766 .pg_doio = pnfs_generic_pg_readpages,
769 static const struct nfs_pageio_ops filelayout_pg_write_ops = {
770 .pg_init = filelayout_pg_init_write,
771 .pg_test = filelayout_pg_test,
772 .pg_doio = pnfs_generic_pg_writepages,
775 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
777 if (fl->stripe_type == STRIPE_SPARSE)
778 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
783 /* The generic layer is about to remove the req from the commit list.
784 * If this will make the bucket empty, it will need to put the lseg reference.
787 filelayout_clear_request_commit(struct nfs_page *req)
789 struct pnfs_layout_segment *freeme = NULL;
790 struct inode *inode = req->wb_context->dentry->d_inode;
792 spin_lock(&inode->i_lock);
793 if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
795 if (list_is_singular(&req->wb_list)) {
796 struct pnfs_layout_segment *lseg;
798 /* From here we can find the bucket, but for the moment,
799 * since there is only one relevant lseg...
801 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
802 if (lseg->pls_range.iomode == IOMODE_RW) {
809 nfs_request_remove_commit_list(req);
810 spin_unlock(&inode->i_lock);
814 static struct list_head *
815 filelayout_choose_commit_list(struct nfs_page *req,
816 struct pnfs_layout_segment *lseg)
818 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
820 struct list_head *list;
822 if (fl->commit_through_mds)
823 return &NFS_I(req->wb_context->dentry->d_inode)->commit_list;
825 /* Note that we are calling nfs4_fl_calc_j_index on each page
826 * that ends up being committed to a data server. An attractive
827 * alternative is to add a field to nfs_write_data and nfs_page
828 * to store the value calculated in filelayout_write_pagelist
829 * and just use that here.
831 j = nfs4_fl_calc_j_index(lseg,
832 (loff_t)req->wb_index << PAGE_CACHE_SHIFT);
833 i = select_bucket_index(fl, j);
834 list = &fl->commit_buckets[i].written;
835 if (list_empty(list)) {
836 /* Non-empty buckets hold a reference on the lseg. That ref
837 * is normally transferred to the COMMIT call and released
838 * there. It could also be released if the last req is pulled
839 * off due to a rewrite, in which case it will be done in
840 * filelayout_remove_commit_req
844 set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
849 filelayout_mark_request_commit(struct nfs_page *req,
850 struct pnfs_layout_segment *lseg)
852 struct list_head *list;
854 list = filelayout_choose_commit_list(req, lseg);
855 nfs_request_add_commit_list(req, list);
858 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
860 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
862 if (flseg->stripe_type == STRIPE_SPARSE)
865 return nfs4_fl_calc_ds_index(lseg, i);
868 static struct nfs_fh *
869 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
871 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
873 if (flseg->stripe_type == STRIPE_SPARSE) {
874 if (flseg->num_fh == 1)
876 else if (flseg->num_fh == 0)
877 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
880 return flseg->fh_array[i];
883 static int filelayout_initiate_commit(struct nfs_write_data *data, int how)
885 struct pnfs_layout_segment *lseg = data->lseg;
886 struct nfs4_pnfs_ds *ds;
890 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
891 ds = nfs4_fl_prepare_ds(lseg, idx);
893 printk(KERN_ERR "NFS: %s: prepare_ds failed, use MDS\n",
895 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
896 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
897 prepare_to_resend_writes(data);
898 filelayout_commit_release(data);
901 dprintk("%s ino %lu, how %d\n", __func__, data->inode->i_ino, how);
902 data->write_done_cb = filelayout_commit_done_cb;
903 data->ds_clp = ds->ds_clp;
904 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
907 return nfs_initiate_commit(data, ds->ds_clp->cl_rpcclient,
908 &filelayout_commit_call_ops, how);
912 * This is only useful while we are using whole file layouts.
914 static struct pnfs_layout_segment *
915 find_only_write_lseg_locked(struct inode *inode)
917 struct pnfs_layout_segment *lseg;
919 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
920 if (lseg->pls_range.iomode == IOMODE_RW)
925 static struct pnfs_layout_segment *find_only_write_lseg(struct inode *inode)
927 struct pnfs_layout_segment *rv;
929 spin_lock(&inode->i_lock);
930 rv = find_only_write_lseg_locked(inode);
933 spin_unlock(&inode->i_lock);
938 filelayout_scan_ds_commit_list(struct nfs4_fl_commit_bucket *bucket, int max,
941 struct list_head *src = &bucket->written;
942 struct list_head *dst = &bucket->committing;
943 struct nfs_page *req, *tmp;
946 list_for_each_entry_safe(req, tmp, src, wb_list) {
947 if (!nfs_lock_request(req))
949 if (cond_resched_lock(lock))
950 list_safe_reset_next(req, tmp, wb_list);
951 nfs_request_remove_commit_list(req);
952 clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
953 nfs_list_add_request(req, dst);
961 /* Move reqs from written to committing lists, returning count of number moved.
962 * Note called with i_lock held.
964 static int filelayout_scan_commit_lists(struct inode *inode, int max,
967 struct pnfs_layout_segment *lseg;
968 struct nfs4_filelayout_segment *fl;
971 lseg = find_only_write_lseg_locked(inode);
974 fl = FILELAYOUT_LSEG(lseg);
975 if (fl->commit_through_mds)
977 for (i = 0; i < fl->number_of_buckets && max != 0; i++) {
978 cnt = filelayout_scan_ds_commit_list(&fl->commit_buckets[i],
988 alloc_ds_commits(struct inode *inode, struct list_head *list)
990 struct pnfs_layout_segment *lseg;
991 struct nfs4_filelayout_segment *fl;
992 struct nfs_write_data *data;
994 unsigned int nreq = 0;
996 /* Won't need this when non-whole file layout segments are supported
997 * instead we will use a pnfs_layout_hdr structure */
998 lseg = find_only_write_lseg(inode);
1001 fl = FILELAYOUT_LSEG(lseg);
1002 for (i = 0; i < fl->number_of_buckets; i++) {
1003 if (list_empty(&fl->commit_buckets[i].committing))
1005 data = nfs_commitdata_alloc();
1008 data->ds_commit_index = i;
1010 list_add(&data->pages, list);
1014 /* Clean up on error */
1015 for (j = i; j < fl->number_of_buckets; j++) {
1016 if (list_empty(&fl->commit_buckets[i].committing))
1018 nfs_retry_commit(&fl->commit_buckets[i].committing, lseg);
1019 put_lseg(lseg); /* associated with emptying bucket */
1022 /* Caller will clean up entries put on list */
1026 /* This follows nfs_commit_list pretty closely */
1028 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1031 struct nfs_write_data *data, *tmp;
1033 unsigned int nreq = 0;
1035 if (!list_empty(mds_pages)) {
1036 data = nfs_commitdata_alloc();
1039 list_add(&data->pages, &list);
1042 nfs_retry_commit(mds_pages, NULL);
1045 nreq += alloc_ds_commits(inode, &list);
1048 nfs_commit_clear_lock(NFS_I(inode));
1052 atomic_add(nreq, &NFS_I(inode)->commits_outstanding);
1054 list_for_each_entry_safe(data, tmp, &list, pages) {
1055 list_del_init(&data->pages);
1057 nfs_init_commit(data, mds_pages, NULL);
1058 nfs_initiate_commit(data, NFS_CLIENT(inode),
1059 data->mds_ops, how);
1061 nfs_init_commit(data, &FILELAYOUT_LSEG(data->lseg)->commit_buckets[data->ds_commit_index].committing, data->lseg);
1062 filelayout_initiate_commit(data, how);
1066 return PNFS_ATTEMPTED;
1070 filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
1072 nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1075 static struct pnfs_layoutdriver_type filelayout_type = {
1076 .id = LAYOUT_NFSV4_1_FILES,
1077 .name = "LAYOUT_NFSV4_1_FILES",
1078 .owner = THIS_MODULE,
1079 .alloc_lseg = filelayout_alloc_lseg,
1080 .free_lseg = filelayout_free_lseg,
1081 .pg_read_ops = &filelayout_pg_read_ops,
1082 .pg_write_ops = &filelayout_pg_write_ops,
1083 .mark_request_commit = filelayout_mark_request_commit,
1084 .clear_request_commit = filelayout_clear_request_commit,
1085 .scan_commit_lists = filelayout_scan_commit_lists,
1086 .commit_pagelist = filelayout_commit_pagelist,
1087 .read_pagelist = filelayout_read_pagelist,
1088 .write_pagelist = filelayout_write_pagelist,
1089 .free_deviceid_node = filelayout_free_deveiceid_node,
1092 static int __init nfs4filelayout_init(void)
1094 printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1096 return pnfs_register_layoutdriver(&filelayout_type);
1099 static void __exit nfs4filelayout_exit(void)
1101 printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1103 pnfs_unregister_layoutdriver(&filelayout_type);
1106 MODULE_ALIAS("nfs-layouttype4-1");
1108 module_init(nfs4filelayout_init);
1109 module_exit(nfs4filelayout_exit);