pnfs_nfs.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Common NFS I/O  operations for the pnfs file based
   4 * layout drivers.
   5 *
   6 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   7 *
   8 * Tom Haynes <loghyr@primarydata.com>
   9 */
  10
  11#include <linux/nfs_fs.h>
  12#include <linux/nfs_page.h>
  13#include <linux/sunrpc/addr.h>
  14#include <linux/module.h>
  15
  16#include "nfs4session.h"
  17#include "internal.h"
  18#include "pnfs.h"
  19#include "netns.h"
  20#include "nfs4trace.h"
  21
  22#define NFSDBG_FACILITY		NFSDBG_PNFS
  23
  24void pnfs_generic_rw_release(void *data)
  25{
  26	struct nfs_pgio_header *hdr = data;
  27
  28	nfs_put_client(hdr->ds_clp);
  29	hdr->mds_ops->rpc_release(data);
  30}
  31EXPORT_SYMBOL_GPL(pnfs_generic_rw_release);
  32
  33/* Fake up some data that will cause nfs_commit_release to retry the writes. */
  34void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data)
  35{
  36	struct nfs_writeverf *verf = data->res.verf;
  37
  38	data->task.tk_status = 0;
  39	memset(&verf->verifier, 0, sizeof(verf->verifier));
  40	verf->committed = NFS_UNSTABLE;
  41}
  42EXPORT_SYMBOL_GPL(pnfs_generic_prepare_to_resend_writes);
  43
  44void pnfs_generic_write_commit_done(struct rpc_task *task, void *data)
  45{
  46	struct nfs_commit_data *wdata = data;
  47
  48	/* Note this may cause RPC to be resent */
  49	wdata->mds_ops->rpc_call_done(task, data);
  50}
  51EXPORT_SYMBOL_GPL(pnfs_generic_write_commit_done);
  52
  53void pnfs_generic_commit_release(void *calldata)
  54{
  55	struct nfs_commit_data *data = calldata;
  56
  57	data->completion_ops->completion(data);
  58	pnfs_put_lseg(data->lseg);
  59	nfs_put_client(data->ds_clp);
  60	nfs_commitdata_release(data);
  61}
  62EXPORT_SYMBOL_GPL(pnfs_generic_commit_release);
  63
  64static struct pnfs_layout_segment *
  65pnfs_free_bucket_lseg(struct pnfs_commit_bucket *bucket)
  66{
  67	if (list_empty(&bucket->committing) && list_empty(&bucket->written)) {
  68		struct pnfs_layout_segment *freeme = bucket->lseg;
  69		bucket->lseg = NULL;
  70		return freeme;
  71	}
  72	return NULL;
  73}
  74
  75/* The generic layer is about to remove the req from the commit list.
  76 * If this will make the bucket empty, it will need to put the lseg reference.
  77 * Note this must be called holding nfsi->commit_mutex
  78 */
  79void
  80pnfs_generic_clear_request_commit(struct nfs_page *req,
  81				  struct nfs_commit_info *cinfo)
  82{
  83	struct pnfs_commit_bucket *bucket = NULL;
  84
  85	if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
  86		goto out;
  87	cinfo->ds->nwritten--;
  88	if (list_is_singular(&req->wb_list))
  89		bucket = list_first_entry(&req->wb_list,
  90					  struct pnfs_commit_bucket, written);
  91out:
  92	nfs_request_remove_commit_list(req, cinfo);
  93	if (bucket)
  94		pnfs_put_lseg(pnfs_free_bucket_lseg(bucket));
  95}
  96EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
  97
  98struct pnfs_commit_array *
  99pnfs_alloc_commit_array(size_t n, gfp_t gfp_flags)
 100{
 101	struct pnfs_commit_array *p;
 102	struct pnfs_commit_bucket *b;
 103
 104	p = kmalloc(struct_size(p, buckets, n), gfp_flags);
 105	if (!p)
 106		return NULL;
 107	p->nbuckets = n;
 108	INIT_LIST_HEAD(&p->cinfo_list);
 109	INIT_LIST_HEAD(&p->lseg_list);
 110	p->lseg = NULL;
 111	for (b = &p->buckets[0]; n != 0; b++, n--) {
 112		INIT_LIST_HEAD(&b->written);
 113		INIT_LIST_HEAD(&b->committing);
 114		b->lseg = NULL;
 115		b->direct_verf.committed = NFS_INVALID_STABLE_HOW;
 116	}
 117	return p;
 118}
 119EXPORT_SYMBOL_GPL(pnfs_alloc_commit_array);
 120
 121void
 122pnfs_free_commit_array(struct pnfs_commit_array *p)
 123{
 124	kfree_rcu(p, rcu);
 125}
 126EXPORT_SYMBOL_GPL(pnfs_free_commit_array);
 127
 128static struct pnfs_commit_array *
 129pnfs_find_commit_array_by_lseg(struct pnfs_ds_commit_info *fl_cinfo,
 130		struct pnfs_layout_segment *lseg)
 131{
 132	struct pnfs_commit_array *array;
 133
 134	list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 135		if (array->lseg == lseg)
 136			return array;
 137	}
 138	return NULL;
 139}
 140
 141struct pnfs_commit_array *
 142pnfs_add_commit_array(struct pnfs_ds_commit_info *fl_cinfo,
 143		struct pnfs_commit_array *new,
 144		struct pnfs_layout_segment *lseg)
 145{
 146	struct pnfs_commit_array *array;
 147
 148	array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
 149	if (array)
 150		return array;
 151	new->lseg = lseg;
 152	refcount_set(&new->refcount, 1);
 153	list_add_rcu(&new->cinfo_list, &fl_cinfo->commits);
 154	list_add(&new->lseg_list, &lseg->pls_commits);
 155	return new;
 156}
 157EXPORT_SYMBOL_GPL(pnfs_add_commit_array);
 158
 159static struct pnfs_commit_array *
 160pnfs_lookup_commit_array(struct pnfs_ds_commit_info *fl_cinfo,
 161		struct pnfs_layout_segment *lseg)
 162{
 163	struct pnfs_commit_array *array;
 164
 165	rcu_read_lock();
 166	array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
 167	if (!array) {
 168		rcu_read_unlock();
 169		fl_cinfo->ops->setup_ds_info(fl_cinfo, lseg);
 170		rcu_read_lock();
 171		array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
 172	}
 173	rcu_read_unlock();
 174	return array;
 175}
 176
 177static void
 178pnfs_release_commit_array_locked(struct pnfs_commit_array *array)
 179{
 180	list_del_rcu(&array->cinfo_list);
 181	list_del(&array->lseg_list);
 182	pnfs_free_commit_array(array);
 183}
 184
 185static void
 186pnfs_put_commit_array_locked(struct pnfs_commit_array *array)
 187{
 188	if (refcount_dec_and_test(&array->refcount))
 189		pnfs_release_commit_array_locked(array);
 190}
 191
 192static void
 193pnfs_put_commit_array(struct pnfs_commit_array *array, struct inode *inode)
 194{
 195	if (refcount_dec_and_lock(&array->refcount, &inode->i_lock)) {
 196		pnfs_release_commit_array_locked(array);
 197		spin_unlock(&inode->i_lock);
 198	}
 199}
 200
 201static struct pnfs_commit_array *
 202pnfs_get_commit_array(struct pnfs_commit_array *array)
 203{
 204	if (refcount_inc_not_zero(&array->refcount))
 205		return array;
 206	return NULL;
 207}
 208
 209static void
 210pnfs_remove_and_free_commit_array(struct pnfs_commit_array *array)
 211{
 212	array->lseg = NULL;
 213	list_del_init(&array->lseg_list);
 214	pnfs_put_commit_array_locked(array);
 215}
 216
 217void
 218pnfs_generic_ds_cinfo_release_lseg(struct pnfs_ds_commit_info *fl_cinfo,
 219		struct pnfs_layout_segment *lseg)
 220{
 221	struct pnfs_commit_array *array, *tmp;
 222
 223	list_for_each_entry_safe(array, tmp, &lseg->pls_commits, lseg_list)
 224		pnfs_remove_and_free_commit_array(array);
 225}
 226EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_release_lseg);
 227
 228void
 229pnfs_generic_ds_cinfo_destroy(struct pnfs_ds_commit_info *fl_cinfo)
 230{
 231	struct pnfs_commit_array *array, *tmp;
 232
 233	list_for_each_entry_safe(array, tmp, &fl_cinfo->commits, cinfo_list)
 234		pnfs_remove_and_free_commit_array(array);
 235}
 236EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_destroy);
 237
 238/*
 239 * Locks the nfs_page requests for commit and moves them to
 240 * @bucket->committing.
 241 */
 242static int
 243pnfs_bucket_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
 244				struct nfs_commit_info *cinfo,
 245				int max)
 246{
 247	struct list_head *src = &bucket->written;
 248	struct list_head *dst = &bucket->committing;
 249	int ret;
 250
 251	lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
 252	ret = nfs_scan_commit_list(src, dst, cinfo, max);
 253	if (ret) {
 254		cinfo->ds->nwritten -= ret;
 255		cinfo->ds->ncommitting += ret;
 256	}
 257	return ret;
 258}
 259
 260static int pnfs_bucket_scan_array(struct nfs_commit_info *cinfo,
 261				  struct pnfs_commit_bucket *buckets,
 262				  unsigned int nbuckets,
 263				  int max)
 264{
 265	unsigned int i;
 266	int rv = 0, cnt;
 267
 268	for (i = 0; i < nbuckets && max != 0; i++) {
 269		cnt = pnfs_bucket_scan_ds_commit_list(&buckets[i], cinfo, max);
 270		rv += cnt;
 271		max -= cnt;
 272	}
 273	return rv;
 274}
 275
 276/* Move reqs from written to committing lists, returning count
 277 * of number moved.
 278 */
 279int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo, int max)
 280{
 281	struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 282	struct pnfs_commit_array *array;
 283	int rv = 0, cnt;
 284
 285	rcu_read_lock();
 286	list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 287		if (!array->lseg || !pnfs_get_commit_array(array))
 288			continue;
 289		rcu_read_unlock();
 290		cnt = pnfs_bucket_scan_array(cinfo, array->buckets,
 291				array->nbuckets, max);
 292		rcu_read_lock();
 293		pnfs_put_commit_array(array, cinfo->inode);
 294		rv += cnt;
 295		max -= cnt;
 296		if (!max)
 297			break;
 298	}
 299	rcu_read_unlock();
 300	return rv;
 301}
 302EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
 303
 304static unsigned int
 305pnfs_bucket_recover_commit_reqs(struct list_head *dst,
 306			        struct pnfs_commit_bucket *buckets,
 307				unsigned int nbuckets,
 308				struct nfs_commit_info *cinfo)
 309{
 310	struct pnfs_commit_bucket *b;
 311	struct pnfs_layout_segment *freeme;
 312	unsigned int nwritten, ret = 0;
 313	unsigned int i;
 314
 315restart:
 316	for (i = 0, b = buckets; i < nbuckets; i++, b++) {
 317		nwritten = nfs_scan_commit_list(&b->written, dst, cinfo, 0);
 318		if (!nwritten)
 319			continue;
 320		ret += nwritten;
 321		freeme = pnfs_free_bucket_lseg(b);
 322		if (freeme) {
 323			pnfs_put_lseg(freeme);
 324			goto restart;
 325		}
 326	}
 327	return ret;
 328}
 329
 330/* Pull everything off the committing lists and dump into @dst.  */
 331void pnfs_generic_recover_commit_reqs(struct list_head *dst,
 332				      struct nfs_commit_info *cinfo)
 333{
 334	struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 335	struct pnfs_commit_array *array;
 336	unsigned int nwritten;
 337
 338	lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
 339	rcu_read_lock();
 340	list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 341		if (!array->lseg || !pnfs_get_commit_array(array))
 342			continue;
 343		rcu_read_unlock();
 344		nwritten = pnfs_bucket_recover_commit_reqs(dst,
 345							   array->buckets,
 346							   array->nbuckets,
 347							   cinfo);
 348		rcu_read_lock();
 349		pnfs_put_commit_array(array, cinfo->inode);
 350		fl_cinfo->nwritten -= nwritten;
 351	}
 352	rcu_read_unlock();
 353}
 354EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
 355
 356static struct pnfs_layout_segment *
 357pnfs_bucket_get_committing(struct list_head *head,
 358			   struct pnfs_commit_bucket *bucket,
 359			   struct nfs_commit_info *cinfo)
 360{
 361	struct pnfs_layout_segment *lseg;
 362	struct list_head *pos;
 363
 364	list_for_each(pos, &bucket->committing)
 365		cinfo->ds->ncommitting--;
 366	list_splice_init(&bucket->committing, head);
 367	lseg = pnfs_free_bucket_lseg(bucket);
 368	if (!lseg)
 369		lseg = pnfs_get_lseg(bucket->lseg);
 370	return lseg;
 371}
 372
 373static struct nfs_commit_data *
 374pnfs_bucket_fetch_commitdata(struct pnfs_commit_bucket *bucket,
 375			     struct nfs_commit_info *cinfo)
 376{
 377	struct nfs_commit_data *data = nfs_commitdata_alloc();
 378
 379	if (!data)
 380		return NULL;
 381	data->lseg = pnfs_bucket_get_committing(&data->pages, bucket, cinfo);
 382	return data;
 383}
 384
 385static void pnfs_generic_retry_commit(struct pnfs_commit_bucket *buckets,
 386				      unsigned int nbuckets,
 387				      struct nfs_commit_info *cinfo,
 388				      unsigned int idx)
 389{
 390	struct pnfs_commit_bucket *bucket;
 391	struct pnfs_layout_segment *freeme;
 392	LIST_HEAD(pages);
 393
 394	for (bucket = buckets; idx < nbuckets; bucket++, idx++) {
 395		if (list_empty(&bucket->committing))
 396			continue;
 397		mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 398		freeme = pnfs_bucket_get_committing(&pages, bucket, cinfo);
 399		mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 400		nfs_retry_commit(&pages, freeme, cinfo, idx);
 401		pnfs_put_lseg(freeme);
 402	}
 403}
 404
 405static unsigned int
 406pnfs_bucket_alloc_ds_commits(struct list_head *list,
 407			     struct pnfs_commit_bucket *buckets,
 408			     unsigned int nbuckets,
 409			     struct nfs_commit_info *cinfo)
 410{
 411	struct pnfs_commit_bucket *bucket;
 412	struct nfs_commit_data *data;
 413	unsigned int i;
 414	unsigned int nreq = 0;
 415
 416	for (i = 0, bucket = buckets; i < nbuckets; i++, bucket++) {
 417		if (list_empty(&bucket->committing))
 418			continue;
 419		mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 420		if (!list_empty(&bucket->committing)) {
 421			data = pnfs_bucket_fetch_commitdata(bucket, cinfo);
 422			if (!data)
 423				goto out_error;
 424			data->ds_commit_index = i;
 425			list_add_tail(&data->list, list);
 426			nreq++;
 427		}
 428		mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 429	}
 430	return nreq;
 431out_error:
 432	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 433	/* Clean up on error */
 434	pnfs_generic_retry_commit(buckets, nbuckets, cinfo, i);
 435	return nreq;
 436}
 437
 438static unsigned int
 439pnfs_alloc_ds_commits_list(struct list_head *list,
 440			   struct pnfs_ds_commit_info *fl_cinfo,
 441			   struct nfs_commit_info *cinfo)
 442{
 443	struct pnfs_commit_array *array;
 444	unsigned int ret = 0;
 445
 446	rcu_read_lock();
 447	list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 448		if (!array->lseg || !pnfs_get_commit_array(array))
 449			continue;
 450		rcu_read_unlock();
 451		ret += pnfs_bucket_alloc_ds_commits(list, array->buckets,
 452				array->nbuckets, cinfo);
 453		rcu_read_lock();
 454		pnfs_put_commit_array(array, cinfo->inode);
 455	}
 456	rcu_read_unlock();
 457	return ret;
 458}
 459
 460/* This follows nfs_commit_list pretty closely */
 461int
 462pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
 463			     int how, struct nfs_commit_info *cinfo,
 464			     int (*initiate_commit)(struct nfs_commit_data *data,
 465						    int how))
 466{
 467	struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 468	struct nfs_commit_data *data, *tmp;
 469	LIST_HEAD(list);
 470	unsigned int nreq = 0;
 471
 472	if (!list_empty(mds_pages)) {
 473		data = nfs_commitdata_alloc();
 474		if (!data) {
 475			nfs_retry_commit(mds_pages, NULL, cinfo, -1);
 476			return -ENOMEM;
 477		}
 478		data->ds_commit_index = -1;
 479		list_splice_init(mds_pages, &data->pages);
 480		list_add_tail(&data->list, &list);
 481		nreq++;
 482	}
 483
 484	nreq += pnfs_alloc_ds_commits_list(&list, fl_cinfo, cinfo);
 485	if (nreq == 0)
 486		goto out;
 487
 488	list_for_each_entry_safe(data, tmp, &list, list) {
 489		list_del(&data->list);
 490		if (data->ds_commit_index < 0) {
 491			nfs_init_commit(data, NULL, NULL, cinfo);
 492			nfs_initiate_commit(NFS_CLIENT(inode), data,
 493					    NFS_PROTO(data->inode),
 494					    data->mds_ops, how,
 495					    RPC_TASK_CRED_NOREF, NULL);
 496		} else {
 497			nfs_init_commit(data, NULL, data->lseg, cinfo);
 498			initiate_commit(data, how);
 499		}
 500	}
 501out:
 502	return PNFS_ATTEMPTED;
 503}
 504EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
 505
 506/*
 507 * Data server cache
 508 *
 509 * Data servers can be mapped to different device ids, but should
 510 * never be shared between net namespaces.
 511 *
 512 * nfs4_pnfs_ds reference counting:
 513 *   - set to 1 on allocation
 514 *   - incremented when a device id maps a data server already in the cache.
 515 *   - decremented when deviceid is removed from the cache.
 516 */
 517
 518/* Debug routines */
 519static void
 520print_ds(struct nfs4_pnfs_ds *ds)
 521{
 522	if (ds == NULL) {
 523		printk(KERN_WARNING "%s NULL device\n", __func__);
 524		return;
 525	}
 526	printk(KERN_WARNING "        ds %s\n"
 527		"        ref count %d\n"
 528		"        client %p\n"
 529		"        cl_exchange_flags %x\n",
 530		ds->ds_remotestr,
 531		refcount_read(&ds->ds_count), ds->ds_clp,
 532		ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
 533}
 534
 535static bool
 536same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
 537{
 538	struct sockaddr_in *a, *b;
 539	struct sockaddr_in6 *a6, *b6;
 540
 541	if (addr1->sa_family != addr2->sa_family)
 542		return false;
 543
 544	switch (addr1->sa_family) {
 545	case AF_INET:
 546		a = (struct sockaddr_in *)addr1;
 547		b = (struct sockaddr_in *)addr2;
 548
 549		if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
 550		    a->sin_port == b->sin_port)
 551			return true;
 552		break;
 553
 554	case AF_INET6:
 555		a6 = (struct sockaddr_in6 *)addr1;
 556		b6 = (struct sockaddr_in6 *)addr2;
 557
 558		/* LINKLOCAL addresses must have matching scope_id */
 559		if (ipv6_addr_src_scope(&a6->sin6_addr) ==
 560		    IPV6_ADDR_SCOPE_LINKLOCAL &&
 561		    a6->sin6_scope_id != b6->sin6_scope_id)
 562			return false;
 563
 564		if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
 565		    a6->sin6_port == b6->sin6_port)
 566			return true;
 567		break;
 568
 569	default:
 570		dprintk("%s: unhandled address family: %u\n",
 571			__func__, addr1->sa_family);
 572		return false;
 573	}
 574
 575	return false;
 576}
 577
 578/*
 579 * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
 580 * declare a match.
 581 */
 582static bool
 583_same_data_server_addrs_locked(const struct list_head *dsaddrs1,
 584			       const struct list_head *dsaddrs2)
 585{
 586	struct nfs4_pnfs_ds_addr *da1, *da2;
 587	struct sockaddr *sa1, *sa2;
 588	bool match = false;
 589
 590	list_for_each_entry(da1, dsaddrs1, da_node) {
 591		sa1 = (struct sockaddr *)&da1->da_addr;
 592		match = false;
 593		list_for_each_entry(da2, dsaddrs2, da_node) {
 594			sa2 = (struct sockaddr *)&da2->da_addr;
 595			match = same_sockaddr(sa1, sa2);
 596			if (match)
 597				break;
 598		}
 599		if (!match)
 600			break;
 601	}
 602	return match;
 603}
 604
 605/*
 606 * Lookup DS by addresses.  nfs4_ds_cache_lock is held
 607 */
 608static struct nfs4_pnfs_ds *
 609_data_server_lookup_locked(const struct nfs_net *nn, const struct list_head *dsaddrs)
 610{
 611	struct nfs4_pnfs_ds *ds;
 612
 613	list_for_each_entry(ds, &nn->nfs4_data_server_cache, ds_node)
 614		if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
 615			return ds;
 616	return NULL;
 617}
 618
 619static struct nfs4_pnfs_ds_addr *nfs4_pnfs_ds_addr_alloc(gfp_t gfp_flags)
 620{
 621	struct nfs4_pnfs_ds_addr *da = kzalloc(sizeof(*da), gfp_flags);
 622	if (da)
 623		INIT_LIST_HEAD(&da->da_node);
 624	return da;
 625}
 626
 627static void nfs4_pnfs_ds_addr_free(struct nfs4_pnfs_ds_addr *da)
 628{
 629	kfree(da->da_remotestr);
 630	kfree(da->da_netid);
 631	kfree(da);
 632}
 633
 634static void destroy_ds(struct nfs4_pnfs_ds *ds)
 635{
 636	struct nfs4_pnfs_ds_addr *da;
 637
 638	dprintk("--> %s\n", __func__);
 639	ifdebug(FACILITY)
 640		print_ds(ds);
 641
 642	nfs_put_client(ds->ds_clp);
 643
 644	while (!list_empty(&ds->ds_addrs)) {
 645		da = list_first_entry(&ds->ds_addrs,
 646				      struct nfs4_pnfs_ds_addr,
 647				      da_node);
 648		list_del_init(&da->da_node);
 649		nfs4_pnfs_ds_addr_free(da);
 650	}
 651
 652	kfree(ds->ds_remotestr);
 653	kfree(ds);
 654}
 655
 656void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
 657{
 658	struct nfs_net *nn = net_generic(ds->ds_net, nfs_net_id);
 659
 660	if (refcount_dec_and_lock(&ds->ds_count, &nn->nfs4_data_server_lock)) {
 661		list_del_init(&ds->ds_node);
 662		spin_unlock(&nn->nfs4_data_server_lock);
 663		destroy_ds(ds);
 664	}
 665}
 666EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
 667
 668/*
 669 * Create a string with a human readable address and port to avoid
 670 * complicated setup around many dprinks.
 671 */
 672static char *
 673nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
 674{
 675	struct nfs4_pnfs_ds_addr *da;
 676	char *remotestr;
 677	size_t len;
 678	char *p;
 679
 680	len = 3;        /* '{', '}' and eol */
 681	list_for_each_entry(da, dsaddrs, da_node) {
 682		len += strlen(da->da_remotestr) + 1;    /* string plus comma */
 683	}
 684
 685	remotestr = kzalloc(len, gfp_flags);
 686	if (!remotestr)
 687		return NULL;
 688
 689	p = remotestr;
 690	*(p++) = '{';
 691	len--;
 692	list_for_each_entry(da, dsaddrs, da_node) {
 693		size_t ll = strlen(da->da_remotestr);
 694
 695		if (ll > len)
 696			goto out_err;
 697
 698		memcpy(p, da->da_remotestr, ll);
 699		p += ll;
 700		len -= ll;
 701
 702		if (len < 1)
 703			goto out_err;
 704		(*p++) = ',';
 705		len--;
 706	}
 707	if (len < 2)
 708		goto out_err;
 709	*(p++) = '}';
 710	*p = '\0';
 711	return remotestr;
 712out_err:
 713	kfree(remotestr);
 714	return NULL;
 715}
 716
 717/*
 718 * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
 719 * uncached and return cached struct nfs4_pnfs_ds.
 720 */
 721struct nfs4_pnfs_ds *
 722nfs4_pnfs_ds_add(const struct net *net, struct list_head *dsaddrs, gfp_t gfp_flags)
 723{
 724	struct nfs_net *nn = net_generic(net, nfs_net_id);
 725	struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
 726	char *remotestr;
 727
 728	if (list_empty(dsaddrs)) {
 729		dprintk("%s: no addresses defined\n", __func__);
 730		goto out;
 731	}
 732
 733	ds = kzalloc(sizeof(*ds), gfp_flags);
 734	if (!ds)
 735		goto out;
 736
 737	/* this is only used for debugging, so it's ok if its NULL */
 738	remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
 739
 740	spin_lock(&nn->nfs4_data_server_lock);
 741	tmp_ds = _data_server_lookup_locked(nn, dsaddrs);
 742	if (tmp_ds == NULL) {
 743		INIT_LIST_HEAD(&ds->ds_addrs);
 744		list_splice_init(dsaddrs, &ds->ds_addrs);
 745		ds->ds_remotestr = remotestr;
 746		refcount_set(&ds->ds_count, 1);
 747		INIT_LIST_HEAD(&ds->ds_node);
 748		ds->ds_net = net;
 749		ds->ds_clp = NULL;
 750		list_add(&ds->ds_node, &nn->nfs4_data_server_cache);
 751		dprintk("%s add new data server %s\n", __func__,
 752			ds->ds_remotestr);
 753	} else {
 754		kfree(remotestr);
 755		kfree(ds);
 756		refcount_inc(&tmp_ds->ds_count);
 757		dprintk("%s data server %s found, inc'ed ds_count to %d\n",
 758			__func__, tmp_ds->ds_remotestr,
 759			refcount_read(&tmp_ds->ds_count));
 760		ds = tmp_ds;
 761	}
 762	spin_unlock(&nn->nfs4_data_server_lock);
 763out:
 764	return ds;
 765}
 766EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
 767
 768static int nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
 769{
 770	might_sleep();
 771	return wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING, TASK_KILLABLE);
 772}
 773
 774static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
 775{
 776	smp_mb__before_atomic();
 777	clear_and_wake_up_bit(NFS4DS_CONNECTING, &ds->ds_state);
 778}
 779
 780static struct nfs_client *(*get_v3_ds_connect)(
 781			struct nfs_server *mds_srv,
 782			const struct sockaddr_storage *ds_addr,
 783			int ds_addrlen,
 784			int ds_proto,
 785			unsigned int ds_timeo,
 786			unsigned int ds_retrans);
 787
 788static bool load_v3_ds_connect(void)
 789{
 790	if (!get_v3_ds_connect) {
 791		get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
 792		WARN_ON_ONCE(!get_v3_ds_connect);
 793	}
 794
 795	return(get_v3_ds_connect != NULL);
 796}
 797
 798void nfs4_pnfs_v3_ds_connect_unload(void)
 799{
 800	if (get_v3_ds_connect) {
 801		symbol_put(nfs3_set_ds_client);
 802		get_v3_ds_connect = NULL;
 803	}
 804}
 805
 806static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
 807				 struct nfs4_pnfs_ds *ds,
 808				 unsigned int timeo,
 809				 unsigned int retrans)
 810{
 811	struct nfs_client *clp = ERR_PTR(-EIO);
 812	struct nfs_client *mds_clp = mds_srv->nfs_client;
 813	enum xprtsec_policies xprtsec_policy = mds_clp->cl_xprtsec.policy;
 814	struct nfs4_pnfs_ds_addr *da;
 815	unsigned long connect_timeout = timeo * (retrans + 1) * HZ / 10;
 816	int ds_proto;
 817	int status = 0;
 818
 819	dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
 820
 821	if (!load_v3_ds_connect())
 822		return -EPROTONOSUPPORT;
 823
 824	list_for_each_entry(da, &ds->ds_addrs, da_node) {
 825		dprintk("%s: DS %s: trying address %s\n",
 826			__func__, ds->ds_remotestr, da->da_remotestr);
 827
 828		if (!IS_ERR(clp)) {
 829			struct xprt_create xprt_args = {
 830				.ident = da->da_transport,
 831				.net = clp->cl_net,
 832				.dstaddr = (struct sockaddr *)&da->da_addr,
 833				.addrlen = da->da_addrlen,
 834				.servername = clp->cl_hostname,
 835				.connect_timeout = connect_timeout,
 836				.reconnect_timeout = connect_timeout,
 837				.xprtsec = clp->cl_xprtsec,
 838			};
 839
 840			if (xprt_args.ident == XPRT_TRANSPORT_TCP &&
 841			    clp->cl_proto == XPRT_TRANSPORT_TCP_TLS)
 842				xprt_args.ident = XPRT_TRANSPORT_TCP_TLS;
 843
 844			if (xprt_args.ident != clp->cl_proto)
 845				continue;
 846			if (xprt_args.dstaddr->sa_family !=
 847			    clp->cl_addr.ss_family)
 848				continue;
 849			/* Add this address as an alias */
 850			rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
 851					  rpc_clnt_test_and_add_xprt, NULL);
 852			continue;
 853		}
 854
 855		ds_proto = da->da_transport;
 856		if (ds_proto == XPRT_TRANSPORT_TCP &&
 857		    xprtsec_policy != RPC_XPRTSEC_NONE)
 858			ds_proto = XPRT_TRANSPORT_TCP_TLS;
 859
 860		clp = get_v3_ds_connect(mds_srv, &da->da_addr, da->da_addrlen,
 861					ds_proto, timeo, retrans);
 862		if (IS_ERR(clp))
 863			continue;
 864		clp->cl_rpcclient->cl_softerr = 0;
 865		clp->cl_rpcclient->cl_softrtry = 0;
 866	}
 867
 868	if (IS_ERR(clp)) {
 869		status = PTR_ERR(clp);
 870		goto out;
 871	}
 872
 873	smp_wmb();
 874	WRITE_ONCE(ds->ds_clp, clp);
 875	dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 876out:
 877	return status;
 878}
 879
 880static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
 881				 struct nfs4_pnfs_ds *ds,
 882				 unsigned int timeo,
 883				 unsigned int retrans,
 884				 u32 minor_version)
 885{
 886	struct nfs_client *clp = ERR_PTR(-EIO);
 887	struct nfs_client *mds_clp = mds_srv->nfs_client;
 888	enum xprtsec_policies xprtsec_policy = mds_clp->cl_xprtsec.policy;
 889	struct nfs4_pnfs_ds_addr *da;
 890	int ds_proto;
 891	int status = 0;
 892
 893	dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
 894
 895	list_for_each_entry(da, &ds->ds_addrs, da_node) {
 896		char servername[48];
 897
 898		dprintk("%s: DS %s: trying address %s\n",
 899			__func__, ds->ds_remotestr, da->da_remotestr);
 900
 901		if (!IS_ERR(clp) && clp->cl_mvops->session_trunk) {
 902			struct xprt_create xprt_args = {
 903				.ident = da->da_transport,
 904				.net = clp->cl_net,
 905				.dstaddr = (struct sockaddr *)&da->da_addr,
 906				.addrlen = da->da_addrlen,
 907				.servername = clp->cl_hostname,
 908				.xprtsec = clp->cl_xprtsec,
 909			};
 910			struct nfs4_add_xprt_data xprtdata = {
 911				.clp = clp,
 912			};
 913			struct rpc_add_xprt_test rpcdata = {
 914				.add_xprt_test = clp->cl_mvops->session_trunk,
 915				.data = &xprtdata,
 916			};
 917
 918			if (xprt_args.ident == XPRT_TRANSPORT_TCP &&
 919			    clp->cl_proto == XPRT_TRANSPORT_TCP_TLS) {
 920				struct sockaddr *addr =
 921					(struct sockaddr *)&da->da_addr;
 922				struct sockaddr_in *sin =
 923					(struct sockaddr_in *)&da->da_addr;
 924				struct sockaddr_in6 *sin6 =
 925					(struct sockaddr_in6 *)&da->da_addr;
 926
 927				/* for NFS with TLS we need to supply a correct
 928				 * servername of the trunked transport, not the
 929				 * servername of the main transport stored in
 930				 * clp->cl_hostname. And set the protocol to
 931				 * indicate to use TLS
 932				 */
 933				servername[0] = '\0';
 934				switch(addr->sa_family) {
 935				case AF_INET:
 936					snprintf(servername, sizeof(servername),
 937						"%pI4", &sin->sin_addr.s_addr);
 938					break;
 939				case AF_INET6:
 940					snprintf(servername, sizeof(servername),
 941						"%pI6", &sin6->sin6_addr);
 942					break;
 943				default:
 944					/* do not consider this address */
 945					continue;
 946				}
 947				xprt_args.ident = XPRT_TRANSPORT_TCP_TLS;
 948				xprt_args.servername = servername;
 949			}
 950			if (xprt_args.ident != clp->cl_proto)
 951				continue;
 952			if (xprt_args.dstaddr->sa_family !=
 953			    clp->cl_addr.ss_family)
 954				continue;
 955
 956			/**
 957			* Test this address for session trunking and
 958			* add as an alias
 959			*/
 960			xprtdata.cred = nfs4_get_clid_cred(clp);
 961			rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
 962					  rpc_clnt_setup_test_and_add_xprt,
 963					  &rpcdata);
 964			if (xprtdata.cred)
 965				put_cred(xprtdata.cred);
 966		} else {
 967			ds_proto = da->da_transport;
 968			if (ds_proto == XPRT_TRANSPORT_TCP &&
 969			    xprtsec_policy != RPC_XPRTSEC_NONE)
 970				ds_proto = XPRT_TRANSPORT_TCP_TLS;
 971
 972			clp = nfs4_set_ds_client(mds_srv, &da->da_addr,
 973						 da->da_addrlen, ds_proto,
 974						 timeo, retrans, minor_version);
 975			if (IS_ERR(clp))
 976				continue;
 977
 978			status = nfs4_init_ds_session(clp,
 979					mds_srv->nfs_client->cl_lease_time);
 980			if (status) {
 981				nfs_put_client(clp);
 982				clp = ERR_PTR(-EIO);
 983				continue;
 984			}
 985		}
 986	}
 987
 988	if (IS_ERR(clp)) {
 989		status = PTR_ERR(clp);
 990		goto out;
 991	}
 992
 993	smp_wmb();
 994	WRITE_ONCE(ds->ds_clp, clp);
 995	dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 996out:
 997	return status;
 998}
 999
1000/*
1001 * Create an rpc connection to the nfs4_pnfs_ds data server.
1002 * Currently only supports IPv4 and IPv6 addresses.
1003 * If connection fails, make devid unavailable and return a -errno.
1004 */
1005int nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
1006			  struct nfs4_deviceid_node *devid, unsigned int timeo,
1007			  unsigned int retrans, u32 version, u32 minor_version)
1008{
1009	int err;
1010
1011	do {
1012		err = nfs4_wait_ds_connect(ds);
1013		if (err || ds->ds_clp)
1014			goto out;
1015		if (nfs4_test_deviceid_unavailable(devid)) {
1016			err = -ENODEV;
1017			goto out;
1018		}
1019	} while (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) != 0);
1020
1021	if (ds->ds_clp)
1022		goto connect_done;
1023
1024	switch (version) {
1025	case 3:
1026		err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo, retrans);
1027		break;
1028	case 4:
1029		err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo, retrans,
1030					       minor_version);
1031		break;
1032	default:
1033		dprintk("%s: unsupported DS version %d\n", __func__, version);
1034		err = -EPROTONOSUPPORT;
1035	}
1036
1037connect_done:
1038	nfs4_clear_ds_conn_bit(ds);
1039out:
1040	/*
1041	 * At this point the ds->ds_clp should be ready, but it might have
1042	 * hit an error.
1043	 */
1044	if (!err) {
1045		if (!ds->ds_clp || !nfs_client_init_is_complete(ds->ds_clp)) {
1046			WARN_ON_ONCE(ds->ds_clp ||
1047				!nfs4_test_deviceid_unavailable(devid));
1048			err = -EINVAL;
1049		} else
1050			err = nfs_client_init_status(ds->ds_clp);
1051	}
1052
1053	trace_pnfs_ds_connect(ds->ds_remotestr, err);
1054	return err;
1055}
1056EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
1057
1058/*
1059 * Currently only supports ipv4, ipv6 and one multi-path address.
1060 */
1061struct nfs4_pnfs_ds_addr *
1062nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
1063{
1064	struct nfs4_pnfs_ds_addr *da = NULL;
1065	char *buf, *portstr;
1066	__be16 port;
1067	ssize_t nlen, rlen;
1068	int tmp[2];
1069	char *netid;
1070	size_t len;
1071	char *startsep = "";
1072	char *endsep = "";
1073
1074
1075	/* r_netid */
1076	nlen = xdr_stream_decode_string_dup(xdr, &netid, XDR_MAX_NETOBJ,
1077					    gfp_flags);
1078	if (unlikely(nlen < 0))
1079		goto out_err;
1080
1081	/* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
1082	/* port is ".ABC.DEF", 8 chars max */
1083	rlen = xdr_stream_decode_string_dup(xdr, &buf, INET6_ADDRSTRLEN +
1084					    IPV6_SCOPE_ID_LEN + 8, gfp_flags);
1085	if (unlikely(rlen < 0))
1086		goto out_free_netid;
1087
1088	/* replace port '.' with '-' */
1089	portstr = strrchr(buf, '.');
1090	if (!portstr) {
1091		dprintk("%s: Failed finding expected dot in port\n",
1092			__func__);
1093		goto out_free_buf;
1094	}
1095	*portstr = '-';
1096
1097	/* find '.' between address and port */
1098	portstr = strrchr(buf, '.');
1099	if (!portstr) {
1100		dprintk("%s: Failed finding expected dot between address and "
1101			"port\n", __func__);
1102		goto out_free_buf;
1103	}
1104	*portstr = '\0';
1105
1106	da = nfs4_pnfs_ds_addr_alloc(gfp_flags);
1107	if (unlikely(!da))
1108		goto out_free_buf;
1109
1110	if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
1111		      sizeof(da->da_addr))) {
1112		dprintk("%s: error parsing address %s\n", __func__, buf);
1113		goto out_free_da;
1114	}
1115
1116	portstr++;
1117	sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
1118	port = htons((tmp[0] << 8) | (tmp[1]));
1119
1120	switch (da->da_addr.ss_family) {
1121	case AF_INET:
1122		((struct sockaddr_in *)&da->da_addr)->sin_port = port;
1123		da->da_addrlen = sizeof(struct sockaddr_in);
1124		break;
1125
1126	case AF_INET6:
1127		((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
1128		da->da_addrlen = sizeof(struct sockaddr_in6);
1129		startsep = "[";
1130		endsep = "]";
1131		break;
1132
1133	default:
1134		dprintk("%s: unsupported address family: %u\n",
1135			__func__, da->da_addr.ss_family);
1136		goto out_free_da;
1137	}
1138
1139	da->da_transport = xprt_find_transport_ident(netid);
1140	if (da->da_transport < 0) {
1141		dprintk("%s: ERROR: unknown r_netid \"%s\"\n",
1142			__func__, netid);
1143		goto out_free_da;
1144	}
1145
1146	da->da_netid = netid;
1147
1148	/* save human readable address */
1149	len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
1150	da->da_remotestr = kzalloc(len, gfp_flags);
1151
1152	/* NULL is ok, only used for dprintk */
1153	if (da->da_remotestr)
1154		snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
1155			 buf, endsep, ntohs(port));
1156
1157	dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
1158	kfree(buf);
1159	return da;
1160
1161out_free_da:
1162	kfree(da);
1163out_free_buf:
1164	dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
1165	kfree(buf);
1166out_free_netid:
1167	kfree(netid);
1168out_err:
1169	return NULL;
1170}
1171EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
1172
1173void
1174pnfs_layout_mark_request_commit(struct nfs_page *req,
1175				struct pnfs_layout_segment *lseg,
1176				struct nfs_commit_info *cinfo,
1177				u32 ds_commit_idx)
1178{
1179	struct list_head *list;
1180	struct pnfs_commit_array *array;
1181	struct pnfs_commit_bucket *bucket;
1182
1183	mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
1184	array = pnfs_lookup_commit_array(cinfo->ds, lseg);
1185	if (!array || !pnfs_is_valid_lseg(lseg))
1186		goto out_resched;
1187	bucket = &array->buckets[ds_commit_idx];
1188	list = &bucket->written;
1189	/* Non-empty buckets hold a reference on the lseg.  That ref
1190	 * is normally transferred to the COMMIT call and released
1191	 * there.  It could also be released if the last req is pulled
1192	 * off due to a rewrite, in which case it will be done in
1193	 * pnfs_common_clear_request_commit
1194	 */
1195	if (!bucket->lseg)
1196		bucket->lseg = pnfs_get_lseg(lseg);
1197	set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1198	cinfo->ds->nwritten++;
1199
1200	nfs_request_add_commit_list_locked(req, list, cinfo);
1201	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1202	nfs_folio_mark_unstable(nfs_page_to_folio(req), cinfo);
1203	return;
1204out_resched:
1205	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1206	cinfo->completion_ops->resched_write(cinfo, req);
1207}
1208EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
1209
1210int
1211pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
1212{
1213	int ret;
1214
1215	if (!pnfs_layoutcommit_outstanding(inode))
1216		return 0;
1217	ret = nfs_commit_inode(inode, FLUSH_SYNC);
1218	if (ret < 0)
1219		return ret;
1220	if (datasync)
1221		return 0;
1222	return pnfs_layoutcommit_inode(inode, true);
1223}
1224EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
1225