1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/fs/nfs/pagelist.c
4 *
5 * A set of helper functions for managing NFS read and write requests.
6 * The main purpose of these routines is to provide support for the
7 * coalescing of several requests into a single RPC call.
8 *
9 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
10 *
11 */
12
13 #include <linux/slab.h>
14 #include <linux/file.h>
15 #include <linux/sched.h>
16 #include <linux/sunrpc/clnt.h>
17 #include <linux/nfs.h>
18 #include <linux/nfs3.h>
19 #include <linux/nfs4.h>
20 #include <linux/nfs_fs.h>
21 #include <linux/nfs_page.h>
22 #include <linux/nfs_mount.h>
23 #include <linux/export.h>
24 #include <linux/filelock.h>
25
26 #include "internal.h"
27 #include "pnfs.h"
28 #include "nfstrace.h"
29 #include "fscache.h"
30
31 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
32
33 static struct kmem_cache *nfs_page_cachep;
34 static const struct rpc_call_ops nfs_pgio_common_ops;
35
36 struct nfs_page_iter_page {
37 const struct nfs_page *req;
38 size_t count;
39 };
40
nfs_page_iter_page_init(struct nfs_page_iter_page * i,const struct nfs_page * req)41 static void nfs_page_iter_page_init(struct nfs_page_iter_page *i,
42 const struct nfs_page *req)
43 {
44 i->req = req;
45 i->count = 0;
46 }
47
nfs_page_iter_page_advance(struct nfs_page_iter_page * i,size_t sz)48 static void nfs_page_iter_page_advance(struct nfs_page_iter_page *i, size_t sz)
49 {
50 const struct nfs_page *req = i->req;
51 size_t tmp = i->count + sz;
52
53 i->count = (tmp < req->wb_bytes) ? tmp : req->wb_bytes;
54 }
55
nfs_page_iter_page_get(struct nfs_page_iter_page * i)56 static struct page *nfs_page_iter_page_get(struct nfs_page_iter_page *i)
57 {
58 const struct nfs_page *req = i->req;
59 struct page *page;
60
61 if (i->count != req->wb_bytes) {
62 size_t base = i->count + req->wb_pgbase;
63 size_t len = PAGE_SIZE - offset_in_page(base);
64
65 page = nfs_page_to_page(req, base);
66 nfs_page_iter_page_advance(i, len);
67 return page;
68 }
69 return NULL;
70 }
71
72 static struct nfs_pgio_mirror *
nfs_pgio_get_mirror(struct nfs_pageio_descriptor * desc,u32 idx)73 nfs_pgio_get_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
74 {
75 if (desc->pg_ops->pg_get_mirror)
76 return desc->pg_ops->pg_get_mirror(desc, idx);
77 return &desc->pg_mirrors[0];
78 }
79
80 struct nfs_pgio_mirror *
nfs_pgio_current_mirror(struct nfs_pageio_descriptor * desc)81 nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
82 {
83 return nfs_pgio_get_mirror(desc, desc->pg_mirror_idx);
84 }
85 EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
86
87 static u32
nfs_pgio_set_current_mirror(struct nfs_pageio_descriptor * desc,u32 idx)88 nfs_pgio_set_current_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
89 {
90 if (desc->pg_ops->pg_set_mirror)
91 return desc->pg_ops->pg_set_mirror(desc, idx);
92 return desc->pg_mirror_idx;
93 }
94
nfs_pgheader_init(struct nfs_pageio_descriptor * desc,struct nfs_pgio_header * hdr,void (* release)(struct nfs_pgio_header * hdr))95 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
96 struct nfs_pgio_header *hdr,
97 void (*release)(struct nfs_pgio_header *hdr))
98 {
99 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
100
101
102 hdr->req = nfs_list_entry(mirror->pg_list.next);
103 hdr->inode = desc->pg_inode;
104 hdr->cred = nfs_req_openctx(hdr->req)->cred;
105 hdr->io_start = req_offset(hdr->req);
106 hdr->good_bytes = mirror->pg_count;
107 hdr->io_completion = desc->pg_io_completion;
108 hdr->dreq = desc->pg_dreq;
109 nfs_netfs_set_pgio_header(hdr, desc);
110 hdr->release = release;
111 hdr->completion_ops = desc->pg_completion_ops;
112 if (hdr->completion_ops->init_hdr)
113 hdr->completion_ops->init_hdr(hdr);
114
115 hdr->pgio_mirror_idx = desc->pg_mirror_idx;
116 }
117 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
118
nfs_set_pgio_error(struct nfs_pgio_header * hdr,int error,loff_t pos)119 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
120 {
121 unsigned int new = pos - hdr->io_start;
122
123 trace_nfs_pgio_error(hdr, error, pos);
124 if (hdr->good_bytes > new) {
125 hdr->good_bytes = new;
126 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
127 if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags))
128 hdr->error = error;
129 }
130 }
131
nfs_page_alloc(void)132 static inline struct nfs_page *nfs_page_alloc(void)
133 {
134 struct nfs_page *p =
135 kmem_cache_zalloc(nfs_page_cachep, nfs_io_gfp_mask());
136 if (p)
137 INIT_LIST_HEAD(&p->wb_list);
138 return p;
139 }
140
141 static inline void
nfs_page_free(struct nfs_page * p)142 nfs_page_free(struct nfs_page *p)
143 {
144 kmem_cache_free(nfs_page_cachep, p);
145 }
146
147 /**
148 * nfs_iocounter_wait - wait for i/o to complete
149 * @l_ctx: nfs_lock_context with io_counter to use
150 *
151 * returns -ERESTARTSYS if interrupted by a fatal signal.
152 * Otherwise returns 0 once the io_count hits 0.
153 */
154 int
nfs_iocounter_wait(struct nfs_lock_context * l_ctx)155 nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
156 {
157 return wait_var_event_killable(&l_ctx->io_count,
158 !atomic_read(&l_ctx->io_count));
159 }
160
161 /**
162 * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O
163 * to complete
164 * @task: the rpc_task that should wait
165 * @l_ctx: nfs_lock_context with io_counter to check
166 *
167 * Returns true if there is outstanding I/O to wait on and the
168 * task has been put to sleep.
169 */
170 bool
nfs_async_iocounter_wait(struct rpc_task * task,struct nfs_lock_context * l_ctx)171 nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx)
172 {
173 struct inode *inode = d_inode(l_ctx->open_context->dentry);
174 bool ret = false;
175
176 if (atomic_read(&l_ctx->io_count) > 0) {
177 rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL);
178 ret = true;
179 }
180
181 if (atomic_read(&l_ctx->io_count) == 0) {
182 rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task);
183 ret = false;
184 }
185
186 return ret;
187 }
188 EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
189
190 /*
191 * nfs_page_set_headlock - set the request PG_HEADLOCK
192 * @req: request that is to be locked
193 *
194 * this lock must be held when modifying req->wb_head
195 *
196 * return 0 on success, < 0 on error
197 */
198 int
nfs_page_set_headlock(struct nfs_page * req)199 nfs_page_set_headlock(struct nfs_page *req)
200 {
201 if (!test_and_set_bit(PG_HEADLOCK, &req->wb_flags))
202 return 0;
203
204 set_bit(PG_CONTENDED1, &req->wb_flags);
205 smp_mb__after_atomic();
206 return wait_on_bit_lock(&req->wb_flags, PG_HEADLOCK,
207 TASK_UNINTERRUPTIBLE);
208 }
209
210 /*
211 * nfs_page_clear_headlock - clear the request PG_HEADLOCK
212 * @req: request that is to be locked
213 */
214 void
nfs_page_clear_headlock(struct nfs_page * req)215 nfs_page_clear_headlock(struct nfs_page *req)
216 {
217 clear_bit_unlock(PG_HEADLOCK, &req->wb_flags);
218 smp_mb__after_atomic();
219 if (!test_bit(PG_CONTENDED1, &req->wb_flags))
220 return;
221 wake_up_bit(&req->wb_flags, PG_HEADLOCK);
222 }
223
224 /*
225 * nfs_page_group_lock - lock the head of the page group
226 * @req: request in group that is to be locked
227 *
228 * this lock must be held when traversing or modifying the page
229 * group list
230 *
231 * return 0 on success, < 0 on error
232 */
233 int
nfs_page_group_lock(struct nfs_page * req)234 nfs_page_group_lock(struct nfs_page *req)
235 {
236 int ret;
237
238 ret = nfs_page_set_headlock(req);
239 if (ret || req->wb_head == req)
240 return ret;
241 return nfs_page_set_headlock(req->wb_head);
242 }
243
244 /*
245 * nfs_page_group_unlock - unlock the head of the page group
246 * @req: request in group that is to be unlocked
247 */
248 void
nfs_page_group_unlock(struct nfs_page * req)249 nfs_page_group_unlock(struct nfs_page *req)
250 {
251 if (req != req->wb_head)
252 nfs_page_clear_headlock(req->wb_head);
253 nfs_page_clear_headlock(req);
254 }
255
256 /*
257 * nfs_page_group_sync_on_bit_locked
258 *
259 * must be called with page group lock held
260 */
261 static bool
nfs_page_group_sync_on_bit_locked(struct nfs_page * req,unsigned int bit)262 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
263 {
264 struct nfs_page *head = req->wb_head;
265 struct nfs_page *tmp;
266
267 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
268 WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
269
270 tmp = req->wb_this_page;
271 while (tmp != req) {
272 if (!test_bit(bit, &tmp->wb_flags))
273 return false;
274 tmp = tmp->wb_this_page;
275 }
276
277 /* true! reset all bits */
278 tmp = req;
279 do {
280 clear_bit(bit, &tmp->wb_flags);
281 tmp = tmp->wb_this_page;
282 } while (tmp != req);
283
284 return true;
285 }
286
287 /*
288 * nfs_page_group_sync_on_bit - set bit on current request, but only
289 * return true if the bit is set for all requests in page group
290 * @req - request in page group
291 * @bit - PG_* bit that is used to sync page group
292 */
nfs_page_group_sync_on_bit(struct nfs_page * req,unsigned int bit)293 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
294 {
295 bool ret;
296
297 nfs_page_group_lock(req);
298 ret = nfs_page_group_sync_on_bit_locked(req, bit);
299 nfs_page_group_unlock(req);
300
301 return ret;
302 }
303
304 /*
305 * nfs_page_group_init - Initialize the page group linkage for @req
306 * @req - a new nfs request
307 * @prev - the previous request in page group, or NULL if @req is the first
308 * or only request in the group (the head).
309 */
310 static inline void
nfs_page_group_init(struct nfs_page * req,struct nfs_page * prev)311 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
312 {
313 struct inode *inode;
314 WARN_ON_ONCE(prev == req);
315
316 if (!prev) {
317 /* a head request */
318 req->wb_head = req;
319 req->wb_this_page = req;
320 } else {
321 /* a subrequest */
322 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
323 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
324 req->wb_head = prev->wb_head;
325 req->wb_this_page = prev->wb_this_page;
326 prev->wb_this_page = req;
327
328 /* All subrequests take a ref on the head request until
329 * nfs_page_group_destroy is called */
330 kref_get(&req->wb_head->wb_kref);
331
332 /* grab extra ref and bump the request count if head request
333 * has extra ref from the write/commit path to handle handoff
334 * between write and commit lists. */
335 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
336 inode = nfs_page_to_inode(req);
337 set_bit(PG_INODE_REF, &req->wb_flags);
338 kref_get(&req->wb_kref);
339 atomic_long_inc(&NFS_I(inode)->nrequests);
340 }
341 }
342 }
343
344 /*
345 * nfs_page_group_destroy - sync the destruction of page groups
346 * @req - request that no longer needs the page group
347 *
348 * releases the page group reference from each member once all
349 * members have called this function.
350 */
351 static void
nfs_page_group_destroy(struct kref * kref)352 nfs_page_group_destroy(struct kref *kref)
353 {
354 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
355 struct nfs_page *head = req->wb_head;
356 struct nfs_page *tmp, *next;
357
358 if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
359 goto out;
360
361 tmp = req;
362 do {
363 next = tmp->wb_this_page;
364 /* unlink and free */
365 tmp->wb_this_page = tmp;
366 tmp->wb_head = tmp;
367 nfs_free_request(tmp);
368 tmp = next;
369 } while (tmp != req);
370 out:
371 /* subrequests must release the ref on the head request */
372 if (head != req)
373 nfs_release_request(head);
374 }
375
nfs_page_create(struct nfs_lock_context * l_ctx,unsigned int pgbase,pgoff_t index,unsigned int offset,unsigned int count)376 static struct nfs_page *nfs_page_create(struct nfs_lock_context *l_ctx,
377 unsigned int pgbase, pgoff_t index,
378 unsigned int offset, unsigned int count)
379 {
380 struct nfs_page *req;
381 struct nfs_open_context *ctx = l_ctx->open_context;
382
383 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
384 return ERR_PTR(-EBADF);
385 /* try to allocate the request struct */
386 req = nfs_page_alloc();
387 if (req == NULL)
388 return ERR_PTR(-ENOMEM);
389
390 req->wb_lock_context = l_ctx;
391 refcount_inc(&l_ctx->count);
392 atomic_inc(&l_ctx->io_count);
393
394 /* Initialize the request struct. Initially, we assume a
395 * long write-back delay. This will be adjusted in
396 * update_nfs_request below if the region is not locked. */
397 req->wb_pgbase = pgbase;
398 req->wb_index = index;
399 req->wb_offset = offset;
400 req->wb_bytes = count;
401 kref_init(&req->wb_kref);
402 req->wb_nio = 0;
403 return req;
404 }
405
nfs_page_assign_folio(struct nfs_page * req,struct folio * folio)406 static void nfs_page_assign_folio(struct nfs_page *req, struct folio *folio)
407 {
408 if (folio != NULL) {
409 req->wb_folio = folio;
410 folio_get(folio);
411 set_bit(PG_FOLIO, &req->wb_flags);
412 }
413 }
414
nfs_page_assign_page(struct nfs_page * req,struct page * page)415 static void nfs_page_assign_page(struct nfs_page *req, struct page *page)
416 {
417 if (page != NULL) {
418 req->wb_page = page;
419 get_page(page);
420 }
421 }
422
423 /**
424 * nfs_page_create_from_page - Create an NFS read/write request.
425 * @ctx: open context to use
426 * @page: page to write
427 * @pgbase: starting offset within the page for the write
428 * @offset: file offset for the write
429 * @count: number of bytes to read/write
430 *
431 * The page must be locked by the caller. This makes sure we never
432 * create two different requests for the same page.
433 * User should ensure it is safe to sleep in this function.
434 */
nfs_page_create_from_page(struct nfs_open_context * ctx,struct page * page,unsigned int pgbase,loff_t offset,unsigned int count)435 struct nfs_page *nfs_page_create_from_page(struct nfs_open_context *ctx,
436 struct page *page,
437 unsigned int pgbase, loff_t offset,
438 unsigned int count)
439 {
440 struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
441 struct nfs_page *ret;
442
443 if (IS_ERR(l_ctx))
444 return ERR_CAST(l_ctx);
445 ret = nfs_page_create(l_ctx, pgbase, offset >> PAGE_SHIFT,
446 offset_in_page(offset), count);
447 if (!IS_ERR(ret)) {
448 nfs_page_assign_page(ret, page);
449 nfs_page_group_init(ret, NULL);
450 }
451 nfs_put_lock_context(l_ctx);
452 return ret;
453 }
454
455 /**
456 * nfs_page_create_from_folio - Create an NFS read/write request.
457 * @ctx: open context to use
458 * @folio: folio to write
459 * @offset: starting offset within the folio for the write
460 * @count: number of bytes to read/write
461 *
462 * The page must be locked by the caller. This makes sure we never
463 * create two different requests for the same page.
464 * User should ensure it is safe to sleep in this function.
465 */
nfs_page_create_from_folio(struct nfs_open_context * ctx,struct folio * folio,unsigned int offset,unsigned int count)466 struct nfs_page *nfs_page_create_from_folio(struct nfs_open_context *ctx,
467 struct folio *folio,
468 unsigned int offset,
469 unsigned int count)
470 {
471 struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
472 struct nfs_page *ret;
473
474 if (IS_ERR(l_ctx))
475 return ERR_CAST(l_ctx);
476 ret = nfs_page_create(l_ctx, offset, folio->index, offset, count);
477 if (!IS_ERR(ret)) {
478 nfs_page_assign_folio(ret, folio);
479 nfs_page_group_init(ret, NULL);
480 }
481 nfs_put_lock_context(l_ctx);
482 return ret;
483 }
484
485 static struct nfs_page *
nfs_create_subreq(struct nfs_page * req,unsigned int pgbase,unsigned int offset,unsigned int count)486 nfs_create_subreq(struct nfs_page *req,
487 unsigned int pgbase,
488 unsigned int offset,
489 unsigned int count)
490 {
491 struct nfs_page *last;
492 struct nfs_page *ret;
493 struct folio *folio = nfs_page_to_folio(req);
494 struct page *page = nfs_page_to_page(req, pgbase);
495
496 ret = nfs_page_create(req->wb_lock_context, pgbase, req->wb_index,
497 offset, count);
498 if (!IS_ERR(ret)) {
499 if (folio)
500 nfs_page_assign_folio(ret, folio);
501 else
502 nfs_page_assign_page(ret, page);
503 /* find the last request */
504 for (last = req->wb_head;
505 last->wb_this_page != req->wb_head;
506 last = last->wb_this_page)
507 ;
508
509 nfs_lock_request(ret);
510 nfs_page_group_init(ret, last);
511 ret->wb_nio = req->wb_nio;
512 }
513 return ret;
514 }
515
516 /**
517 * nfs_unlock_request - Unlock request and wake up sleepers.
518 * @req: pointer to request
519 */
nfs_unlock_request(struct nfs_page * req)520 void nfs_unlock_request(struct nfs_page *req)
521 {
522 clear_bit_unlock(PG_BUSY, &req->wb_flags);
523 smp_mb__after_atomic();
524 if (!test_bit(PG_CONTENDED2, &req->wb_flags))
525 return;
526 wake_up_bit(&req->wb_flags, PG_BUSY);
527 }
528
529 /**
530 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
531 * @req: pointer to request
532 */
nfs_unlock_and_release_request(struct nfs_page * req)533 void nfs_unlock_and_release_request(struct nfs_page *req)
534 {
535 nfs_unlock_request(req);
536 nfs_release_request(req);
537 }
538
539 /*
540 * nfs_clear_request - Free up all resources allocated to the request
541 * @req:
542 *
543 * Release page and open context resources associated with a read/write
544 * request after it has completed.
545 */
nfs_clear_request(struct nfs_page * req)546 static void nfs_clear_request(struct nfs_page *req)
547 {
548 struct folio *folio = nfs_page_to_folio(req);
549 struct page *page = req->wb_page;
550 struct nfs_lock_context *l_ctx = req->wb_lock_context;
551 struct nfs_open_context *ctx;
552
553 if (folio != NULL) {
554 folio_put(folio);
555 req->wb_folio = NULL;
556 clear_bit(PG_FOLIO, &req->wb_flags);
557 } else if (page != NULL) {
558 put_page(page);
559 req->wb_page = NULL;
560 }
561 if (l_ctx != NULL) {
562 if (atomic_dec_and_test(&l_ctx->io_count)) {
563 wake_up_var(&l_ctx->io_count);
564 ctx = l_ctx->open_context;
565 if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
566 rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
567 }
568 nfs_put_lock_context(l_ctx);
569 req->wb_lock_context = NULL;
570 }
571 }
572
573 /**
574 * nfs_free_request - Release the count on an NFS read/write request
575 * @req: request to release
576 *
577 * Note: Should never be called with the spinlock held!
578 */
nfs_free_request(struct nfs_page * req)579 void nfs_free_request(struct nfs_page *req)
580 {
581 WARN_ON_ONCE(req->wb_this_page != req);
582
583 /* extra debug: make sure no sync bits are still set */
584 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
585 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
586 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
587 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
588 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
589
590 /* Release struct file and open context */
591 nfs_clear_request(req);
592 nfs_page_free(req);
593 }
594
nfs_release_request(struct nfs_page * req)595 void nfs_release_request(struct nfs_page *req)
596 {
597 kref_put(&req->wb_kref, nfs_page_group_destroy);
598 }
599 EXPORT_SYMBOL_GPL(nfs_release_request);
600
601 /*
602 * nfs_generic_pg_test - determine if requests can be coalesced
603 * @desc: pointer to descriptor
604 * @prev: previous request in desc, or NULL
605 * @req: this request
606 *
607 * Returns zero if @req cannot be coalesced into @desc, otherwise it returns
608 * the size of the request.
609 */
nfs_generic_pg_test(struct nfs_pageio_descriptor * desc,struct nfs_page * prev,struct nfs_page * req)610 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
611 struct nfs_page *prev, struct nfs_page *req)
612 {
613 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
614
615
616 if (mirror->pg_count > mirror->pg_bsize) {
617 /* should never happen */
618 WARN_ON_ONCE(1);
619 return 0;
620 }
621
622 /*
623 * Limit the request size so that we can still allocate a page array
624 * for it without upsetting the slab allocator.
625 */
626 if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
627 sizeof(struct page *) > PAGE_SIZE)
628 return 0;
629
630 return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
631 }
632 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
633
nfs_pgio_header_alloc(const struct nfs_rw_ops * ops)634 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
635 {
636 struct nfs_pgio_header *hdr = ops->rw_alloc_header();
637
638 if (hdr) {
639 INIT_LIST_HEAD(&hdr->pages);
640 hdr->rw_ops = ops;
641 }
642 return hdr;
643 }
644 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
645
646 /**
647 * nfs_pgio_data_destroy - make @hdr suitable for reuse
648 *
649 * Frees memory and releases refs from nfs_generic_pgio, so that it may
650 * be called again.
651 *
652 * @hdr: A header that has had nfs_generic_pgio called
653 */
nfs_pgio_data_destroy(struct nfs_pgio_header * hdr)654 static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
655 {
656 if (hdr->args.context)
657 put_nfs_open_context(hdr->args.context);
658 if (hdr->page_array.pagevec != hdr->page_array.page_array)
659 kfree(hdr->page_array.pagevec);
660 }
661
662 /*
663 * nfs_pgio_header_free - Free a read or write header
664 * @hdr: The header to free
665 */
nfs_pgio_header_free(struct nfs_pgio_header * hdr)666 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
667 {
668 nfs_pgio_data_destroy(hdr);
669 hdr->rw_ops->rw_free_header(hdr);
670 }
671 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
672
673 /**
674 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
675 * @hdr: The pageio hdr
676 * @pgbase: base
677 * @count: Number of bytes to read
678 * @how: How to commit data (writes only)
679 * @cinfo: Commit information for the call (writes only)
680 */
nfs_pgio_rpcsetup(struct nfs_pgio_header * hdr,unsigned int pgbase,unsigned int count,int how,struct nfs_commit_info * cinfo)681 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr, unsigned int pgbase,
682 unsigned int count, int how,
683 struct nfs_commit_info *cinfo)
684 {
685 struct nfs_page *req = hdr->req;
686
687 /* Set up the RPC argument and reply structs
688 * NB: take care not to mess about with hdr->commit et al. */
689
690 hdr->args.fh = NFS_FH(hdr->inode);
691 hdr->args.offset = req_offset(req);
692 /* pnfs_set_layoutcommit needs this */
693 hdr->mds_offset = hdr->args.offset;
694 hdr->args.pgbase = pgbase;
695 hdr->args.pages = hdr->page_array.pagevec;
696 hdr->args.count = count;
697 hdr->args.context = get_nfs_open_context(nfs_req_openctx(req));
698 hdr->args.lock_context = req->wb_lock_context;
699 hdr->args.stable = NFS_UNSTABLE;
700 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
701 case 0:
702 break;
703 case FLUSH_COND_STABLE:
704 if (nfs_reqs_to_commit(cinfo))
705 break;
706 fallthrough;
707 default:
708 hdr->args.stable = NFS_FILE_SYNC;
709 }
710
711 hdr->res.fattr = &hdr->fattr;
712 hdr->res.count = 0;
713 hdr->res.eof = 0;
714 hdr->res.verf = &hdr->verf;
715 nfs_fattr_init(&hdr->fattr);
716 }
717
718 /**
719 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
720 * @task: The current task
721 * @calldata: pageio header to prepare
722 */
nfs_pgio_prepare(struct rpc_task * task,void * calldata)723 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
724 {
725 struct nfs_pgio_header *hdr = calldata;
726 int err;
727 err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
728 if (err)
729 rpc_exit(task, err);
730 }
731
nfs_initiate_pgio(struct rpc_clnt * clnt,struct nfs_pgio_header * hdr,const struct cred * cred,const struct nfs_rpc_ops * rpc_ops,const struct rpc_call_ops * call_ops,int how,int flags)732 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
733 const struct cred *cred, const struct nfs_rpc_ops *rpc_ops,
734 const struct rpc_call_ops *call_ops, int how, int flags)
735 {
736 struct rpc_task *task;
737 struct rpc_message msg = {
738 .rpc_argp = &hdr->args,
739 .rpc_resp = &hdr->res,
740 .rpc_cred = cred,
741 };
742 struct rpc_task_setup task_setup_data = {
743 .rpc_client = clnt,
744 .task = &hdr->task,
745 .rpc_message = &msg,
746 .callback_ops = call_ops,
747 .callback_data = hdr,
748 .workqueue = nfsiod_workqueue,
749 .flags = RPC_TASK_ASYNC | flags,
750 };
751
752 if (nfs_server_capable(hdr->inode, NFS_CAP_MOVEABLE))
753 task_setup_data.flags |= RPC_TASK_MOVEABLE;
754
755 hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
756
757 dprintk("NFS: initiated pgio call "
758 "(req %s/%llu, %u bytes @ offset %llu)\n",
759 hdr->inode->i_sb->s_id,
760 (unsigned long long)NFS_FILEID(hdr->inode),
761 hdr->args.count,
762 (unsigned long long)hdr->args.offset);
763
764 task = rpc_run_task(&task_setup_data);
765 if (IS_ERR(task))
766 return PTR_ERR(task);
767 rpc_put_task(task);
768 return 0;
769 }
770 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
771
772 /**
773 * nfs_pgio_error - Clean up from a pageio error
774 * @hdr: pageio header
775 */
nfs_pgio_error(struct nfs_pgio_header * hdr)776 static void nfs_pgio_error(struct nfs_pgio_header *hdr)
777 {
778 set_bit(NFS_IOHDR_REDO, &hdr->flags);
779 hdr->completion_ops->completion(hdr);
780 }
781
782 /**
783 * nfs_pgio_release - Release pageio data
784 * @calldata: The pageio header to release
785 */
nfs_pgio_release(void * calldata)786 static void nfs_pgio_release(void *calldata)
787 {
788 struct nfs_pgio_header *hdr = calldata;
789 hdr->completion_ops->completion(hdr);
790 }
791
nfs_pageio_mirror_init(struct nfs_pgio_mirror * mirror,unsigned int bsize)792 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
793 unsigned int bsize)
794 {
795 INIT_LIST_HEAD(&mirror->pg_list);
796 mirror->pg_bytes_written = 0;
797 mirror->pg_count = 0;
798 mirror->pg_bsize = bsize;
799 mirror->pg_base = 0;
800 mirror->pg_recoalesce = 0;
801 }
802
803 /**
804 * nfs_pageio_init - initialise a page io descriptor
805 * @desc: pointer to descriptor
806 * @inode: pointer to inode
807 * @pg_ops: pointer to pageio operations
808 * @compl_ops: pointer to pageio completion operations
809 * @rw_ops: pointer to nfs read/write operations
810 * @bsize: io block size
811 * @io_flags: extra parameters for the io function
812 */
nfs_pageio_init(struct nfs_pageio_descriptor * desc,struct inode * inode,const struct nfs_pageio_ops * pg_ops,const struct nfs_pgio_completion_ops * compl_ops,const struct nfs_rw_ops * rw_ops,size_t bsize,int io_flags)813 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
814 struct inode *inode,
815 const struct nfs_pageio_ops *pg_ops,
816 const struct nfs_pgio_completion_ops *compl_ops,
817 const struct nfs_rw_ops *rw_ops,
818 size_t bsize,
819 int io_flags)
820 {
821 desc->pg_moreio = 0;
822 desc->pg_inode = inode;
823 desc->pg_ops = pg_ops;
824 desc->pg_completion_ops = compl_ops;
825 desc->pg_rw_ops = rw_ops;
826 desc->pg_ioflags = io_flags;
827 desc->pg_error = 0;
828 desc->pg_lseg = NULL;
829 desc->pg_io_completion = NULL;
830 desc->pg_dreq = NULL;
831 nfs_netfs_reset_pageio_descriptor(desc);
832 desc->pg_bsize = bsize;
833
834 desc->pg_mirror_count = 1;
835 desc->pg_mirror_idx = 0;
836
837 desc->pg_mirrors_dynamic = NULL;
838 desc->pg_mirrors = desc->pg_mirrors_static;
839 nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
840 desc->pg_maxretrans = 0;
841 }
842
843 /**
844 * nfs_pgio_result - Basic pageio error handling
845 * @task: The task that ran
846 * @calldata: Pageio header to check
847 */
nfs_pgio_result(struct rpc_task * task,void * calldata)848 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
849 {
850 struct nfs_pgio_header *hdr = calldata;
851 struct inode *inode = hdr->inode;
852
853 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
854 return;
855 if (task->tk_status < 0)
856 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
857 else
858 hdr->rw_ops->rw_result(task, hdr);
859 }
860
861 /*
862 * Create an RPC task for the given read or write request and kick it.
863 * The page must have been locked by the caller.
864 *
865 * It may happen that the page we're passed is not marked dirty.
866 * This is the case if nfs_updatepage detects a conflicting request
867 * that has been written but not committed.
868 */
nfs_generic_pgio(struct nfs_pageio_descriptor * desc,struct nfs_pgio_header * hdr)869 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
870 struct nfs_pgio_header *hdr)
871 {
872 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
873
874 struct nfs_page *req;
875 struct page **pages,
876 *last_page;
877 struct list_head *head = &mirror->pg_list;
878 struct nfs_commit_info cinfo;
879 struct nfs_page_array *pg_array = &hdr->page_array;
880 unsigned int pagecount, pageused;
881 unsigned int pg_base = offset_in_page(mirror->pg_base);
882 gfp_t gfp_flags = nfs_io_gfp_mask();
883
884 pagecount = nfs_page_array_len(pg_base, mirror->pg_count);
885 pg_array->npages = pagecount;
886
887 if (pagecount <= ARRAY_SIZE(pg_array->page_array))
888 pg_array->pagevec = pg_array->page_array;
889 else {
890 pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
891 if (!pg_array->pagevec) {
892 pg_array->npages = 0;
893 nfs_pgio_error(hdr);
894 desc->pg_error = -ENOMEM;
895 return desc->pg_error;
896 }
897 }
898
899 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
900 pages = hdr->page_array.pagevec;
901 last_page = NULL;
902 pageused = 0;
903 while (!list_empty(head)) {
904 struct nfs_page_iter_page i;
905 struct page *page;
906
907 req = nfs_list_entry(head->next);
908 nfs_list_move_request(req, &hdr->pages);
909
910 if (req->wb_pgbase == 0)
911 last_page = NULL;
912
913 nfs_page_iter_page_init(&i, req);
914 while ((page = nfs_page_iter_page_get(&i)) != NULL) {
915 if (last_page != page) {
916 pageused++;
917 if (pageused > pagecount)
918 goto full;
919 *pages++ = last_page = page;
920 }
921 }
922 }
923 full:
924 if (WARN_ON_ONCE(pageused != pagecount)) {
925 nfs_pgio_error(hdr);
926 desc->pg_error = -EINVAL;
927 return desc->pg_error;
928 }
929
930 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
931 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
932 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
933
934 /* Set up the argument struct */
935 nfs_pgio_rpcsetup(hdr, pg_base, mirror->pg_count, desc->pg_ioflags,
936 &cinfo);
937 desc->pg_rpc_callops = &nfs_pgio_common_ops;
938 return 0;
939 }
940 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
941
nfs_generic_pg_pgios(struct nfs_pageio_descriptor * desc)942 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
943 {
944 struct nfs_pgio_header *hdr;
945 int ret;
946 unsigned short task_flags = 0;
947
948 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
949 if (!hdr) {
950 desc->pg_error = -ENOMEM;
951 return desc->pg_error;
952 }
953 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
954 ret = nfs_generic_pgio(desc, hdr);
955 if (ret == 0) {
956 if (NFS_SERVER(hdr->inode)->nfs_client->cl_minorversion)
957 task_flags = RPC_TASK_MOVEABLE;
958 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
959 hdr,
960 hdr->cred,
961 NFS_PROTO(hdr->inode),
962 desc->pg_rpc_callops,
963 desc->pg_ioflags,
964 RPC_TASK_CRED_NOREF | task_flags);
965 }
966 return ret;
967 }
968
969 static struct nfs_pgio_mirror *
nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor * desc,unsigned int mirror_count)970 nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
971 unsigned int mirror_count)
972 {
973 struct nfs_pgio_mirror *ret;
974 unsigned int i;
975
976 kfree(desc->pg_mirrors_dynamic);
977 desc->pg_mirrors_dynamic = NULL;
978 if (mirror_count == 1)
979 return desc->pg_mirrors_static;
980 ret = kmalloc_array(mirror_count, sizeof(*ret), nfs_io_gfp_mask());
981 if (ret != NULL) {
982 for (i = 0; i < mirror_count; i++)
983 nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
984 desc->pg_mirrors_dynamic = ret;
985 }
986 return ret;
987 }
988
989 /*
990 * nfs_pageio_setup_mirroring - determine if mirroring is to be used
991 * by calling the pg_get_mirror_count op
992 */
nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor * pgio,struct nfs_page * req)993 static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
994 struct nfs_page *req)
995 {
996 unsigned int mirror_count = 1;
997
998 if (pgio->pg_ops->pg_get_mirror_count)
999 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
1000 if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
1001 return;
1002
1003 if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
1004 pgio->pg_error = -EINVAL;
1005 return;
1006 }
1007
1008 pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
1009 if (pgio->pg_mirrors == NULL) {
1010 pgio->pg_error = -ENOMEM;
1011 pgio->pg_mirrors = pgio->pg_mirrors_static;
1012 mirror_count = 1;
1013 }
1014 pgio->pg_mirror_count = mirror_count;
1015 }
1016
nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor * pgio)1017 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
1018 {
1019 pgio->pg_mirror_count = 1;
1020 pgio->pg_mirror_idx = 0;
1021 pgio->pg_mirrors = pgio->pg_mirrors_static;
1022 kfree(pgio->pg_mirrors_dynamic);
1023 pgio->pg_mirrors_dynamic = NULL;
1024 }
1025
nfs_match_lock_context(const struct nfs_lock_context * l1,const struct nfs_lock_context * l2)1026 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
1027 const struct nfs_lock_context *l2)
1028 {
1029 return l1->lockowner == l2->lockowner;
1030 }
1031
nfs_page_is_contiguous(const struct nfs_page * prev,const struct nfs_page * req)1032 static bool nfs_page_is_contiguous(const struct nfs_page *prev,
1033 const struct nfs_page *req)
1034 {
1035 size_t prev_end = prev->wb_pgbase + prev->wb_bytes;
1036
1037 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
1038 return false;
1039 if (req->wb_pgbase == 0)
1040 return prev_end == nfs_page_max_length(prev);
1041 if (req->wb_pgbase == prev_end) {
1042 struct folio *folio = nfs_page_to_folio(req);
1043 if (folio)
1044 return folio == nfs_page_to_folio(prev);
1045 return req->wb_page == prev->wb_page;
1046 }
1047 return false;
1048 }
1049
1050 /**
1051 * nfs_coalesce_size - test two requests for compatibility
1052 * @prev: pointer to nfs_page
1053 * @req: pointer to nfs_page
1054 * @pgio: pointer to nfs_pagio_descriptor
1055 *
1056 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
1057 * page data area they describe is contiguous, and that their RPC
1058 * credentials, NFSv4 open state, and lockowners are the same.
1059 *
1060 * Returns size of the request that can be coalesced
1061 */
nfs_coalesce_size(struct nfs_page * prev,struct nfs_page * req,struct nfs_pageio_descriptor * pgio)1062 static unsigned int nfs_coalesce_size(struct nfs_page *prev,
1063 struct nfs_page *req,
1064 struct nfs_pageio_descriptor *pgio)
1065 {
1066 struct file_lock_context *flctx;
1067
1068 if (prev) {
1069 if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
1070 return 0;
1071 flctx = locks_inode_context(d_inode(nfs_req_openctx(req)->dentry));
1072 if (flctx != NULL &&
1073 !(list_empty_careful(&flctx->flc_posix) &&
1074 list_empty_careful(&flctx->flc_flock)) &&
1075 !nfs_match_lock_context(req->wb_lock_context,
1076 prev->wb_lock_context))
1077 return 0;
1078 if (!nfs_page_is_contiguous(prev, req))
1079 return 0;
1080 }
1081 return pgio->pg_ops->pg_test(pgio, prev, req);
1082 }
1083
1084 /**
1085 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
1086 * @desc: destination io descriptor
1087 * @req: request
1088 *
1089 * If the request 'req' was successfully coalesced into the existing list
1090 * of pages 'desc', it returns the size of req.
1091 */
1092 static unsigned int
nfs_pageio_do_add_request(struct nfs_pageio_descriptor * desc,struct nfs_page * req)1093 nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
1094 struct nfs_page *req)
1095 {
1096 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1097 struct nfs_page *prev = NULL;
1098 unsigned int size;
1099
1100 if (list_empty(&mirror->pg_list)) {
1101 if (desc->pg_ops->pg_init)
1102 desc->pg_ops->pg_init(desc, req);
1103 if (desc->pg_error < 0)
1104 return 0;
1105 mirror->pg_base = req->wb_pgbase;
1106 mirror->pg_count = 0;
1107 mirror->pg_recoalesce = 0;
1108 } else
1109 prev = nfs_list_entry(mirror->pg_list.prev);
1110
1111 if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
1112 if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
1113 desc->pg_error = -ETIMEDOUT;
1114 else
1115 desc->pg_error = -EIO;
1116 return 0;
1117 }
1118
1119 size = nfs_coalesce_size(prev, req, desc);
1120 if (size < req->wb_bytes)
1121 return size;
1122 nfs_list_move_request(req, &mirror->pg_list);
1123 mirror->pg_count += req->wb_bytes;
1124 return req->wb_bytes;
1125 }
1126
1127 /*
1128 * Helper for nfs_pageio_add_request and nfs_pageio_complete
1129 */
nfs_pageio_doio(struct nfs_pageio_descriptor * desc)1130 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
1131 {
1132 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1133
1134 if (!list_empty(&mirror->pg_list)) {
1135 int error = desc->pg_ops->pg_doio(desc);
1136 if (error < 0)
1137 desc->pg_error = error;
1138 if (list_empty(&mirror->pg_list))
1139 mirror->pg_bytes_written += mirror->pg_count;
1140 }
1141 }
1142
1143 static void
nfs_pageio_cleanup_request(struct nfs_pageio_descriptor * desc,struct nfs_page * req)1144 nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1145 struct nfs_page *req)
1146 {
1147 LIST_HEAD(head);
1148
1149 nfs_list_move_request(req, &head);
1150 desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1151 }
1152
1153 /**
1154 * __nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1155 * @desc: destination io descriptor
1156 * @req: request
1157 *
1158 * This may split a request into subrequests which are all part of the
1159 * same page group. If so, it will submit @req as the last one, to ensure
1160 * the pointer to @req is still valid in case of failure.
1161 *
1162 * Returns true if the request 'req' was successfully coalesced into the
1163 * existing list of pages 'desc'.
1164 */
__nfs_pageio_add_request(struct nfs_pageio_descriptor * desc,struct nfs_page * req)1165 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1166 struct nfs_page *req)
1167 {
1168 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1169 struct nfs_page *subreq;
1170 unsigned int size, subreq_size;
1171
1172 nfs_page_group_lock(req);
1173
1174 subreq = req;
1175 subreq_size = subreq->wb_bytes;
1176 for(;;) {
1177 size = nfs_pageio_do_add_request(desc, subreq);
1178 if (size == subreq_size) {
1179 /* We successfully submitted a request */
1180 if (subreq == req)
1181 break;
1182 req->wb_pgbase += size;
1183 req->wb_bytes -= size;
1184 req->wb_offset += size;
1185 subreq_size = req->wb_bytes;
1186 subreq = req;
1187 continue;
1188 }
1189 if (WARN_ON_ONCE(subreq != req)) {
1190 nfs_page_group_unlock(req);
1191 nfs_pageio_cleanup_request(desc, subreq);
1192 subreq = req;
1193 subreq_size = req->wb_bytes;
1194 nfs_page_group_lock(req);
1195 }
1196 if (!size) {
1197 /* Can't coalesce any more, so do I/O */
1198 nfs_page_group_unlock(req);
1199 desc->pg_moreio = 1;
1200 nfs_pageio_doio(desc);
1201 if (desc->pg_error < 0 || mirror->pg_recoalesce)
1202 return 0;
1203 /* retry add_request for this subreq */
1204 nfs_page_group_lock(req);
1205 continue;
1206 }
1207 subreq = nfs_create_subreq(req, req->wb_pgbase,
1208 req->wb_offset, size);
1209 if (IS_ERR(subreq))
1210 goto err_ptr;
1211 subreq_size = size;
1212 }
1213
1214 nfs_page_group_unlock(req);
1215 return 1;
1216 err_ptr:
1217 desc->pg_error = PTR_ERR(subreq);
1218 nfs_page_group_unlock(req);
1219 return 0;
1220 }
1221
nfs_do_recoalesce(struct nfs_pageio_descriptor * desc)1222 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1223 {
1224 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1225 LIST_HEAD(head);
1226
1227 do {
1228 list_splice_init(&mirror->pg_list, &head);
1229 mirror->pg_recoalesce = 0;
1230
1231 while (!list_empty(&head)) {
1232 struct nfs_page *req;
1233
1234 req = list_first_entry(&head, struct nfs_page, wb_list);
1235 if (__nfs_pageio_add_request(desc, req))
1236 continue;
1237 if (desc->pg_error < 0) {
1238 list_splice_tail(&head, &mirror->pg_list);
1239 mirror->pg_recoalesce = 1;
1240 return 0;
1241 }
1242 break;
1243 }
1244 } while (mirror->pg_recoalesce);
1245 return 1;
1246 }
1247
nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor * desc,struct nfs_page * req)1248 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1249 struct nfs_page *req)
1250 {
1251 int ret;
1252
1253 do {
1254 ret = __nfs_pageio_add_request(desc, req);
1255 if (ret)
1256 break;
1257 if (desc->pg_error < 0)
1258 break;
1259 ret = nfs_do_recoalesce(desc);
1260 } while (ret);
1261
1262 return ret;
1263 }
1264
nfs_pageio_error_cleanup(struct nfs_pageio_descriptor * desc)1265 static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1266 {
1267 u32 midx;
1268 struct nfs_pgio_mirror *mirror;
1269
1270 if (!desc->pg_error)
1271 return;
1272
1273 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1274 mirror = nfs_pgio_get_mirror(desc, midx);
1275 desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1276 desc->pg_error);
1277 }
1278 }
1279
nfs_pageio_add_request(struct nfs_pageio_descriptor * desc,struct nfs_page * req)1280 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1281 struct nfs_page *req)
1282 {
1283 u32 midx;
1284 unsigned int pgbase, offset, bytes;
1285 struct nfs_page *dupreq;
1286
1287 pgbase = req->wb_pgbase;
1288 offset = req->wb_offset;
1289 bytes = req->wb_bytes;
1290
1291 nfs_pageio_setup_mirroring(desc, req);
1292 if (desc->pg_error < 0)
1293 goto out_failed;
1294
1295 /* Create the mirror instances first, and fire them off */
1296 for (midx = 1; midx < desc->pg_mirror_count; midx++) {
1297 nfs_page_group_lock(req);
1298
1299 dupreq = nfs_create_subreq(req,
1300 pgbase, offset, bytes);
1301
1302 nfs_page_group_unlock(req);
1303 if (IS_ERR(dupreq)) {
1304 desc->pg_error = PTR_ERR(dupreq);
1305 goto out_failed;
1306 }
1307
1308 nfs_pgio_set_current_mirror(desc, midx);
1309 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1310 goto out_cleanup_subreq;
1311 }
1312
1313 nfs_pgio_set_current_mirror(desc, 0);
1314 if (!nfs_pageio_add_request_mirror(desc, req))
1315 goto out_failed;
1316
1317 return 1;
1318
1319 out_cleanup_subreq:
1320 nfs_pageio_cleanup_request(desc, dupreq);
1321 out_failed:
1322 nfs_pageio_error_cleanup(desc);
1323 return 0;
1324 }
1325
1326 /*
1327 * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1328 * nfs_pageio_descriptor
1329 * @desc: pointer to io descriptor
1330 * @mirror_idx: pointer to mirror index
1331 */
nfs_pageio_complete_mirror(struct nfs_pageio_descriptor * desc,u32 mirror_idx)1332 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1333 u32 mirror_idx)
1334 {
1335 struct nfs_pgio_mirror *mirror;
1336 u32 restore_idx;
1337
1338 restore_idx = nfs_pgio_set_current_mirror(desc, mirror_idx);
1339 mirror = nfs_pgio_current_mirror(desc);
1340
1341 for (;;) {
1342 nfs_pageio_doio(desc);
1343 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1344 break;
1345 if (!nfs_do_recoalesce(desc))
1346 break;
1347 }
1348 nfs_pgio_set_current_mirror(desc, restore_idx);
1349 }
1350
1351 /*
1352 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1353 * @hdr - the pgio header to move request from
1354 * @desc - the pageio descriptor to add requests to
1355 *
1356 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1357 * to send them.
1358 *
1359 * Returns 0 on success and < 0 on error.
1360 */
nfs_pageio_resend(struct nfs_pageio_descriptor * desc,struct nfs_pgio_header * hdr)1361 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1362 struct nfs_pgio_header *hdr)
1363 {
1364 LIST_HEAD(pages);
1365
1366 desc->pg_io_completion = hdr->io_completion;
1367 desc->pg_dreq = hdr->dreq;
1368 nfs_netfs_set_pageio_descriptor(desc, hdr);
1369 list_splice_init(&hdr->pages, &pages);
1370 while (!list_empty(&pages)) {
1371 struct nfs_page *req = nfs_list_entry(pages.next);
1372
1373 if (!nfs_pageio_add_request(desc, req))
1374 break;
1375 }
1376 nfs_pageio_complete(desc);
1377 if (!list_empty(&pages)) {
1378 int err = desc->pg_error < 0 ? desc->pg_error : -EIO;
1379 hdr->completion_ops->error_cleanup(&pages, err);
1380 nfs_set_pgio_error(hdr, err, hdr->io_start);
1381 return err;
1382 }
1383 return 0;
1384 }
1385 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1386
1387 /**
1388 * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1389 * @desc: pointer to io descriptor
1390 */
nfs_pageio_complete(struct nfs_pageio_descriptor * desc)1391 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1392 {
1393 u32 midx;
1394
1395 for (midx = 0; midx < desc->pg_mirror_count; midx++)
1396 nfs_pageio_complete_mirror(desc, midx);
1397
1398 if (desc->pg_error < 0)
1399 nfs_pageio_error_cleanup(desc);
1400 if (desc->pg_ops->pg_cleanup)
1401 desc->pg_ops->pg_cleanup(desc);
1402 nfs_pageio_cleanup_mirroring(desc);
1403 }
1404
1405 /**
1406 * nfs_pageio_cond_complete - Conditional I/O completion
1407 * @desc: pointer to io descriptor
1408 * @index: page index
1409 *
1410 * It is important to ensure that processes don't try to take locks
1411 * on non-contiguous ranges of pages as that might deadlock. This
1412 * function should be called before attempting to wait on a locked
1413 * nfs_page. It will complete the I/O if the page index 'index'
1414 * is not contiguous with the existing list of pages in 'desc'.
1415 */
nfs_pageio_cond_complete(struct nfs_pageio_descriptor * desc,pgoff_t index)1416 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1417 {
1418 struct nfs_pgio_mirror *mirror;
1419 struct nfs_page *prev;
1420 struct folio *folio;
1421 u32 midx;
1422
1423 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1424 mirror = nfs_pgio_get_mirror(desc, midx);
1425 if (!list_empty(&mirror->pg_list)) {
1426 prev = nfs_list_entry(mirror->pg_list.prev);
1427 folio = nfs_page_to_folio(prev);
1428 if (folio) {
1429 if (index == folio_next_index(folio))
1430 continue;
1431 } else if (index == prev->wb_index + 1)
1432 continue;
1433 /*
1434 * We will submit more requests after these. Indicate
1435 * this to the underlying layers.
1436 */
1437 desc->pg_moreio = 1;
1438 nfs_pageio_complete(desc);
1439 break;
1440 }
1441 }
1442 }
1443
1444 /*
1445 * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
1446 */
nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor * pgio)1447 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1448 {
1449 nfs_pageio_complete(pgio);
1450 }
1451
nfs_init_nfspagecache(void)1452 int __init nfs_init_nfspagecache(void)
1453 {
1454 nfs_page_cachep = kmem_cache_create("nfs_page",
1455 sizeof(struct nfs_page),
1456 0, SLAB_HWCACHE_ALIGN,
1457 NULL);
1458 if (nfs_page_cachep == NULL)
1459 return -ENOMEM;
1460
1461 return 0;
1462 }
1463
nfs_destroy_nfspagecache(void)1464 void nfs_destroy_nfspagecache(void)
1465 {
1466 kmem_cache_destroy(nfs_page_cachep);
1467 }
1468
1469 static const struct rpc_call_ops nfs_pgio_common_ops = {
1470 .rpc_call_prepare = nfs_pgio_prepare,
1471 .rpc_call_done = nfs_pgio_result,
1472 .rpc_release = nfs_pgio_release,
1473 };
1474
1475 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1476 .pg_test = nfs_generic_pg_test,
1477 .pg_doio = nfs_generic_pg_pgios,
1478 };
1479