Loading...
1/*
2 * linux/fs/nfs/pagelist.c
3 *
4 * A set of helper functions for managing NFS read and write requests.
5 * The main purpose of these routines is to provide support for the
6 * coalescing of several requests into a single RPC call.
7 *
8 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
9 *
10 */
11
12#include <linux/slab.h>
13#include <linux/file.h>
14#include <linux/sched.h>
15#include <linux/sunrpc/clnt.h>
16#include <linux/nfs.h>
17#include <linux/nfs3.h>
18#include <linux/nfs4.h>
19#include <linux/nfs_page.h>
20#include <linux/nfs_fs.h>
21#include <linux/nfs_mount.h>
22#include <linux/export.h>
23
24#include "internal.h"
25#include "pnfs.h"
26
27static struct kmem_cache *nfs_page_cachep;
28
29bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
30{
31 p->npages = pagecount;
32 if (pagecount <= ARRAY_SIZE(p->page_array))
33 p->pagevec = p->page_array;
34 else {
35 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
36 if (!p->pagevec)
37 p->npages = 0;
38 }
39 return p->pagevec != NULL;
40}
41
42void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
43 struct nfs_pgio_header *hdr,
44 void (*release)(struct nfs_pgio_header *hdr))
45{
46 hdr->req = nfs_list_entry(desc->pg_list.next);
47 hdr->inode = desc->pg_inode;
48 hdr->cred = hdr->req->wb_context->cred;
49 hdr->io_start = req_offset(hdr->req);
50 hdr->good_bytes = desc->pg_count;
51 hdr->dreq = desc->pg_dreq;
52 hdr->release = release;
53 hdr->completion_ops = desc->pg_completion_ops;
54 if (hdr->completion_ops->init_hdr)
55 hdr->completion_ops->init_hdr(hdr);
56}
57
58void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
59{
60 spin_lock(&hdr->lock);
61 if (pos < hdr->io_start + hdr->good_bytes) {
62 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
63 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
64 hdr->good_bytes = pos - hdr->io_start;
65 hdr->error = error;
66 }
67 spin_unlock(&hdr->lock);
68}
69
70static inline struct nfs_page *
71nfs_page_alloc(void)
72{
73 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_KERNEL);
74 if (p)
75 INIT_LIST_HEAD(&p->wb_list);
76 return p;
77}
78
79static inline void
80nfs_page_free(struct nfs_page *p)
81{
82 kmem_cache_free(nfs_page_cachep, p);
83}
84
85/**
86 * nfs_create_request - Create an NFS read/write request.
87 * @ctx: open context to use
88 * @inode: inode to which the request is attached
89 * @page: page to write
90 * @offset: starting offset within the page for the write
91 * @count: number of bytes to read/write
92 *
93 * The page must be locked by the caller. This makes sure we never
94 * create two different requests for the same page.
95 * User should ensure it is safe to sleep in this function.
96 */
97struct nfs_page *
98nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
99 struct page *page,
100 unsigned int offset, unsigned int count)
101{
102 struct nfs_page *req;
103
104 /* try to allocate the request struct */
105 req = nfs_page_alloc();
106 if (req == NULL)
107 return ERR_PTR(-ENOMEM);
108
109 /* get lock context early so we can deal with alloc failures */
110 req->wb_lock_context = nfs_get_lock_context(ctx);
111 if (req->wb_lock_context == NULL) {
112 nfs_page_free(req);
113 return ERR_PTR(-ENOMEM);
114 }
115
116 /* Initialize the request struct. Initially, we assume a
117 * long write-back delay. This will be adjusted in
118 * update_nfs_request below if the region is not locked. */
119 req->wb_page = page;
120 req->wb_index = page->index;
121 page_cache_get(page);
122 req->wb_offset = offset;
123 req->wb_pgbase = offset;
124 req->wb_bytes = count;
125 req->wb_context = get_nfs_open_context(ctx);
126 kref_init(&req->wb_kref);
127 return req;
128}
129
130/**
131 * nfs_unlock_request - Unlock request and wake up sleepers.
132 * @req:
133 */
134void nfs_unlock_request(struct nfs_page *req)
135{
136 if (!NFS_WBACK_BUSY(req)) {
137 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
138 BUG();
139 }
140 smp_mb__before_clear_bit();
141 clear_bit(PG_BUSY, &req->wb_flags);
142 smp_mb__after_clear_bit();
143 wake_up_bit(&req->wb_flags, PG_BUSY);
144}
145
146/**
147 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
148 * @req:
149 */
150void nfs_unlock_and_release_request(struct nfs_page *req)
151{
152 nfs_unlock_request(req);
153 nfs_release_request(req);
154}
155
156/*
157 * nfs_clear_request - Free up all resources allocated to the request
158 * @req:
159 *
160 * Release page and open context resources associated with a read/write
161 * request after it has completed.
162 */
163static void nfs_clear_request(struct nfs_page *req)
164{
165 struct page *page = req->wb_page;
166 struct nfs_open_context *ctx = req->wb_context;
167 struct nfs_lock_context *l_ctx = req->wb_lock_context;
168
169 if (page != NULL) {
170 page_cache_release(page);
171 req->wb_page = NULL;
172 }
173 if (l_ctx != NULL) {
174 nfs_put_lock_context(l_ctx);
175 req->wb_lock_context = NULL;
176 }
177 if (ctx != NULL) {
178 put_nfs_open_context(ctx);
179 req->wb_context = NULL;
180 }
181}
182
183
184/**
185 * nfs_release_request - Release the count on an NFS read/write request
186 * @req: request to release
187 *
188 * Note: Should never be called with the spinlock held!
189 */
190static void nfs_free_request(struct kref *kref)
191{
192 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
193
194 /* Release struct file and open context */
195 nfs_clear_request(req);
196 nfs_page_free(req);
197}
198
199void nfs_release_request(struct nfs_page *req)
200{
201 kref_put(&req->wb_kref, nfs_free_request);
202}
203
204static int nfs_wait_bit_uninterruptible(void *word)
205{
206 io_schedule();
207 return 0;
208}
209
210/**
211 * nfs_wait_on_request - Wait for a request to complete.
212 * @req: request to wait upon.
213 *
214 * Interruptible by fatal signals only.
215 * The user is responsible for holding a count on the request.
216 */
217int
218nfs_wait_on_request(struct nfs_page *req)
219{
220 return wait_on_bit(&req->wb_flags, PG_BUSY,
221 nfs_wait_bit_uninterruptible,
222 TASK_UNINTERRUPTIBLE);
223}
224
225bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req)
226{
227 /*
228 * FIXME: ideally we should be able to coalesce all requests
229 * that are not block boundary aligned, but currently this
230 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
231 * since nfs_flush_multi and nfs_pagein_multi assume you
232 * can have only one struct nfs_page.
233 */
234 if (desc->pg_bsize < PAGE_SIZE)
235 return 0;
236
237 return desc->pg_count + req->wb_bytes <= desc->pg_bsize;
238}
239EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
240
241/**
242 * nfs_pageio_init - initialise a page io descriptor
243 * @desc: pointer to descriptor
244 * @inode: pointer to inode
245 * @doio: pointer to io function
246 * @bsize: io block size
247 * @io_flags: extra parameters for the io function
248 */
249void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
250 struct inode *inode,
251 const struct nfs_pageio_ops *pg_ops,
252 const struct nfs_pgio_completion_ops *compl_ops,
253 size_t bsize,
254 int io_flags)
255{
256 INIT_LIST_HEAD(&desc->pg_list);
257 desc->pg_bytes_written = 0;
258 desc->pg_count = 0;
259 desc->pg_bsize = bsize;
260 desc->pg_base = 0;
261 desc->pg_moreio = 0;
262 desc->pg_recoalesce = 0;
263 desc->pg_inode = inode;
264 desc->pg_ops = pg_ops;
265 desc->pg_completion_ops = compl_ops;
266 desc->pg_ioflags = io_flags;
267 desc->pg_error = 0;
268 desc->pg_lseg = NULL;
269 desc->pg_dreq = NULL;
270}
271
272/**
273 * nfs_can_coalesce_requests - test two requests for compatibility
274 * @prev: pointer to nfs_page
275 * @req: pointer to nfs_page
276 *
277 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
278 * page data area they describe is contiguous, and that their RPC
279 * credentials, NFSv4 open state, and lockowners are the same.
280 *
281 * Return 'true' if this is the case, else return 'false'.
282 */
283static bool nfs_can_coalesce_requests(struct nfs_page *prev,
284 struct nfs_page *req,
285 struct nfs_pageio_descriptor *pgio)
286{
287 if (req->wb_context->cred != prev->wb_context->cred)
288 return false;
289 if (req->wb_lock_context->lockowner != prev->wb_lock_context->lockowner)
290 return false;
291 if (req->wb_context->state != prev->wb_context->state)
292 return false;
293 if (req->wb_pgbase != 0)
294 return false;
295 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
296 return false;
297 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
298 return false;
299 return pgio->pg_ops->pg_test(pgio, prev, req);
300}
301
302/**
303 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
304 * @desc: destination io descriptor
305 * @req: request
306 *
307 * Returns true if the request 'req' was successfully coalesced into the
308 * existing list of pages 'desc'.
309 */
310static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
311 struct nfs_page *req)
312{
313 if (desc->pg_count != 0) {
314 struct nfs_page *prev;
315
316 prev = nfs_list_entry(desc->pg_list.prev);
317 if (!nfs_can_coalesce_requests(prev, req, desc))
318 return 0;
319 } else {
320 if (desc->pg_ops->pg_init)
321 desc->pg_ops->pg_init(desc, req);
322 desc->pg_base = req->wb_pgbase;
323 }
324 nfs_list_remove_request(req);
325 nfs_list_add_request(req, &desc->pg_list);
326 desc->pg_count += req->wb_bytes;
327 return 1;
328}
329
330/*
331 * Helper for nfs_pageio_add_request and nfs_pageio_complete
332 */
333static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
334{
335 if (!list_empty(&desc->pg_list)) {
336 int error = desc->pg_ops->pg_doio(desc);
337 if (error < 0)
338 desc->pg_error = error;
339 else
340 desc->pg_bytes_written += desc->pg_count;
341 }
342 if (list_empty(&desc->pg_list)) {
343 desc->pg_count = 0;
344 desc->pg_base = 0;
345 }
346}
347
348/**
349 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
350 * @desc: destination io descriptor
351 * @req: request
352 *
353 * Returns true if the request 'req' was successfully coalesced into the
354 * existing list of pages 'desc'.
355 */
356static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
357 struct nfs_page *req)
358{
359 while (!nfs_pageio_do_add_request(desc, req)) {
360 desc->pg_moreio = 1;
361 nfs_pageio_doio(desc);
362 if (desc->pg_error < 0)
363 return 0;
364 desc->pg_moreio = 0;
365 if (desc->pg_recoalesce)
366 return 0;
367 }
368 return 1;
369}
370
371static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
372{
373 LIST_HEAD(head);
374
375 do {
376 list_splice_init(&desc->pg_list, &head);
377 desc->pg_bytes_written -= desc->pg_count;
378 desc->pg_count = 0;
379 desc->pg_base = 0;
380 desc->pg_recoalesce = 0;
381
382 while (!list_empty(&head)) {
383 struct nfs_page *req;
384
385 req = list_first_entry(&head, struct nfs_page, wb_list);
386 nfs_list_remove_request(req);
387 if (__nfs_pageio_add_request(desc, req))
388 continue;
389 if (desc->pg_error < 0)
390 return 0;
391 break;
392 }
393 } while (desc->pg_recoalesce);
394 return 1;
395}
396
397int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
398 struct nfs_page *req)
399{
400 int ret;
401
402 do {
403 ret = __nfs_pageio_add_request(desc, req);
404 if (ret)
405 break;
406 if (desc->pg_error < 0)
407 break;
408 ret = nfs_do_recoalesce(desc);
409 } while (ret);
410 return ret;
411}
412
413/**
414 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
415 * @desc: pointer to io descriptor
416 */
417void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
418{
419 for (;;) {
420 nfs_pageio_doio(desc);
421 if (!desc->pg_recoalesce)
422 break;
423 if (!nfs_do_recoalesce(desc))
424 break;
425 }
426}
427
428/**
429 * nfs_pageio_cond_complete - Conditional I/O completion
430 * @desc: pointer to io descriptor
431 * @index: page index
432 *
433 * It is important to ensure that processes don't try to take locks
434 * on non-contiguous ranges of pages as that might deadlock. This
435 * function should be called before attempting to wait on a locked
436 * nfs_page. It will complete the I/O if the page index 'index'
437 * is not contiguous with the existing list of pages in 'desc'.
438 */
439void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
440{
441 if (!list_empty(&desc->pg_list)) {
442 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
443 if (index != prev->wb_index + 1)
444 nfs_pageio_complete(desc);
445 }
446}
447
448int __init nfs_init_nfspagecache(void)
449{
450 nfs_page_cachep = kmem_cache_create("nfs_page",
451 sizeof(struct nfs_page),
452 0, SLAB_HWCACHE_ALIGN,
453 NULL);
454 if (nfs_page_cachep == NULL)
455 return -ENOMEM;
456
457 return 0;
458}
459
460void nfs_destroy_nfspagecache(void)
461{
462 kmem_cache_destroy(nfs_page_cachep);
463}
464
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
33static struct kmem_cache *nfs_page_cachep;
34static const struct rpc_call_ops nfs_pgio_common_ops;
35
36struct nfs_page_iter_page {
37 const struct nfs_page *req;
38 size_t count;
39};
40
41static 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
48static 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
56static 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
72static struct nfs_pgio_mirror *
73nfs_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
80struct nfs_pgio_mirror *
81nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
82{
83 return nfs_pgio_get_mirror(desc, desc->pg_mirror_idx);
84}
85EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
86
87static u32
88nfs_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
95void 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}
117EXPORT_SYMBOL_GPL(nfs_pgheader_init);
118
119void 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
132static 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
141static inline void
142nfs_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 */
154int
155nfs_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 */
170bool
171nfs_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}
188EXPORT_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 */
198int
199nfs_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 */
214void
215nfs_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 */
233int
234nfs_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 */
248void
249nfs_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 */
261static bool
262nfs_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 */
293bool 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 */
310static inline void
311nfs_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 */
351static void
352nfs_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);
370out:
371 /* subrequests must release the ref on the head request */
372 if (head != req)
373 nfs_release_request(head);
374}
375
376static 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
406static 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
415static 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 */
435struct 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 */
466struct 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
485static struct nfs_page *
486nfs_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 */
520void 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 */
533void 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 */
546static 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 */
579void 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
595void nfs_release_request(struct nfs_page *req)
596{
597 kref_put(&req->wb_kref, nfs_page_group_destroy);
598}
599EXPORT_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 */
610size_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}
632EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
633
634struct 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}
644EXPORT_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 */
654static 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 */
666void 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}
671EXPORT_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 */
681static 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 */
723static 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
732int 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 struct nfsd_file *localio)
736{
737 struct rpc_task *task;
738 struct rpc_message msg = {
739 .rpc_argp = &hdr->args,
740 .rpc_resp = &hdr->res,
741 .rpc_cred = cred,
742 };
743 struct rpc_task_setup task_setup_data = {
744 .rpc_client = clnt,
745 .task = &hdr->task,
746 .rpc_message = &msg,
747 .callback_ops = call_ops,
748 .callback_data = hdr,
749 .workqueue = nfsiod_workqueue,
750 .flags = RPC_TASK_ASYNC | flags,
751 };
752
753 if (nfs_server_capable(hdr->inode, NFS_CAP_MOVEABLE))
754 task_setup_data.flags |= RPC_TASK_MOVEABLE;
755
756 hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
757
758 dprintk("NFS: initiated pgio call "
759 "(req %s/%llu, %u bytes @ offset %llu)\n",
760 hdr->inode->i_sb->s_id,
761 (unsigned long long)NFS_FILEID(hdr->inode),
762 hdr->args.count,
763 (unsigned long long)hdr->args.offset);
764
765 if (localio)
766 return nfs_local_doio(NFS_SERVER(hdr->inode)->nfs_client,
767 localio, hdr, call_ops);
768
769 task = rpc_run_task(&task_setup_data);
770 if (IS_ERR(task))
771 return PTR_ERR(task);
772 rpc_put_task(task);
773 return 0;
774}
775EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
776
777/**
778 * nfs_pgio_error - Clean up from a pageio error
779 * @hdr: pageio header
780 */
781static void nfs_pgio_error(struct nfs_pgio_header *hdr)
782{
783 set_bit(NFS_IOHDR_REDO, &hdr->flags);
784 hdr->completion_ops->completion(hdr);
785}
786
787/**
788 * nfs_pgio_release - Release pageio data
789 * @calldata: The pageio header to release
790 */
791static void nfs_pgio_release(void *calldata)
792{
793 struct nfs_pgio_header *hdr = calldata;
794 hdr->completion_ops->completion(hdr);
795}
796
797static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
798 unsigned int bsize)
799{
800 INIT_LIST_HEAD(&mirror->pg_list);
801 mirror->pg_bytes_written = 0;
802 mirror->pg_count = 0;
803 mirror->pg_bsize = bsize;
804 mirror->pg_base = 0;
805 mirror->pg_recoalesce = 0;
806}
807
808/**
809 * nfs_pageio_init - initialise a page io descriptor
810 * @desc: pointer to descriptor
811 * @inode: pointer to inode
812 * @pg_ops: pointer to pageio operations
813 * @compl_ops: pointer to pageio completion operations
814 * @rw_ops: pointer to nfs read/write operations
815 * @bsize: io block size
816 * @io_flags: extra parameters for the io function
817 */
818void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
819 struct inode *inode,
820 const struct nfs_pageio_ops *pg_ops,
821 const struct nfs_pgio_completion_ops *compl_ops,
822 const struct nfs_rw_ops *rw_ops,
823 size_t bsize,
824 int io_flags)
825{
826 desc->pg_moreio = 0;
827 desc->pg_inode = inode;
828 desc->pg_ops = pg_ops;
829 desc->pg_completion_ops = compl_ops;
830 desc->pg_rw_ops = rw_ops;
831 desc->pg_ioflags = io_flags;
832 desc->pg_error = 0;
833 desc->pg_lseg = NULL;
834 desc->pg_io_completion = NULL;
835 desc->pg_dreq = NULL;
836 nfs_netfs_reset_pageio_descriptor(desc);
837 desc->pg_bsize = bsize;
838
839 desc->pg_mirror_count = 1;
840 desc->pg_mirror_idx = 0;
841
842 desc->pg_mirrors_dynamic = NULL;
843 desc->pg_mirrors = desc->pg_mirrors_static;
844 nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
845 desc->pg_maxretrans = 0;
846}
847
848/**
849 * nfs_pgio_result - Basic pageio error handling
850 * @task: The task that ran
851 * @calldata: Pageio header to check
852 */
853static void nfs_pgio_result(struct rpc_task *task, void *calldata)
854{
855 struct nfs_pgio_header *hdr = calldata;
856 struct inode *inode = hdr->inode;
857
858 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
859 return;
860 if (task->tk_status < 0)
861 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
862 else
863 hdr->rw_ops->rw_result(task, hdr);
864}
865
866/*
867 * Create an RPC task for the given read or write request and kick it.
868 * The page must have been locked by the caller.
869 *
870 * It may happen that the page we're passed is not marked dirty.
871 * This is the case if nfs_updatepage detects a conflicting request
872 * that has been written but not committed.
873 */
874int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
875 struct nfs_pgio_header *hdr)
876{
877 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
878
879 struct nfs_page *req;
880 struct page **pages,
881 *last_page;
882 struct list_head *head = &mirror->pg_list;
883 struct nfs_commit_info cinfo;
884 struct nfs_page_array *pg_array = &hdr->page_array;
885 unsigned int pagecount, pageused;
886 unsigned int pg_base = offset_in_page(mirror->pg_base);
887 gfp_t gfp_flags = nfs_io_gfp_mask();
888
889 pagecount = nfs_page_array_len(pg_base, mirror->pg_count);
890 pg_array->npages = pagecount;
891
892 if (pagecount <= ARRAY_SIZE(pg_array->page_array))
893 pg_array->pagevec = pg_array->page_array;
894 else {
895 pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
896 if (!pg_array->pagevec) {
897 pg_array->npages = 0;
898 nfs_pgio_error(hdr);
899 desc->pg_error = -ENOMEM;
900 return desc->pg_error;
901 }
902 }
903
904 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
905 pages = hdr->page_array.pagevec;
906 last_page = NULL;
907 pageused = 0;
908 while (!list_empty(head)) {
909 struct nfs_page_iter_page i;
910 struct page *page;
911
912 req = nfs_list_entry(head->next);
913 nfs_list_move_request(req, &hdr->pages);
914
915 if (req->wb_pgbase == 0)
916 last_page = NULL;
917
918 nfs_page_iter_page_init(&i, req);
919 while ((page = nfs_page_iter_page_get(&i)) != NULL) {
920 if (last_page != page) {
921 pageused++;
922 if (pageused > pagecount)
923 goto full;
924 *pages++ = last_page = page;
925 }
926 }
927 }
928full:
929 if (WARN_ON_ONCE(pageused != pagecount)) {
930 nfs_pgio_error(hdr);
931 desc->pg_error = -EINVAL;
932 return desc->pg_error;
933 }
934
935 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
936 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
937 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
938
939 /* Set up the argument struct */
940 nfs_pgio_rpcsetup(hdr, pg_base, mirror->pg_count, desc->pg_ioflags,
941 &cinfo);
942 desc->pg_rpc_callops = &nfs_pgio_common_ops;
943 return 0;
944}
945EXPORT_SYMBOL_GPL(nfs_generic_pgio);
946
947static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
948{
949 struct nfs_pgio_header *hdr;
950 int ret;
951 unsigned short task_flags = 0;
952
953 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
954 if (!hdr) {
955 desc->pg_error = -ENOMEM;
956 return desc->pg_error;
957 }
958 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
959 ret = nfs_generic_pgio(desc, hdr);
960 if (ret == 0) {
961 struct nfs_client *clp = NFS_SERVER(hdr->inode)->nfs_client;
962
963 struct nfsd_file *localio =
964 nfs_local_open_fh(clp, hdr->cred,
965 hdr->args.fh, hdr->args.context->mode);
966
967 if (NFS_SERVER(hdr->inode)->nfs_client->cl_minorversion)
968 task_flags = RPC_TASK_MOVEABLE;
969 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
970 hdr,
971 hdr->cred,
972 NFS_PROTO(hdr->inode),
973 desc->pg_rpc_callops,
974 desc->pg_ioflags,
975 RPC_TASK_CRED_NOREF | task_flags,
976 localio);
977 }
978 return ret;
979}
980
981static struct nfs_pgio_mirror *
982nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
983 unsigned int mirror_count)
984{
985 struct nfs_pgio_mirror *ret;
986 unsigned int i;
987
988 kfree(desc->pg_mirrors_dynamic);
989 desc->pg_mirrors_dynamic = NULL;
990 if (mirror_count == 1)
991 return desc->pg_mirrors_static;
992 ret = kmalloc_array(mirror_count, sizeof(*ret), nfs_io_gfp_mask());
993 if (ret != NULL) {
994 for (i = 0; i < mirror_count; i++)
995 nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
996 desc->pg_mirrors_dynamic = ret;
997 }
998 return ret;
999}
1000
1001/*
1002 * nfs_pageio_setup_mirroring - determine if mirroring is to be used
1003 * by calling the pg_get_mirror_count op
1004 */
1005static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
1006 struct nfs_page *req)
1007{
1008 unsigned int mirror_count = 1;
1009
1010 if (pgio->pg_ops->pg_get_mirror_count)
1011 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
1012 if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
1013 return;
1014
1015 if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
1016 pgio->pg_error = -EINVAL;
1017 return;
1018 }
1019
1020 pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
1021 if (pgio->pg_mirrors == NULL) {
1022 pgio->pg_error = -ENOMEM;
1023 pgio->pg_mirrors = pgio->pg_mirrors_static;
1024 mirror_count = 1;
1025 }
1026 pgio->pg_mirror_count = mirror_count;
1027}
1028
1029static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
1030{
1031 pgio->pg_mirror_count = 1;
1032 pgio->pg_mirror_idx = 0;
1033 pgio->pg_mirrors = pgio->pg_mirrors_static;
1034 kfree(pgio->pg_mirrors_dynamic);
1035 pgio->pg_mirrors_dynamic = NULL;
1036}
1037
1038static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
1039 const struct nfs_lock_context *l2)
1040{
1041 return l1->lockowner == l2->lockowner;
1042}
1043
1044static bool nfs_page_is_contiguous(const struct nfs_page *prev,
1045 const struct nfs_page *req)
1046{
1047 size_t prev_end = prev->wb_pgbase + prev->wb_bytes;
1048
1049 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
1050 return false;
1051 if (req->wb_pgbase == 0)
1052 return prev_end == nfs_page_max_length(prev);
1053 if (req->wb_pgbase == prev_end) {
1054 struct folio *folio = nfs_page_to_folio(req);
1055 if (folio)
1056 return folio == nfs_page_to_folio(prev);
1057 return req->wb_page == prev->wb_page;
1058 }
1059 return false;
1060}
1061
1062/**
1063 * nfs_coalesce_size - test two requests for compatibility
1064 * @prev: pointer to nfs_page
1065 * @req: pointer to nfs_page
1066 * @pgio: pointer to nfs_pagio_descriptor
1067 *
1068 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
1069 * page data area they describe is contiguous, and that their RPC
1070 * credentials, NFSv4 open state, and lockowners are the same.
1071 *
1072 * Returns size of the request that can be coalesced
1073 */
1074static unsigned int nfs_coalesce_size(struct nfs_page *prev,
1075 struct nfs_page *req,
1076 struct nfs_pageio_descriptor *pgio)
1077{
1078 struct file_lock_context *flctx;
1079
1080 if (prev) {
1081 if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
1082 return 0;
1083 flctx = locks_inode_context(d_inode(nfs_req_openctx(req)->dentry));
1084 if (flctx != NULL &&
1085 !(list_empty_careful(&flctx->flc_posix) &&
1086 list_empty_careful(&flctx->flc_flock)) &&
1087 !nfs_match_lock_context(req->wb_lock_context,
1088 prev->wb_lock_context))
1089 return 0;
1090 if (!nfs_page_is_contiguous(prev, req))
1091 return 0;
1092 }
1093 return pgio->pg_ops->pg_test(pgio, prev, req);
1094}
1095
1096/**
1097 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
1098 * @desc: destination io descriptor
1099 * @req: request
1100 *
1101 * If the request 'req' was successfully coalesced into the existing list
1102 * of pages 'desc', it returns the size of req.
1103 */
1104static unsigned int
1105nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
1106 struct nfs_page *req)
1107{
1108 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1109 struct nfs_page *prev = NULL;
1110 unsigned int size;
1111
1112 if (list_empty(&mirror->pg_list)) {
1113 if (desc->pg_ops->pg_init)
1114 desc->pg_ops->pg_init(desc, req);
1115 if (desc->pg_error < 0)
1116 return 0;
1117 mirror->pg_base = req->wb_pgbase;
1118 mirror->pg_count = 0;
1119 mirror->pg_recoalesce = 0;
1120 } else
1121 prev = nfs_list_entry(mirror->pg_list.prev);
1122
1123 if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
1124 if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
1125 desc->pg_error = -ETIMEDOUT;
1126 else
1127 desc->pg_error = -EIO;
1128 return 0;
1129 }
1130
1131 size = nfs_coalesce_size(prev, req, desc);
1132 if (size < req->wb_bytes)
1133 return size;
1134 nfs_list_move_request(req, &mirror->pg_list);
1135 mirror->pg_count += req->wb_bytes;
1136 return req->wb_bytes;
1137}
1138
1139/*
1140 * Helper for nfs_pageio_add_request and nfs_pageio_complete
1141 */
1142static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
1143{
1144 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1145
1146 if (!list_empty(&mirror->pg_list)) {
1147 int error = desc->pg_ops->pg_doio(desc);
1148 if (error < 0)
1149 desc->pg_error = error;
1150 if (list_empty(&mirror->pg_list))
1151 mirror->pg_bytes_written += mirror->pg_count;
1152 }
1153}
1154
1155static void
1156nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1157 struct nfs_page *req)
1158{
1159 LIST_HEAD(head);
1160
1161 nfs_list_move_request(req, &head);
1162 desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1163}
1164
1165/**
1166 * __nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1167 * @desc: destination io descriptor
1168 * @req: request
1169 *
1170 * This may split a request into subrequests which are all part of the
1171 * same page group. If so, it will submit @req as the last one, to ensure
1172 * the pointer to @req is still valid in case of failure.
1173 *
1174 * Returns true if the request 'req' was successfully coalesced into the
1175 * existing list of pages 'desc'.
1176 */
1177static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1178 struct nfs_page *req)
1179{
1180 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1181 struct nfs_page *subreq;
1182 unsigned int size, subreq_size;
1183
1184 nfs_page_group_lock(req);
1185
1186 subreq = req;
1187 subreq_size = subreq->wb_bytes;
1188 for(;;) {
1189 size = nfs_pageio_do_add_request(desc, subreq);
1190 if (size == subreq_size) {
1191 /* We successfully submitted a request */
1192 if (subreq == req)
1193 break;
1194 req->wb_pgbase += size;
1195 req->wb_bytes -= size;
1196 req->wb_offset += size;
1197 subreq_size = req->wb_bytes;
1198 subreq = req;
1199 continue;
1200 }
1201 if (WARN_ON_ONCE(subreq != req)) {
1202 nfs_page_group_unlock(req);
1203 nfs_pageio_cleanup_request(desc, subreq);
1204 subreq = req;
1205 subreq_size = req->wb_bytes;
1206 nfs_page_group_lock(req);
1207 }
1208 if (!size) {
1209 /* Can't coalesce any more, so do I/O */
1210 nfs_page_group_unlock(req);
1211 desc->pg_moreio = 1;
1212 nfs_pageio_doio(desc);
1213 if (desc->pg_error < 0 || mirror->pg_recoalesce)
1214 return 0;
1215 /* retry add_request for this subreq */
1216 nfs_page_group_lock(req);
1217 continue;
1218 }
1219 subreq = nfs_create_subreq(req, req->wb_pgbase,
1220 req->wb_offset, size);
1221 if (IS_ERR(subreq))
1222 goto err_ptr;
1223 subreq_size = size;
1224 }
1225
1226 nfs_page_group_unlock(req);
1227 return 1;
1228err_ptr:
1229 desc->pg_error = PTR_ERR(subreq);
1230 nfs_page_group_unlock(req);
1231 return 0;
1232}
1233
1234static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1235{
1236 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1237 LIST_HEAD(head);
1238
1239 do {
1240 list_splice_init(&mirror->pg_list, &head);
1241 mirror->pg_recoalesce = 0;
1242
1243 while (!list_empty(&head)) {
1244 struct nfs_page *req;
1245
1246 req = list_first_entry(&head, struct nfs_page, wb_list);
1247 if (__nfs_pageio_add_request(desc, req))
1248 continue;
1249 if (desc->pg_error < 0) {
1250 list_splice_tail(&head, &mirror->pg_list);
1251 mirror->pg_recoalesce = 1;
1252 return 0;
1253 }
1254 break;
1255 }
1256 } while (mirror->pg_recoalesce);
1257 return 1;
1258}
1259
1260static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1261 struct nfs_page *req)
1262{
1263 int ret;
1264
1265 do {
1266 ret = __nfs_pageio_add_request(desc, req);
1267 if (ret)
1268 break;
1269 if (desc->pg_error < 0)
1270 break;
1271 ret = nfs_do_recoalesce(desc);
1272 } while (ret);
1273
1274 return ret;
1275}
1276
1277static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1278{
1279 u32 midx;
1280 struct nfs_pgio_mirror *mirror;
1281
1282 if (!desc->pg_error)
1283 return;
1284
1285 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1286 mirror = nfs_pgio_get_mirror(desc, midx);
1287 desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1288 desc->pg_error);
1289 }
1290}
1291
1292int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1293 struct nfs_page *req)
1294{
1295 u32 midx;
1296 unsigned int pgbase, offset, bytes;
1297 struct nfs_page *dupreq;
1298
1299 pgbase = req->wb_pgbase;
1300 offset = req->wb_offset;
1301 bytes = req->wb_bytes;
1302
1303 nfs_pageio_setup_mirroring(desc, req);
1304 if (desc->pg_error < 0)
1305 goto out_failed;
1306
1307 /* Create the mirror instances first, and fire them off */
1308 for (midx = 1; midx < desc->pg_mirror_count; midx++) {
1309 nfs_page_group_lock(req);
1310
1311 dupreq = nfs_create_subreq(req,
1312 pgbase, offset, bytes);
1313
1314 nfs_page_group_unlock(req);
1315 if (IS_ERR(dupreq)) {
1316 desc->pg_error = PTR_ERR(dupreq);
1317 goto out_failed;
1318 }
1319
1320 nfs_pgio_set_current_mirror(desc, midx);
1321 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1322 goto out_cleanup_subreq;
1323 }
1324
1325 nfs_pgio_set_current_mirror(desc, 0);
1326 if (!nfs_pageio_add_request_mirror(desc, req))
1327 goto out_failed;
1328
1329 return 1;
1330
1331out_cleanup_subreq:
1332 nfs_pageio_cleanup_request(desc, dupreq);
1333out_failed:
1334 nfs_pageio_error_cleanup(desc);
1335 return 0;
1336}
1337
1338/*
1339 * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1340 * nfs_pageio_descriptor
1341 * @desc: pointer to io descriptor
1342 * @mirror_idx: pointer to mirror index
1343 */
1344static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1345 u32 mirror_idx)
1346{
1347 struct nfs_pgio_mirror *mirror;
1348 u32 restore_idx;
1349
1350 restore_idx = nfs_pgio_set_current_mirror(desc, mirror_idx);
1351 mirror = nfs_pgio_current_mirror(desc);
1352
1353 for (;;) {
1354 nfs_pageio_doio(desc);
1355 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1356 break;
1357 if (!nfs_do_recoalesce(desc))
1358 break;
1359 }
1360 nfs_pgio_set_current_mirror(desc, restore_idx);
1361}
1362
1363/*
1364 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1365 * @hdr - the pgio header to move request from
1366 * @desc - the pageio descriptor to add requests to
1367 *
1368 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1369 * to send them.
1370 *
1371 * Returns 0 on success and < 0 on error.
1372 */
1373int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1374 struct nfs_pgio_header *hdr)
1375{
1376 LIST_HEAD(pages);
1377
1378 desc->pg_io_completion = hdr->io_completion;
1379 desc->pg_dreq = hdr->dreq;
1380 nfs_netfs_set_pageio_descriptor(desc, hdr);
1381 list_splice_init(&hdr->pages, &pages);
1382 while (!list_empty(&pages)) {
1383 struct nfs_page *req = nfs_list_entry(pages.next);
1384
1385 if (!nfs_pageio_add_request(desc, req))
1386 break;
1387 }
1388 nfs_pageio_complete(desc);
1389 if (!list_empty(&pages)) {
1390 int err = desc->pg_error < 0 ? desc->pg_error : -EIO;
1391 hdr->completion_ops->error_cleanup(&pages, err);
1392 nfs_set_pgio_error(hdr, err, hdr->io_start);
1393 return err;
1394 }
1395 return 0;
1396}
1397EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1398
1399/**
1400 * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1401 * @desc: pointer to io descriptor
1402 */
1403void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1404{
1405 u32 midx;
1406
1407 for (midx = 0; midx < desc->pg_mirror_count; midx++)
1408 nfs_pageio_complete_mirror(desc, midx);
1409
1410 if (desc->pg_error < 0)
1411 nfs_pageio_error_cleanup(desc);
1412 if (desc->pg_ops->pg_cleanup)
1413 desc->pg_ops->pg_cleanup(desc);
1414 nfs_pageio_cleanup_mirroring(desc);
1415}
1416
1417/**
1418 * nfs_pageio_cond_complete - Conditional I/O completion
1419 * @desc: pointer to io descriptor
1420 * @index: page index
1421 *
1422 * It is important to ensure that processes don't try to take locks
1423 * on non-contiguous ranges of pages as that might deadlock. This
1424 * function should be called before attempting to wait on a locked
1425 * nfs_page. It will complete the I/O if the page index 'index'
1426 * is not contiguous with the existing list of pages in 'desc'.
1427 */
1428void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1429{
1430 struct nfs_pgio_mirror *mirror;
1431 struct nfs_page *prev;
1432 struct folio *folio;
1433 u32 midx;
1434
1435 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1436 mirror = nfs_pgio_get_mirror(desc, midx);
1437 if (!list_empty(&mirror->pg_list)) {
1438 prev = nfs_list_entry(mirror->pg_list.prev);
1439 folio = nfs_page_to_folio(prev);
1440 if (folio) {
1441 if (index == folio_next_index(folio))
1442 continue;
1443 } else if (index == prev->wb_index + 1)
1444 continue;
1445 /*
1446 * We will submit more requests after these. Indicate
1447 * this to the underlying layers.
1448 */
1449 desc->pg_moreio = 1;
1450 nfs_pageio_complete(desc);
1451 break;
1452 }
1453 }
1454}
1455
1456/*
1457 * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
1458 */
1459void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1460{
1461 nfs_pageio_complete(pgio);
1462}
1463
1464int __init nfs_init_nfspagecache(void)
1465{
1466 nfs_page_cachep = kmem_cache_create("nfs_page",
1467 sizeof(struct nfs_page),
1468 0, SLAB_HWCACHE_ALIGN,
1469 NULL);
1470 if (nfs_page_cachep == NULL)
1471 return -ENOMEM;
1472
1473 return 0;
1474}
1475
1476void nfs_destroy_nfspagecache(void)
1477{
1478 kmem_cache_destroy(nfs_page_cachep);
1479}
1480
1481static const struct rpc_call_ops nfs_pgio_common_ops = {
1482 .rpc_call_prepare = nfs_pgio_prepare,
1483 .rpc_call_done = nfs_pgio_result,
1484 .rpc_release = nfs_pgio_release,
1485};
1486
1487const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1488 .pg_test = nfs_generic_pg_test,
1489 .pg_doio = nfs_generic_pg_pgios,
1490};