Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* handling of writes to regular files and writing back to the server
3 *
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#include <linux/backing-dev.h>
9#include <linux/slab.h>
10#include <linux/fs.h>
11#include <linux/pagemap.h>
12#include <linux/writeback.h>
13#include <linux/pagevec.h>
14#include <linux/netfs.h>
15#include <trace/events/netfs.h>
16#include "internal.h"
17
18/*
19 * completion of write to server
20 */
21static void afs_pages_written_back(struct afs_vnode *vnode, loff_t start, unsigned int len)
22{
23 _enter("{%llx:%llu},{%x @%llx}",
24 vnode->fid.vid, vnode->fid.vnode, len, start);
25
26 afs_prune_wb_keys(vnode);
27 _leave("");
28}
29
30/*
31 * Find a key to use for the writeback. We cached the keys used to author the
32 * writes on the vnode. wreq->netfs_priv2 will contain the last writeback key
33 * record used or NULL and we need to start from there if it's set.
34 * wreq->netfs_priv will be set to the key itself or NULL.
35 */
36static void afs_get_writeback_key(struct netfs_io_request *wreq)
37{
38 struct afs_wb_key *wbk, *old = wreq->netfs_priv2;
39 struct afs_vnode *vnode = AFS_FS_I(wreq->inode);
40
41 key_put(wreq->netfs_priv);
42 wreq->netfs_priv = NULL;
43 wreq->netfs_priv2 = NULL;
44
45 spin_lock(&vnode->wb_lock);
46 if (old)
47 wbk = list_next_entry(old, vnode_link);
48 else
49 wbk = list_first_entry(&vnode->wb_keys, struct afs_wb_key, vnode_link);
50
51 list_for_each_entry_from(wbk, &vnode->wb_keys, vnode_link) {
52 _debug("wbk %u", key_serial(wbk->key));
53 if (key_validate(wbk->key) == 0) {
54 refcount_inc(&wbk->usage);
55 wreq->netfs_priv = key_get(wbk->key);
56 wreq->netfs_priv2 = wbk;
57 _debug("USE WB KEY %u", key_serial(wbk->key));
58 break;
59 }
60 }
61
62 spin_unlock(&vnode->wb_lock);
63
64 afs_put_wb_key(old);
65}
66
67static void afs_store_data_success(struct afs_operation *op)
68{
69 struct afs_vnode *vnode = op->file[0].vnode;
70
71 op->ctime = op->file[0].scb.status.mtime_client;
72 afs_vnode_commit_status(op, &op->file[0]);
73 if (!afs_op_error(op)) {
74 afs_pages_written_back(vnode, op->store.pos, op->store.size);
75 afs_stat_v(vnode, n_stores);
76 atomic_long_add(op->store.size, &afs_v2net(vnode)->n_store_bytes);
77 }
78}
79
80static const struct afs_operation_ops afs_store_data_operation = {
81 .issue_afs_rpc = afs_fs_store_data,
82 .issue_yfs_rpc = yfs_fs_store_data,
83 .success = afs_store_data_success,
84};
85
86/*
87 * Prepare a subrequest to write to the server. This sets the max_len
88 * parameter.
89 */
90void afs_prepare_write(struct netfs_io_subrequest *subreq)
91{
92 struct netfs_io_stream *stream = &subreq->rreq->io_streams[subreq->stream_nr];
93
94 //if (test_bit(NETFS_SREQ_RETRYING, &subreq->flags))
95 // subreq->max_len = 512 * 1024;
96 //else
97 stream->sreq_max_len = 256 * 1024 * 1024;
98}
99
100/*
101 * Issue a subrequest to write to the server.
102 */
103static void afs_issue_write_worker(struct work_struct *work)
104{
105 struct netfs_io_subrequest *subreq = container_of(work, struct netfs_io_subrequest, work);
106 struct netfs_io_request *wreq = subreq->rreq;
107 struct afs_operation *op;
108 struct afs_vnode *vnode = AFS_FS_I(wreq->inode);
109 unsigned long long pos = subreq->start + subreq->transferred;
110 size_t len = subreq->len - subreq->transferred;
111 int ret = -ENOKEY;
112
113 _enter("R=%x[%x],%s{%llx:%llu.%u},%llx,%zx",
114 wreq->debug_id, subreq->debug_index,
115 vnode->volume->name,
116 vnode->fid.vid,
117 vnode->fid.vnode,
118 vnode->fid.unique,
119 pos, len);
120
121#if 0 // Error injection
122 if (subreq->debug_index == 3)
123 return netfs_write_subrequest_terminated(subreq, -ENOANO, false);
124
125 if (!subreq->retry_count) {
126 set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
127 return netfs_write_subrequest_terminated(subreq, -EAGAIN, false);
128 }
129#endif
130
131 op = afs_alloc_operation(wreq->netfs_priv, vnode->volume);
132 if (IS_ERR(op))
133 return netfs_write_subrequest_terminated(subreq, -EAGAIN, false);
134
135 afs_op_set_vnode(op, 0, vnode);
136 op->file[0].dv_delta = 1;
137 op->file[0].modification = true;
138 op->store.pos = pos;
139 op->store.size = len;
140 op->flags |= AFS_OPERATION_UNINTR;
141 op->ops = &afs_store_data_operation;
142
143 afs_begin_vnode_operation(op);
144
145 op->store.write_iter = &subreq->io_iter;
146 op->store.i_size = umax(pos + len, vnode->netfs.remote_i_size);
147 op->mtime = inode_get_mtime(&vnode->netfs.inode);
148
149 afs_wait_for_operation(op);
150 ret = afs_put_operation(op);
151 switch (ret) {
152 case 0:
153 __set_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
154 break;
155 case -EACCES:
156 case -EPERM:
157 case -ENOKEY:
158 case -EKEYEXPIRED:
159 case -EKEYREJECTED:
160 case -EKEYREVOKED:
161 /* If there are more keys we can try, use the retry algorithm
162 * to rotate the keys.
163 */
164 if (wreq->netfs_priv2)
165 set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
166 break;
167 }
168
169 netfs_write_subrequest_terminated(subreq, ret < 0 ? ret : subreq->len, false);
170}
171
172void afs_issue_write(struct netfs_io_subrequest *subreq)
173{
174 subreq->work.func = afs_issue_write_worker;
175 if (!queue_work(system_unbound_wq, &subreq->work))
176 WARN_ON_ONCE(1);
177}
178
179/*
180 * Writeback calls this when it finds a folio that needs uploading. This isn't
181 * called if writeback only has copy-to-cache to deal with.
182 */
183void afs_begin_writeback(struct netfs_io_request *wreq)
184{
185 afs_get_writeback_key(wreq);
186 wreq->io_streams[0].avail = true;
187}
188
189/*
190 * Prepare to retry the writes in request. Use this to try rotating the
191 * available writeback keys.
192 */
193void afs_retry_request(struct netfs_io_request *wreq, struct netfs_io_stream *stream)
194{
195 struct netfs_io_subrequest *subreq =
196 list_first_entry(&stream->subrequests,
197 struct netfs_io_subrequest, rreq_link);
198
199 switch (subreq->error) {
200 case -EACCES:
201 case -EPERM:
202 case -ENOKEY:
203 case -EKEYEXPIRED:
204 case -EKEYREJECTED:
205 case -EKEYREVOKED:
206 afs_get_writeback_key(wreq);
207 if (!wreq->netfs_priv)
208 stream->failed = true;
209 break;
210 }
211}
212
213/*
214 * write some of the pending data back to the server
215 */
216int afs_writepages(struct address_space *mapping, struct writeback_control *wbc)
217{
218 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
219 int ret;
220
221 /* We have to be careful as we can end up racing with setattr()
222 * truncating the pagecache since the caller doesn't take a lock here
223 * to prevent it.
224 */
225 if (wbc->sync_mode == WB_SYNC_ALL)
226 down_read(&vnode->validate_lock);
227 else if (!down_read_trylock(&vnode->validate_lock))
228 return 0;
229
230 ret = netfs_writepages(mapping, wbc);
231 up_read(&vnode->validate_lock);
232 return ret;
233}
234
235/*
236 * flush any dirty pages for this process, and check for write errors.
237 * - the return status from this call provides a reliable indication of
238 * whether any write errors occurred for this process.
239 */
240int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
241{
242 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
243 struct afs_file *af = file->private_data;
244 int ret;
245
246 _enter("{%llx:%llu},{n=%pD},%d",
247 vnode->fid.vid, vnode->fid.vnode, file,
248 datasync);
249
250 ret = afs_validate(vnode, af->key);
251 if (ret < 0)
252 return ret;
253
254 return file_write_and_wait_range(file, start, end);
255}
256
257/*
258 * notification that a previously read-only page is about to become writable
259 * - if it returns an error, the caller will deliver a bus error signal
260 */
261vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
262{
263 struct file *file = vmf->vma->vm_file;
264
265 if (afs_validate(AFS_FS_I(file_inode(file)), afs_file_key(file)) < 0)
266 return VM_FAULT_SIGBUS;
267 return netfs_page_mkwrite(vmf, NULL);
268}
269
270/*
271 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
272 */
273void afs_prune_wb_keys(struct afs_vnode *vnode)
274{
275 LIST_HEAD(graveyard);
276 struct afs_wb_key *wbk, *tmp;
277
278 /* Discard unused keys */
279 spin_lock(&vnode->wb_lock);
280
281 if (!mapping_tagged(&vnode->netfs.inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
282 !mapping_tagged(&vnode->netfs.inode.i_data, PAGECACHE_TAG_DIRTY)) {
283 list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
284 if (refcount_read(&wbk->usage) == 1)
285 list_move(&wbk->vnode_link, &graveyard);
286 }
287 }
288
289 spin_unlock(&vnode->wb_lock);
290
291 while (!list_empty(&graveyard)) {
292 wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
293 list_del(&wbk->vnode_link);
294 afs_put_wb_key(wbk);
295 }
296}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* handling of writes to regular files and writing back to the server
3 *
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#include <linux/backing-dev.h>
9#include <linux/slab.h>
10#include <linux/fs.h>
11#include <linux/pagemap.h>
12#include <linux/writeback.h>
13#include <linux/pagevec.h>
14#include <linux/netfs.h>
15#include <linux/fscache.h>
16#include "internal.h"
17
18/*
19 * mark a page as having been made dirty and thus needing writeback
20 */
21int afs_set_page_dirty(struct page *page)
22{
23 _enter("");
24 return __set_page_dirty_nobuffers(page);
25}
26
27/*
28 * prepare to perform part of a write to a page
29 */
30int afs_write_begin(struct file *file, struct address_space *mapping,
31 loff_t pos, unsigned len, unsigned flags,
32 struct page **_page, void **fsdata)
33{
34 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
35 struct page *page;
36 unsigned long priv;
37 unsigned f, from;
38 unsigned t, to;
39 pgoff_t index;
40 int ret;
41
42 _enter("{%llx:%llu},%llx,%x",
43 vnode->fid.vid, vnode->fid.vnode, pos, len);
44
45 /* Prefetch area to be written into the cache if we're caching this
46 * file. We need to do this before we get a lock on the page in case
47 * there's more than one writer competing for the same cache block.
48 */
49 ret = netfs_write_begin(file, mapping, pos, len, flags, &page, fsdata,
50 &afs_req_ops, NULL);
51 if (ret < 0)
52 return ret;
53
54 index = page->index;
55 from = pos - index * PAGE_SIZE;
56 to = from + len;
57
58try_again:
59 /* See if this page is already partially written in a way that we can
60 * merge the new write with.
61 */
62 if (PagePrivate(page)) {
63 priv = page_private(page);
64 f = afs_page_dirty_from(page, priv);
65 t = afs_page_dirty_to(page, priv);
66 ASSERTCMP(f, <=, t);
67
68 if (PageWriteback(page)) {
69 trace_afs_page_dirty(vnode, tracepoint_string("alrdy"), page);
70 goto flush_conflicting_write;
71 }
72 /* If the file is being filled locally, allow inter-write
73 * spaces to be merged into writes. If it's not, only write
74 * back what the user gives us.
75 */
76 if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
77 (to < f || from > t))
78 goto flush_conflicting_write;
79 }
80
81 *_page = page;
82 _leave(" = 0");
83 return 0;
84
85 /* The previous write and this write aren't adjacent or overlapping, so
86 * flush the page out.
87 */
88flush_conflicting_write:
89 _debug("flush conflict");
90 ret = write_one_page(page);
91 if (ret < 0)
92 goto error;
93
94 ret = lock_page_killable(page);
95 if (ret < 0)
96 goto error;
97 goto try_again;
98
99error:
100 put_page(page);
101 _leave(" = %d", ret);
102 return ret;
103}
104
105/*
106 * finalise part of a write to a page
107 */
108int afs_write_end(struct file *file, struct address_space *mapping,
109 loff_t pos, unsigned len, unsigned copied,
110 struct page *page, void *fsdata)
111{
112 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
113 unsigned long priv;
114 unsigned int f, from = pos & (thp_size(page) - 1);
115 unsigned int t, to = from + copied;
116 loff_t i_size, maybe_i_size;
117
118 _enter("{%llx:%llu},{%lx}",
119 vnode->fid.vid, vnode->fid.vnode, page->index);
120
121 if (!PageUptodate(page)) {
122 if (copied < len) {
123 copied = 0;
124 goto out;
125 }
126
127 SetPageUptodate(page);
128 }
129
130 if (copied == 0)
131 goto out;
132
133 maybe_i_size = pos + copied;
134
135 i_size = i_size_read(&vnode->vfs_inode);
136 if (maybe_i_size > i_size) {
137 write_seqlock(&vnode->cb_lock);
138 i_size = i_size_read(&vnode->vfs_inode);
139 if (maybe_i_size > i_size)
140 afs_set_i_size(vnode, maybe_i_size);
141 write_sequnlock(&vnode->cb_lock);
142 }
143
144 if (PagePrivate(page)) {
145 priv = page_private(page);
146 f = afs_page_dirty_from(page, priv);
147 t = afs_page_dirty_to(page, priv);
148 if (from < f)
149 f = from;
150 if (to > t)
151 t = to;
152 priv = afs_page_dirty(page, f, t);
153 set_page_private(page, priv);
154 trace_afs_page_dirty(vnode, tracepoint_string("dirty+"), page);
155 } else {
156 priv = afs_page_dirty(page, from, to);
157 attach_page_private(page, (void *)priv);
158 trace_afs_page_dirty(vnode, tracepoint_string("dirty"), page);
159 }
160
161 if (set_page_dirty(page))
162 _debug("dirtied %lx", page->index);
163
164out:
165 unlock_page(page);
166 put_page(page);
167 return copied;
168}
169
170/*
171 * kill all the pages in the given range
172 */
173static void afs_kill_pages(struct address_space *mapping,
174 loff_t start, loff_t len)
175{
176 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
177 struct pagevec pv;
178 unsigned int loop, psize;
179
180 _enter("{%llx:%llu},%llx @%llx",
181 vnode->fid.vid, vnode->fid.vnode, len, start);
182
183 pagevec_init(&pv);
184
185 do {
186 _debug("kill %llx @%llx", len, start);
187
188 pv.nr = find_get_pages_contig(mapping, start / PAGE_SIZE,
189 PAGEVEC_SIZE, pv.pages);
190 if (pv.nr == 0)
191 break;
192
193 for (loop = 0; loop < pv.nr; loop++) {
194 struct page *page = pv.pages[loop];
195
196 if (page->index * PAGE_SIZE >= start + len)
197 break;
198
199 psize = thp_size(page);
200 start += psize;
201 len -= psize;
202 ClearPageUptodate(page);
203 end_page_writeback(page);
204 lock_page(page);
205 generic_error_remove_page(mapping, page);
206 unlock_page(page);
207 }
208
209 __pagevec_release(&pv);
210 } while (len > 0);
211
212 _leave("");
213}
214
215/*
216 * Redirty all the pages in a given range.
217 */
218static void afs_redirty_pages(struct writeback_control *wbc,
219 struct address_space *mapping,
220 loff_t start, loff_t len)
221{
222 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
223 struct pagevec pv;
224 unsigned int loop, psize;
225
226 _enter("{%llx:%llu},%llx @%llx",
227 vnode->fid.vid, vnode->fid.vnode, len, start);
228
229 pagevec_init(&pv);
230
231 do {
232 _debug("redirty %llx @%llx", len, start);
233
234 pv.nr = find_get_pages_contig(mapping, start / PAGE_SIZE,
235 PAGEVEC_SIZE, pv.pages);
236 if (pv.nr == 0)
237 break;
238
239 for (loop = 0; loop < pv.nr; loop++) {
240 struct page *page = pv.pages[loop];
241
242 if (page->index * PAGE_SIZE >= start + len)
243 break;
244
245 psize = thp_size(page);
246 start += psize;
247 len -= psize;
248 redirty_page_for_writepage(wbc, page);
249 end_page_writeback(page);
250 }
251
252 __pagevec_release(&pv);
253 } while (len > 0);
254
255 _leave("");
256}
257
258/*
259 * completion of write to server
260 */
261static void afs_pages_written_back(struct afs_vnode *vnode, loff_t start, unsigned int len)
262{
263 struct address_space *mapping = vnode->vfs_inode.i_mapping;
264 struct page *page;
265 pgoff_t end;
266
267 XA_STATE(xas, &mapping->i_pages, start / PAGE_SIZE);
268
269 _enter("{%llx:%llu},{%x @%llx}",
270 vnode->fid.vid, vnode->fid.vnode, len, start);
271
272 rcu_read_lock();
273
274 end = (start + len - 1) / PAGE_SIZE;
275 xas_for_each(&xas, page, end) {
276 if (!PageWriteback(page)) {
277 kdebug("bad %x @%llx page %lx %lx", len, start, page->index, end);
278 ASSERT(PageWriteback(page));
279 }
280
281 trace_afs_page_dirty(vnode, tracepoint_string("clear"), page);
282 detach_page_private(page);
283 page_endio(page, true, 0);
284 }
285
286 rcu_read_unlock();
287
288 afs_prune_wb_keys(vnode);
289 _leave("");
290}
291
292/*
293 * Find a key to use for the writeback. We cached the keys used to author the
294 * writes on the vnode. *_wbk will contain the last writeback key used or NULL
295 * and we need to start from there if it's set.
296 */
297static int afs_get_writeback_key(struct afs_vnode *vnode,
298 struct afs_wb_key **_wbk)
299{
300 struct afs_wb_key *wbk = NULL;
301 struct list_head *p;
302 int ret = -ENOKEY, ret2;
303
304 spin_lock(&vnode->wb_lock);
305 if (*_wbk)
306 p = (*_wbk)->vnode_link.next;
307 else
308 p = vnode->wb_keys.next;
309
310 while (p != &vnode->wb_keys) {
311 wbk = list_entry(p, struct afs_wb_key, vnode_link);
312 _debug("wbk %u", key_serial(wbk->key));
313 ret2 = key_validate(wbk->key);
314 if (ret2 == 0) {
315 refcount_inc(&wbk->usage);
316 _debug("USE WB KEY %u", key_serial(wbk->key));
317 break;
318 }
319
320 wbk = NULL;
321 if (ret == -ENOKEY)
322 ret = ret2;
323 p = p->next;
324 }
325
326 spin_unlock(&vnode->wb_lock);
327 if (*_wbk)
328 afs_put_wb_key(*_wbk);
329 *_wbk = wbk;
330 return 0;
331}
332
333static void afs_store_data_success(struct afs_operation *op)
334{
335 struct afs_vnode *vnode = op->file[0].vnode;
336
337 op->ctime = op->file[0].scb.status.mtime_client;
338 afs_vnode_commit_status(op, &op->file[0]);
339 if (op->error == 0) {
340 if (!op->store.laundering)
341 afs_pages_written_back(vnode, op->store.pos, op->store.size);
342 afs_stat_v(vnode, n_stores);
343 atomic_long_add(op->store.size, &afs_v2net(vnode)->n_store_bytes);
344 }
345}
346
347static const struct afs_operation_ops afs_store_data_operation = {
348 .issue_afs_rpc = afs_fs_store_data,
349 .issue_yfs_rpc = yfs_fs_store_data,
350 .success = afs_store_data_success,
351};
352
353/*
354 * write to a file
355 */
356static int afs_store_data(struct afs_vnode *vnode, struct iov_iter *iter, loff_t pos,
357 bool laundering)
358{
359 struct afs_operation *op;
360 struct afs_wb_key *wbk = NULL;
361 loff_t size = iov_iter_count(iter), i_size;
362 int ret = -ENOKEY;
363
364 _enter("%s{%llx:%llu.%u},%llx,%llx",
365 vnode->volume->name,
366 vnode->fid.vid,
367 vnode->fid.vnode,
368 vnode->fid.unique,
369 size, pos);
370
371 ret = afs_get_writeback_key(vnode, &wbk);
372 if (ret) {
373 _leave(" = %d [no keys]", ret);
374 return ret;
375 }
376
377 op = afs_alloc_operation(wbk->key, vnode->volume);
378 if (IS_ERR(op)) {
379 afs_put_wb_key(wbk);
380 return -ENOMEM;
381 }
382
383 i_size = i_size_read(&vnode->vfs_inode);
384
385 afs_op_set_vnode(op, 0, vnode);
386 op->file[0].dv_delta = 1;
387 op->file[0].modification = true;
388 op->store.write_iter = iter;
389 op->store.pos = pos;
390 op->store.size = size;
391 op->store.i_size = max(pos + size, i_size);
392 op->store.laundering = laundering;
393 op->mtime = vnode->vfs_inode.i_mtime;
394 op->flags |= AFS_OPERATION_UNINTR;
395 op->ops = &afs_store_data_operation;
396
397try_next_key:
398 afs_begin_vnode_operation(op);
399 afs_wait_for_operation(op);
400
401 switch (op->error) {
402 case -EACCES:
403 case -EPERM:
404 case -ENOKEY:
405 case -EKEYEXPIRED:
406 case -EKEYREJECTED:
407 case -EKEYREVOKED:
408 _debug("next");
409
410 ret = afs_get_writeback_key(vnode, &wbk);
411 if (ret == 0) {
412 key_put(op->key);
413 op->key = key_get(wbk->key);
414 goto try_next_key;
415 }
416 break;
417 }
418
419 afs_put_wb_key(wbk);
420 _leave(" = %d", op->error);
421 return afs_put_operation(op);
422}
423
424/*
425 * Extend the region to be written back to include subsequent contiguously
426 * dirty pages if possible, but don't sleep while doing so.
427 *
428 * If this page holds new content, then we can include filler zeros in the
429 * writeback.
430 */
431static void afs_extend_writeback(struct address_space *mapping,
432 struct afs_vnode *vnode,
433 long *_count,
434 loff_t start,
435 loff_t max_len,
436 bool new_content,
437 unsigned int *_len)
438{
439 struct pagevec pvec;
440 struct page *page;
441 unsigned long priv;
442 unsigned int psize, filler = 0;
443 unsigned int f, t;
444 loff_t len = *_len;
445 pgoff_t index = (start + len) / PAGE_SIZE;
446 bool stop = true;
447 unsigned int i;
448
449 XA_STATE(xas, &mapping->i_pages, index);
450 pagevec_init(&pvec);
451
452 do {
453 /* Firstly, we gather up a batch of contiguous dirty pages
454 * under the RCU read lock - but we can't clear the dirty flags
455 * there if any of those pages are mapped.
456 */
457 rcu_read_lock();
458
459 xas_for_each(&xas, page, ULONG_MAX) {
460 stop = true;
461 if (xas_retry(&xas, page))
462 continue;
463 if (xa_is_value(page))
464 break;
465 if (page->index != index)
466 break;
467
468 if (!page_cache_get_speculative(page)) {
469 xas_reset(&xas);
470 continue;
471 }
472
473 /* Has the page moved or been split? */
474 if (unlikely(page != xas_reload(&xas))) {
475 put_page(page);
476 break;
477 }
478
479 if (!trylock_page(page)) {
480 put_page(page);
481 break;
482 }
483 if (!PageDirty(page) || PageWriteback(page)) {
484 unlock_page(page);
485 put_page(page);
486 break;
487 }
488
489 psize = thp_size(page);
490 priv = page_private(page);
491 f = afs_page_dirty_from(page, priv);
492 t = afs_page_dirty_to(page, priv);
493 if (f != 0 && !new_content) {
494 unlock_page(page);
495 put_page(page);
496 break;
497 }
498
499 len += filler + t;
500 filler = psize - t;
501 if (len >= max_len || *_count <= 0)
502 stop = true;
503 else if (t == psize || new_content)
504 stop = false;
505
506 index += thp_nr_pages(page);
507 if (!pagevec_add(&pvec, page))
508 break;
509 if (stop)
510 break;
511 }
512
513 if (!stop)
514 xas_pause(&xas);
515 rcu_read_unlock();
516
517 /* Now, if we obtained any pages, we can shift them to being
518 * writable and mark them for caching.
519 */
520 if (!pagevec_count(&pvec))
521 break;
522
523 for (i = 0; i < pagevec_count(&pvec); i++) {
524 page = pvec.pages[i];
525 trace_afs_page_dirty(vnode, tracepoint_string("store+"), page);
526
527 if (!clear_page_dirty_for_io(page))
528 BUG();
529 if (test_set_page_writeback(page))
530 BUG();
531
532 *_count -= thp_nr_pages(page);
533 unlock_page(page);
534 }
535
536 pagevec_release(&pvec);
537 cond_resched();
538 } while (!stop);
539
540 *_len = len;
541}
542
543/*
544 * Synchronously write back the locked page and any subsequent non-locked dirty
545 * pages.
546 */
547static ssize_t afs_write_back_from_locked_page(struct address_space *mapping,
548 struct writeback_control *wbc,
549 struct page *page,
550 loff_t start, loff_t end)
551{
552 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
553 struct iov_iter iter;
554 unsigned long priv;
555 unsigned int offset, to, len, max_len;
556 loff_t i_size = i_size_read(&vnode->vfs_inode);
557 bool new_content = test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
558 long count = wbc->nr_to_write;
559 int ret;
560
561 _enter(",%lx,%llx-%llx", page->index, start, end);
562
563 if (test_set_page_writeback(page))
564 BUG();
565
566 count -= thp_nr_pages(page);
567
568 /* Find all consecutive lockable dirty pages that have contiguous
569 * written regions, stopping when we find a page that is not
570 * immediately lockable, is not dirty or is missing, or we reach the
571 * end of the range.
572 */
573 priv = page_private(page);
574 offset = afs_page_dirty_from(page, priv);
575 to = afs_page_dirty_to(page, priv);
576 trace_afs_page_dirty(vnode, tracepoint_string("store"), page);
577
578 len = to - offset;
579 start += offset;
580 if (start < i_size) {
581 /* Trim the write to the EOF; the extra data is ignored. Also
582 * put an upper limit on the size of a single storedata op.
583 */
584 max_len = 65536 * 4096;
585 max_len = min_t(unsigned long long, max_len, end - start + 1);
586 max_len = min_t(unsigned long long, max_len, i_size - start);
587
588 if (len < max_len &&
589 (to == thp_size(page) || new_content))
590 afs_extend_writeback(mapping, vnode, &count,
591 start, max_len, new_content, &len);
592 len = min_t(loff_t, len, max_len);
593 }
594
595 /* We now have a contiguous set of dirty pages, each with writeback
596 * set; the first page is still locked at this point, but all the rest
597 * have been unlocked.
598 */
599 unlock_page(page);
600
601 if (start < i_size) {
602 _debug("write back %x @%llx [%llx]", len, start, i_size);
603
604 iov_iter_xarray(&iter, WRITE, &mapping->i_pages, start, len);
605 ret = afs_store_data(vnode, &iter, start, false);
606 } else {
607 _debug("write discard %x @%llx [%llx]", len, start, i_size);
608
609 /* The dirty region was entirely beyond the EOF. */
610 afs_pages_written_back(vnode, start, len);
611 ret = 0;
612 }
613
614 switch (ret) {
615 case 0:
616 wbc->nr_to_write = count;
617 ret = len;
618 break;
619
620 default:
621 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
622 fallthrough;
623 case -EACCES:
624 case -EPERM:
625 case -ENOKEY:
626 case -EKEYEXPIRED:
627 case -EKEYREJECTED:
628 case -EKEYREVOKED:
629 afs_redirty_pages(wbc, mapping, start, len);
630 mapping_set_error(mapping, ret);
631 break;
632
633 case -EDQUOT:
634 case -ENOSPC:
635 afs_redirty_pages(wbc, mapping, start, len);
636 mapping_set_error(mapping, -ENOSPC);
637 break;
638
639 case -EROFS:
640 case -EIO:
641 case -EREMOTEIO:
642 case -EFBIG:
643 case -ENOENT:
644 case -ENOMEDIUM:
645 case -ENXIO:
646 trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
647 afs_kill_pages(mapping, start, len);
648 mapping_set_error(mapping, ret);
649 break;
650 }
651
652 _leave(" = %d", ret);
653 return ret;
654}
655
656/*
657 * write a page back to the server
658 * - the caller locked the page for us
659 */
660int afs_writepage(struct page *page, struct writeback_control *wbc)
661{
662 ssize_t ret;
663 loff_t start;
664
665 _enter("{%lx},", page->index);
666
667 start = page->index * PAGE_SIZE;
668 ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
669 start, LLONG_MAX - start);
670 if (ret < 0) {
671 _leave(" = %zd", ret);
672 return ret;
673 }
674
675 _leave(" = 0");
676 return 0;
677}
678
679/*
680 * write a region of pages back to the server
681 */
682static int afs_writepages_region(struct address_space *mapping,
683 struct writeback_control *wbc,
684 loff_t start, loff_t end, loff_t *_next)
685{
686 struct page *page;
687 ssize_t ret;
688 int n;
689
690 _enter("%llx,%llx,", start, end);
691
692 do {
693 pgoff_t index = start / PAGE_SIZE;
694
695 n = find_get_pages_range_tag(mapping, &index, end / PAGE_SIZE,
696 PAGECACHE_TAG_DIRTY, 1, &page);
697 if (!n)
698 break;
699
700 start = (loff_t)page->index * PAGE_SIZE; /* May regress with THPs */
701
702 _debug("wback %lx", page->index);
703
704 /* At this point we hold neither the i_pages lock nor the
705 * page lock: the page may be truncated or invalidated
706 * (changing page->mapping to NULL), or even swizzled
707 * back from swapper_space to tmpfs file mapping
708 */
709 if (wbc->sync_mode != WB_SYNC_NONE) {
710 ret = lock_page_killable(page);
711 if (ret < 0) {
712 put_page(page);
713 return ret;
714 }
715 } else {
716 if (!trylock_page(page)) {
717 put_page(page);
718 return 0;
719 }
720 }
721
722 if (page->mapping != mapping || !PageDirty(page)) {
723 start += thp_size(page);
724 unlock_page(page);
725 put_page(page);
726 continue;
727 }
728
729 if (PageWriteback(page)) {
730 unlock_page(page);
731 if (wbc->sync_mode != WB_SYNC_NONE)
732 wait_on_page_writeback(page);
733 put_page(page);
734 continue;
735 }
736
737 if (!clear_page_dirty_for_io(page))
738 BUG();
739 ret = afs_write_back_from_locked_page(mapping, wbc, page, start, end);
740 put_page(page);
741 if (ret < 0) {
742 _leave(" = %zd", ret);
743 return ret;
744 }
745
746 start += ret;
747
748 cond_resched();
749 } while (wbc->nr_to_write > 0);
750
751 *_next = start;
752 _leave(" = 0 [%llx]", *_next);
753 return 0;
754}
755
756/*
757 * write some of the pending data back to the server
758 */
759int afs_writepages(struct address_space *mapping,
760 struct writeback_control *wbc)
761{
762 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
763 loff_t start, next;
764 int ret;
765
766 _enter("");
767
768 /* We have to be careful as we can end up racing with setattr()
769 * truncating the pagecache since the caller doesn't take a lock here
770 * to prevent it.
771 */
772 if (wbc->sync_mode == WB_SYNC_ALL)
773 down_read(&vnode->validate_lock);
774 else if (!down_read_trylock(&vnode->validate_lock))
775 return 0;
776
777 if (wbc->range_cyclic) {
778 start = mapping->writeback_index * PAGE_SIZE;
779 ret = afs_writepages_region(mapping, wbc, start, LLONG_MAX, &next);
780 if (ret == 0) {
781 mapping->writeback_index = next / PAGE_SIZE;
782 if (start > 0 && wbc->nr_to_write > 0) {
783 ret = afs_writepages_region(mapping, wbc, 0,
784 start, &next);
785 if (ret == 0)
786 mapping->writeback_index =
787 next / PAGE_SIZE;
788 }
789 }
790 } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
791 ret = afs_writepages_region(mapping, wbc, 0, LLONG_MAX, &next);
792 if (wbc->nr_to_write > 0 && ret == 0)
793 mapping->writeback_index = next / PAGE_SIZE;
794 } else {
795 ret = afs_writepages_region(mapping, wbc,
796 wbc->range_start, wbc->range_end, &next);
797 }
798
799 up_read(&vnode->validate_lock);
800 _leave(" = %d", ret);
801 return ret;
802}
803
804/*
805 * write to an AFS file
806 */
807ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
808{
809 struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
810 struct afs_file *af = iocb->ki_filp->private_data;
811 ssize_t result;
812 size_t count = iov_iter_count(from);
813
814 _enter("{%llx:%llu},{%zu},",
815 vnode->fid.vid, vnode->fid.vnode, count);
816
817 if (IS_SWAPFILE(&vnode->vfs_inode)) {
818 printk(KERN_INFO
819 "AFS: Attempt to write to active swap file!\n");
820 return -EBUSY;
821 }
822
823 if (!count)
824 return 0;
825
826 result = afs_validate(vnode, af->key);
827 if (result < 0)
828 return result;
829
830 result = generic_file_write_iter(iocb, from);
831
832 _leave(" = %zd", result);
833 return result;
834}
835
836/*
837 * flush any dirty pages for this process, and check for write errors.
838 * - the return status from this call provides a reliable indication of
839 * whether any write errors occurred for this process.
840 */
841int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
842{
843 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
844 struct afs_file *af = file->private_data;
845 int ret;
846
847 _enter("{%llx:%llu},{n=%pD},%d",
848 vnode->fid.vid, vnode->fid.vnode, file,
849 datasync);
850
851 ret = afs_validate(vnode, af->key);
852 if (ret < 0)
853 return ret;
854
855 return file_write_and_wait_range(file, start, end);
856}
857
858/*
859 * notification that a previously read-only page is about to become writable
860 * - if it returns an error, the caller will deliver a bus error signal
861 */
862vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
863{
864 struct page *page = thp_head(vmf->page);
865 struct file *file = vmf->vma->vm_file;
866 struct inode *inode = file_inode(file);
867 struct afs_vnode *vnode = AFS_FS_I(inode);
868 struct afs_file *af = file->private_data;
869 unsigned long priv;
870 vm_fault_t ret = VM_FAULT_RETRY;
871
872 _enter("{{%llx:%llu}},{%lx}", vnode->fid.vid, vnode->fid.vnode, page->index);
873
874 afs_validate(vnode, af->key);
875
876 sb_start_pagefault(inode->i_sb);
877
878 /* Wait for the page to be written to the cache before we allow it to
879 * be modified. We then assume the entire page will need writing back.
880 */
881#ifdef CONFIG_AFS_FSCACHE
882 if (PageFsCache(page) &&
883 wait_on_page_fscache_killable(page) < 0)
884 goto out;
885#endif
886
887 if (wait_on_page_writeback_killable(page))
888 goto out;
889
890 if (lock_page_killable(page) < 0)
891 goto out;
892
893 /* We mustn't change page->private until writeback is complete as that
894 * details the portion of the page we need to write back and we might
895 * need to redirty the page if there's a problem.
896 */
897 if (wait_on_page_writeback_killable(page) < 0) {
898 unlock_page(page);
899 goto out;
900 }
901
902 priv = afs_page_dirty(page, 0, thp_size(page));
903 priv = afs_page_dirty_mmapped(priv);
904 if (PagePrivate(page)) {
905 set_page_private(page, priv);
906 trace_afs_page_dirty(vnode, tracepoint_string("mkwrite+"), page);
907 } else {
908 attach_page_private(page, (void *)priv);
909 trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"), page);
910 }
911 file_update_time(file);
912
913 ret = VM_FAULT_LOCKED;
914out:
915 sb_end_pagefault(inode->i_sb);
916 return ret;
917}
918
919/*
920 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
921 */
922void afs_prune_wb_keys(struct afs_vnode *vnode)
923{
924 LIST_HEAD(graveyard);
925 struct afs_wb_key *wbk, *tmp;
926
927 /* Discard unused keys */
928 spin_lock(&vnode->wb_lock);
929
930 if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
931 !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
932 list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
933 if (refcount_read(&wbk->usage) == 1)
934 list_move(&wbk->vnode_link, &graveyard);
935 }
936 }
937
938 spin_unlock(&vnode->wb_lock);
939
940 while (!list_empty(&graveyard)) {
941 wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
942 list_del(&wbk->vnode_link);
943 afs_put_wb_key(wbk);
944 }
945}
946
947/*
948 * Clean up a page during invalidation.
949 */
950int afs_launder_page(struct page *page)
951{
952 struct address_space *mapping = page->mapping;
953 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
954 struct iov_iter iter;
955 struct bio_vec bv[1];
956 unsigned long priv;
957 unsigned int f, t;
958 int ret = 0;
959
960 _enter("{%lx}", page->index);
961
962 priv = page_private(page);
963 if (clear_page_dirty_for_io(page)) {
964 f = 0;
965 t = thp_size(page);
966 if (PagePrivate(page)) {
967 f = afs_page_dirty_from(page, priv);
968 t = afs_page_dirty_to(page, priv);
969 }
970
971 bv[0].bv_page = page;
972 bv[0].bv_offset = f;
973 bv[0].bv_len = t - f;
974 iov_iter_bvec(&iter, WRITE, bv, 1, bv[0].bv_len);
975
976 trace_afs_page_dirty(vnode, tracepoint_string("launder"), page);
977 ret = afs_store_data(vnode, &iter, page_offset(page) + f, true);
978 }
979
980 trace_afs_page_dirty(vnode, tracepoint_string("laundered"), page);
981 detach_page_private(page);
982 wait_on_page_fscache(page);
983 return ret;
984}