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 "internal.h"
15
16/*
17 * mark a page as having been made dirty and thus needing writeback
18 */
19int afs_set_page_dirty(struct page *page)
20{
21 _enter("");
22 return __set_page_dirty_nobuffers(page);
23}
24
25/*
26 * partly or wholly fill a page that's under preparation for writing
27 */
28static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
29 loff_t pos, unsigned int len, struct page *page)
30{
31 struct afs_read *req;
32 size_t p;
33 void *data;
34 int ret;
35
36 _enter(",,%llu", (unsigned long long)pos);
37
38 if (pos >= vnode->vfs_inode.i_size) {
39 p = pos & ~PAGE_MASK;
40 ASSERTCMP(p + len, <=, PAGE_SIZE);
41 data = kmap(page);
42 memset(data + p, 0, len);
43 kunmap(page);
44 return 0;
45 }
46
47 req = kzalloc(struct_size(req, array, 1), GFP_KERNEL);
48 if (!req)
49 return -ENOMEM;
50
51 refcount_set(&req->usage, 1);
52 req->pos = pos;
53 req->len = len;
54 req->nr_pages = 1;
55 req->pages = req->array;
56 req->pages[0] = page;
57 get_page(page);
58
59 ret = afs_fetch_data(vnode, key, req);
60 afs_put_read(req);
61 if (ret < 0) {
62 if (ret == -ENOENT) {
63 _debug("got NOENT from server"
64 " - marking file deleted and stale");
65 set_bit(AFS_VNODE_DELETED, &vnode->flags);
66 ret = -ESTALE;
67 }
68 }
69
70 _leave(" = %d", ret);
71 return ret;
72}
73
74/*
75 * prepare to perform part of a write to a page
76 */
77int afs_write_begin(struct file *file, struct address_space *mapping,
78 loff_t pos, unsigned len, unsigned flags,
79 struct page **pagep, void **fsdata)
80{
81 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
82 struct page *page;
83 struct key *key = afs_file_key(file);
84 unsigned long priv;
85 unsigned f, from = pos & (PAGE_SIZE - 1);
86 unsigned t, to = from + len;
87 pgoff_t index = pos >> PAGE_SHIFT;
88 int ret;
89
90 _enter("{%llx:%llu},{%lx},%u,%u",
91 vnode->fid.vid, vnode->fid.vnode, index, from, to);
92
93 /* We want to store information about how much of a page is altered in
94 * page->private.
95 */
96 BUILD_BUG_ON(PAGE_SIZE > 32768 && sizeof(page->private) < 8);
97
98 page = grab_cache_page_write_begin(mapping, index, flags);
99 if (!page)
100 return -ENOMEM;
101
102 if (!PageUptodate(page) && len != PAGE_SIZE) {
103 ret = afs_fill_page(vnode, key, pos & PAGE_MASK, PAGE_SIZE, page);
104 if (ret < 0) {
105 unlock_page(page);
106 put_page(page);
107 _leave(" = %d [prep]", ret);
108 return ret;
109 }
110 SetPageUptodate(page);
111 }
112
113 /* page won't leak in error case: it eventually gets cleaned off LRU */
114 *pagep = page;
115
116try_again:
117 /* See if this page is already partially written in a way that we can
118 * merge the new write with.
119 */
120 t = f = 0;
121 if (PagePrivate(page)) {
122 priv = page_private(page);
123 f = priv & AFS_PRIV_MAX;
124 t = priv >> AFS_PRIV_SHIFT;
125 ASSERTCMP(f, <=, t);
126 }
127
128 if (f != t) {
129 if (PageWriteback(page)) {
130 trace_afs_page_dirty(vnode, tracepoint_string("alrdy"),
131 page->index, priv);
132 goto flush_conflicting_write;
133 }
134 /* If the file is being filled locally, allow inter-write
135 * spaces to be merged into writes. If it's not, only write
136 * back what the user gives us.
137 */
138 if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
139 (to < f || from > t))
140 goto flush_conflicting_write;
141 if (from < f)
142 f = from;
143 if (to > t)
144 t = to;
145 } else {
146 f = from;
147 t = to;
148 }
149
150 priv = (unsigned long)t << AFS_PRIV_SHIFT;
151 priv |= f;
152 trace_afs_page_dirty(vnode, tracepoint_string("begin"),
153 page->index, priv);
154 SetPagePrivate(page);
155 set_page_private(page, priv);
156 _leave(" = 0");
157 return 0;
158
159 /* The previous write and this write aren't adjacent or overlapping, so
160 * flush the page out.
161 */
162flush_conflicting_write:
163 _debug("flush conflict");
164 ret = write_one_page(page);
165 if (ret < 0) {
166 _leave(" = %d", ret);
167 return ret;
168 }
169
170 ret = lock_page_killable(page);
171 if (ret < 0) {
172 _leave(" = %d", ret);
173 return ret;
174 }
175 goto try_again;
176}
177
178/*
179 * finalise part of a write to a page
180 */
181int afs_write_end(struct file *file, struct address_space *mapping,
182 loff_t pos, unsigned len, unsigned copied,
183 struct page *page, void *fsdata)
184{
185 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
186 struct key *key = afs_file_key(file);
187 loff_t i_size, maybe_i_size;
188 int ret;
189
190 _enter("{%llx:%llu},{%lx}",
191 vnode->fid.vid, vnode->fid.vnode, page->index);
192
193 maybe_i_size = pos + copied;
194
195 i_size = i_size_read(&vnode->vfs_inode);
196 if (maybe_i_size > i_size) {
197 spin_lock(&vnode->wb_lock);
198 i_size = i_size_read(&vnode->vfs_inode);
199 if (maybe_i_size > i_size)
200 i_size_write(&vnode->vfs_inode, maybe_i_size);
201 spin_unlock(&vnode->wb_lock);
202 }
203
204 if (!PageUptodate(page)) {
205 if (copied < len) {
206 /* Try and load any missing data from the server. The
207 * unmarshalling routine will take care of clearing any
208 * bits that are beyond the EOF.
209 */
210 ret = afs_fill_page(vnode, key, pos + copied,
211 len - copied, page);
212 if (ret < 0)
213 goto out;
214 }
215 SetPageUptodate(page);
216 }
217
218 set_page_dirty(page);
219 if (PageDirty(page))
220 _debug("dirtied");
221 ret = copied;
222
223out:
224 unlock_page(page);
225 put_page(page);
226 return ret;
227}
228
229/*
230 * kill all the pages in the given range
231 */
232static void afs_kill_pages(struct address_space *mapping,
233 pgoff_t first, pgoff_t last)
234{
235 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
236 struct pagevec pv;
237 unsigned count, loop;
238
239 _enter("{%llx:%llu},%lx-%lx",
240 vnode->fid.vid, vnode->fid.vnode, first, last);
241
242 pagevec_init(&pv);
243
244 do {
245 _debug("kill %lx-%lx", first, last);
246
247 count = last - first + 1;
248 if (count > PAGEVEC_SIZE)
249 count = PAGEVEC_SIZE;
250 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
251 ASSERTCMP(pv.nr, ==, count);
252
253 for (loop = 0; loop < count; loop++) {
254 struct page *page = pv.pages[loop];
255 ClearPageUptodate(page);
256 SetPageError(page);
257 end_page_writeback(page);
258 if (page->index >= first)
259 first = page->index + 1;
260 lock_page(page);
261 generic_error_remove_page(mapping, page);
262 unlock_page(page);
263 }
264
265 __pagevec_release(&pv);
266 } while (first <= last);
267
268 _leave("");
269}
270
271/*
272 * Redirty all the pages in a given range.
273 */
274static void afs_redirty_pages(struct writeback_control *wbc,
275 struct address_space *mapping,
276 pgoff_t first, pgoff_t last)
277{
278 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
279 struct pagevec pv;
280 unsigned count, loop;
281
282 _enter("{%llx:%llu},%lx-%lx",
283 vnode->fid.vid, vnode->fid.vnode, first, last);
284
285 pagevec_init(&pv);
286
287 do {
288 _debug("redirty %lx-%lx", first, last);
289
290 count = last - first + 1;
291 if (count > PAGEVEC_SIZE)
292 count = PAGEVEC_SIZE;
293 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
294 ASSERTCMP(pv.nr, ==, count);
295
296 for (loop = 0; loop < count; loop++) {
297 struct page *page = pv.pages[loop];
298
299 redirty_page_for_writepage(wbc, page);
300 end_page_writeback(page);
301 if (page->index >= first)
302 first = page->index + 1;
303 }
304
305 __pagevec_release(&pv);
306 } while (first <= last);
307
308 _leave("");
309}
310
311/*
312 * completion of write to server
313 */
314static void afs_pages_written_back(struct afs_vnode *vnode,
315 pgoff_t first, pgoff_t last)
316{
317 struct pagevec pv;
318 unsigned long priv;
319 unsigned count, loop;
320
321 _enter("{%llx:%llu},{%lx-%lx}",
322 vnode->fid.vid, vnode->fid.vnode, first, last);
323
324 pagevec_init(&pv);
325
326 do {
327 _debug("done %lx-%lx", first, last);
328
329 count = last - first + 1;
330 if (count > PAGEVEC_SIZE)
331 count = PAGEVEC_SIZE;
332 pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
333 first, count, pv.pages);
334 ASSERTCMP(pv.nr, ==, count);
335
336 for (loop = 0; loop < count; loop++) {
337 priv = page_private(pv.pages[loop]);
338 trace_afs_page_dirty(vnode, tracepoint_string("clear"),
339 pv.pages[loop]->index, priv);
340 set_page_private(pv.pages[loop], 0);
341 end_page_writeback(pv.pages[loop]);
342 }
343 first += count;
344 __pagevec_release(&pv);
345 } while (first <= last);
346
347 afs_prune_wb_keys(vnode);
348 _leave("");
349}
350
351/*
352 * write to a file
353 */
354static int afs_store_data(struct address_space *mapping,
355 pgoff_t first, pgoff_t last,
356 unsigned offset, unsigned to)
357{
358 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
359 struct afs_fs_cursor fc;
360 struct afs_status_cb *scb;
361 struct afs_wb_key *wbk = NULL;
362 struct list_head *p;
363 int ret = -ENOKEY, ret2;
364
365 _enter("%s{%llx:%llu.%u},%lx,%lx,%x,%x",
366 vnode->volume->name,
367 vnode->fid.vid,
368 vnode->fid.vnode,
369 vnode->fid.unique,
370 first, last, offset, to);
371
372 scb = kzalloc(sizeof(struct afs_status_cb), GFP_NOFS);
373 if (!scb)
374 return -ENOMEM;
375
376 spin_lock(&vnode->wb_lock);
377 p = vnode->wb_keys.next;
378
379 /* Iterate through the list looking for a valid key to use. */
380try_next_key:
381 while (p != &vnode->wb_keys) {
382 wbk = list_entry(p, struct afs_wb_key, vnode_link);
383 _debug("wbk %u", key_serial(wbk->key));
384 ret2 = key_validate(wbk->key);
385 if (ret2 == 0)
386 goto found_key;
387 if (ret == -ENOKEY)
388 ret = ret2;
389 p = p->next;
390 }
391
392 spin_unlock(&vnode->wb_lock);
393 afs_put_wb_key(wbk);
394 kfree(scb);
395 _leave(" = %d [no keys]", ret);
396 return ret;
397
398found_key:
399 refcount_inc(&wbk->usage);
400 spin_unlock(&vnode->wb_lock);
401
402 _debug("USE WB KEY %u", key_serial(wbk->key));
403
404 ret = -ERESTARTSYS;
405 if (afs_begin_vnode_operation(&fc, vnode, wbk->key, false)) {
406 afs_dataversion_t data_version = vnode->status.data_version + 1;
407
408 while (afs_select_fileserver(&fc)) {
409 fc.cb_break = afs_calc_vnode_cb_break(vnode);
410 afs_fs_store_data(&fc, mapping, first, last, offset, to, scb);
411 }
412
413 afs_check_for_remote_deletion(&fc, vnode);
414 afs_vnode_commit_status(&fc, vnode, fc.cb_break,
415 &data_version, scb);
416 if (fc.ac.error == 0)
417 afs_pages_written_back(vnode, first, last);
418 ret = afs_end_vnode_operation(&fc);
419 }
420
421 switch (ret) {
422 case 0:
423 afs_stat_v(vnode, n_stores);
424 atomic_long_add((last * PAGE_SIZE + to) -
425 (first * PAGE_SIZE + offset),
426 &afs_v2net(vnode)->n_store_bytes);
427 break;
428 case -EACCES:
429 case -EPERM:
430 case -ENOKEY:
431 case -EKEYEXPIRED:
432 case -EKEYREJECTED:
433 case -EKEYREVOKED:
434 _debug("next");
435 spin_lock(&vnode->wb_lock);
436 p = wbk->vnode_link.next;
437 afs_put_wb_key(wbk);
438 goto try_next_key;
439 }
440
441 afs_put_wb_key(wbk);
442 kfree(scb);
443 _leave(" = %d", ret);
444 return ret;
445}
446
447/*
448 * Synchronously write back the locked page and any subsequent non-locked dirty
449 * pages.
450 */
451static int afs_write_back_from_locked_page(struct address_space *mapping,
452 struct writeback_control *wbc,
453 struct page *primary_page,
454 pgoff_t final_page)
455{
456 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
457 struct page *pages[8], *page;
458 unsigned long count, priv;
459 unsigned n, offset, to, f, t;
460 pgoff_t start, first, last;
461 int loop, ret;
462
463 _enter(",%lx", primary_page->index);
464
465 count = 1;
466 if (test_set_page_writeback(primary_page))
467 BUG();
468
469 /* Find all consecutive lockable dirty pages that have contiguous
470 * written regions, stopping when we find a page that is not
471 * immediately lockable, is not dirty or is missing, or we reach the
472 * end of the range.
473 */
474 start = primary_page->index;
475 priv = page_private(primary_page);
476 offset = priv & AFS_PRIV_MAX;
477 to = priv >> AFS_PRIV_SHIFT;
478 trace_afs_page_dirty(vnode, tracepoint_string("store"),
479 primary_page->index, priv);
480
481 WARN_ON(offset == to);
482 if (offset == to)
483 trace_afs_page_dirty(vnode, tracepoint_string("WARN"),
484 primary_page->index, priv);
485
486 if (start >= final_page ||
487 (to < PAGE_SIZE && !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)))
488 goto no_more;
489
490 start++;
491 do {
492 _debug("more %lx [%lx]", start, count);
493 n = final_page - start + 1;
494 if (n > ARRAY_SIZE(pages))
495 n = ARRAY_SIZE(pages);
496 n = find_get_pages_contig(mapping, start, ARRAY_SIZE(pages), pages);
497 _debug("fgpc %u", n);
498 if (n == 0)
499 goto no_more;
500 if (pages[0]->index != start) {
501 do {
502 put_page(pages[--n]);
503 } while (n > 0);
504 goto no_more;
505 }
506
507 for (loop = 0; loop < n; loop++) {
508 page = pages[loop];
509 if (to != PAGE_SIZE &&
510 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags))
511 break;
512 if (page->index > final_page)
513 break;
514 if (!trylock_page(page))
515 break;
516 if (!PageDirty(page) || PageWriteback(page)) {
517 unlock_page(page);
518 break;
519 }
520
521 priv = page_private(page);
522 f = priv & AFS_PRIV_MAX;
523 t = priv >> AFS_PRIV_SHIFT;
524 if (f != 0 &&
525 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)) {
526 unlock_page(page);
527 break;
528 }
529 to = t;
530
531 trace_afs_page_dirty(vnode, tracepoint_string("store+"),
532 page->index, priv);
533
534 if (!clear_page_dirty_for_io(page))
535 BUG();
536 if (test_set_page_writeback(page))
537 BUG();
538 unlock_page(page);
539 put_page(page);
540 }
541 count += loop;
542 if (loop < n) {
543 for (; loop < n; loop++)
544 put_page(pages[loop]);
545 goto no_more;
546 }
547
548 start += loop;
549 } while (start <= final_page && count < 65536);
550
551no_more:
552 /* We now have a contiguous set of dirty pages, each with writeback
553 * set; the first page is still locked at this point, but all the rest
554 * have been unlocked.
555 */
556 unlock_page(primary_page);
557
558 first = primary_page->index;
559 last = first + count - 1;
560
561 _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
562
563 ret = afs_store_data(mapping, first, last, offset, to);
564 switch (ret) {
565 case 0:
566 ret = count;
567 break;
568
569 default:
570 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
571 /* Fall through */
572 case -EACCES:
573 case -EPERM:
574 case -ENOKEY:
575 case -EKEYEXPIRED:
576 case -EKEYREJECTED:
577 case -EKEYREVOKED:
578 afs_redirty_pages(wbc, mapping, first, last);
579 mapping_set_error(mapping, ret);
580 break;
581
582 case -EDQUOT:
583 case -ENOSPC:
584 afs_redirty_pages(wbc, mapping, first, last);
585 mapping_set_error(mapping, -ENOSPC);
586 break;
587
588 case -EROFS:
589 case -EIO:
590 case -EREMOTEIO:
591 case -EFBIG:
592 case -ENOENT:
593 case -ENOMEDIUM:
594 case -ENXIO:
595 trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
596 afs_kill_pages(mapping, first, last);
597 mapping_set_error(mapping, ret);
598 break;
599 }
600
601 _leave(" = %d", ret);
602 return ret;
603}
604
605/*
606 * write a page back to the server
607 * - the caller locked the page for us
608 */
609int afs_writepage(struct page *page, struct writeback_control *wbc)
610{
611 int ret;
612
613 _enter("{%lx},", page->index);
614
615 ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
616 wbc->range_end >> PAGE_SHIFT);
617 if (ret < 0) {
618 _leave(" = %d", ret);
619 return 0;
620 }
621
622 wbc->nr_to_write -= ret;
623
624 _leave(" = 0");
625 return 0;
626}
627
628/*
629 * write a region of pages back to the server
630 */
631static int afs_writepages_region(struct address_space *mapping,
632 struct writeback_control *wbc,
633 pgoff_t index, pgoff_t end, pgoff_t *_next)
634{
635 struct page *page;
636 int ret, n;
637
638 _enter(",,%lx,%lx,", index, end);
639
640 do {
641 n = find_get_pages_range_tag(mapping, &index, end,
642 PAGECACHE_TAG_DIRTY, 1, &page);
643 if (!n)
644 break;
645
646 _debug("wback %lx", page->index);
647
648 /*
649 * at this point we hold neither the i_pages lock nor the
650 * page lock: the page may be truncated or invalidated
651 * (changing page->mapping to NULL), or even swizzled
652 * back from swapper_space to tmpfs file mapping
653 */
654 ret = lock_page_killable(page);
655 if (ret < 0) {
656 put_page(page);
657 _leave(" = %d", ret);
658 return ret;
659 }
660
661 if (page->mapping != mapping || !PageDirty(page)) {
662 unlock_page(page);
663 put_page(page);
664 continue;
665 }
666
667 if (PageWriteback(page)) {
668 unlock_page(page);
669 if (wbc->sync_mode != WB_SYNC_NONE)
670 wait_on_page_writeback(page);
671 put_page(page);
672 continue;
673 }
674
675 if (!clear_page_dirty_for_io(page))
676 BUG();
677 ret = afs_write_back_from_locked_page(mapping, wbc, page, end);
678 put_page(page);
679 if (ret < 0) {
680 _leave(" = %d", ret);
681 return ret;
682 }
683
684 wbc->nr_to_write -= ret;
685
686 cond_resched();
687 } while (index < end && wbc->nr_to_write > 0);
688
689 *_next = index;
690 _leave(" = 0 [%lx]", *_next);
691 return 0;
692}
693
694/*
695 * write some of the pending data back to the server
696 */
697int afs_writepages(struct address_space *mapping,
698 struct writeback_control *wbc)
699{
700 pgoff_t start, end, next;
701 int ret;
702
703 _enter("");
704
705 if (wbc->range_cyclic) {
706 start = mapping->writeback_index;
707 end = -1;
708 ret = afs_writepages_region(mapping, wbc, start, end, &next);
709 if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
710 ret = afs_writepages_region(mapping, wbc, 0, start,
711 &next);
712 mapping->writeback_index = next;
713 } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
714 end = (pgoff_t)(LLONG_MAX >> PAGE_SHIFT);
715 ret = afs_writepages_region(mapping, wbc, 0, end, &next);
716 if (wbc->nr_to_write > 0)
717 mapping->writeback_index = next;
718 } else {
719 start = wbc->range_start >> PAGE_SHIFT;
720 end = wbc->range_end >> PAGE_SHIFT;
721 ret = afs_writepages_region(mapping, wbc, start, end, &next);
722 }
723
724 _leave(" = %d", ret);
725 return ret;
726}
727
728/*
729 * write to an AFS file
730 */
731ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
732{
733 struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
734 ssize_t result;
735 size_t count = iov_iter_count(from);
736
737 _enter("{%llx:%llu},{%zu},",
738 vnode->fid.vid, vnode->fid.vnode, count);
739
740 if (IS_SWAPFILE(&vnode->vfs_inode)) {
741 printk(KERN_INFO
742 "AFS: Attempt to write to active swap file!\n");
743 return -EBUSY;
744 }
745
746 if (!count)
747 return 0;
748
749 result = generic_file_write_iter(iocb, from);
750
751 _leave(" = %zd", result);
752 return result;
753}
754
755/*
756 * flush any dirty pages for this process, and check for write errors.
757 * - the return status from this call provides a reliable indication of
758 * whether any write errors occurred for this process.
759 */
760int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
761{
762 struct inode *inode = file_inode(file);
763 struct afs_vnode *vnode = AFS_FS_I(inode);
764
765 _enter("{%llx:%llu},{n=%pD},%d",
766 vnode->fid.vid, vnode->fid.vnode, file,
767 datasync);
768
769 return file_write_and_wait_range(file, start, end);
770}
771
772/*
773 * notification that a previously read-only page is about to become writable
774 * - if it returns an error, the caller will deliver a bus error signal
775 */
776vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
777{
778 struct file *file = vmf->vma->vm_file;
779 struct inode *inode = file_inode(file);
780 struct afs_vnode *vnode = AFS_FS_I(inode);
781 unsigned long priv;
782
783 _enter("{{%llx:%llu}},{%lx}",
784 vnode->fid.vid, vnode->fid.vnode, vmf->page->index);
785
786 sb_start_pagefault(inode->i_sb);
787
788 /* Wait for the page to be written to the cache before we allow it to
789 * be modified. We then assume the entire page will need writing back.
790 */
791#ifdef CONFIG_AFS_FSCACHE
792 fscache_wait_on_page_write(vnode->cache, vmf->page);
793#endif
794
795 if (PageWriteback(vmf->page) &&
796 wait_on_page_bit_killable(vmf->page, PG_writeback) < 0)
797 return VM_FAULT_RETRY;
798
799 if (lock_page_killable(vmf->page) < 0)
800 return VM_FAULT_RETRY;
801
802 /* We mustn't change page->private until writeback is complete as that
803 * details the portion of the page we need to write back and we might
804 * need to redirty the page if there's a problem.
805 */
806 wait_on_page_writeback(vmf->page);
807
808 priv = (unsigned long)PAGE_SIZE << AFS_PRIV_SHIFT; /* To */
809 priv |= 0; /* From */
810 trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"),
811 vmf->page->index, priv);
812 SetPagePrivate(vmf->page);
813 set_page_private(vmf->page, priv);
814
815 sb_end_pagefault(inode->i_sb);
816 return VM_FAULT_LOCKED;
817}
818
819/*
820 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
821 */
822void afs_prune_wb_keys(struct afs_vnode *vnode)
823{
824 LIST_HEAD(graveyard);
825 struct afs_wb_key *wbk, *tmp;
826
827 /* Discard unused keys */
828 spin_lock(&vnode->wb_lock);
829
830 if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
831 !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
832 list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
833 if (refcount_read(&wbk->usage) == 1)
834 list_move(&wbk->vnode_link, &graveyard);
835 }
836 }
837
838 spin_unlock(&vnode->wb_lock);
839
840 while (!list_empty(&graveyard)) {
841 wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
842 list_del(&wbk->vnode_link);
843 afs_put_wb_key(wbk);
844 }
845}
846
847/*
848 * Clean up a page during invalidation.
849 */
850int afs_launder_page(struct page *page)
851{
852 struct address_space *mapping = page->mapping;
853 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
854 unsigned long priv;
855 unsigned int f, t;
856 int ret = 0;
857
858 _enter("{%lx}", page->index);
859
860 priv = page_private(page);
861 if (clear_page_dirty_for_io(page)) {
862 f = 0;
863 t = PAGE_SIZE;
864 if (PagePrivate(page)) {
865 f = priv & AFS_PRIV_MAX;
866 t = priv >> AFS_PRIV_SHIFT;
867 }
868
869 trace_afs_page_dirty(vnode, tracepoint_string("launder"),
870 page->index, priv);
871 ret = afs_store_data(mapping, page->index, page->index, t, f);
872 }
873
874 trace_afs_page_dirty(vnode, tracepoint_string("laundered"),
875 page->index, priv);
876 set_page_private(page, 0);
877 ClearPagePrivate(page);
878
879#ifdef CONFIG_AFS_FSCACHE
880 if (PageFsCache(page)) {
881 fscache_wait_on_page_write(vnode->cache, page);
882 fscache_uncache_page(vnode->cache, page);
883 }
884#endif
885 return ret;
886}