blob: dcad5e2105252fa277a1c66ef83d8a960d2a082b [file] [log] [blame]
Miklos Szeredib6aeade2005-09-09 13:10:30 -07001/*
2 FUSE: Filesystem in Userspace
Miklos Szeredi1729a162008-11-26 12:03:54 +01003 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
Miklos Szeredib6aeade2005-09-09 13:10:30 -07004
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/pagemap.h>
12#include <linux/slab.h>
13#include <linux/kernel.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040014#include <linux/sched.h>
Tejun Heo08cbf542009-04-14 10:54:53 +090015#include <linux/module.h>
Miklos Szeredid9d318d2010-11-30 16:39:27 +010016#include <linux/compat.h>
Johannes Weiner478e0842011-07-25 22:35:35 +020017#include <linux/swap.h>
Brian Foster3634a632013-05-17 09:30:32 -040018#include <linux/falloc.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080019#include <linux/uio.h>
Miklos Szeredib6aeade2005-09-09 13:10:30 -070020
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080021static const struct file_operations fuse_direct_io_file_operations;
Miklos Szeredi45323fb2005-09-09 13:10:37 -070022
Miklos Szeredi91fe96b2009-04-28 16:56:37 +020023static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
24 int opcode, struct fuse_open_out *outargp)
Miklos Szeredib6aeade2005-09-09 13:10:30 -070025{
Miklos Szeredib6aeade2005-09-09 13:10:30 -070026 struct fuse_open_in inarg;
Miklos Szeredi70781872014-12-12 09:49:05 +010027 FUSE_ARGS(args);
Miklos Szeredifd72faa2005-11-07 00:59:51 -080028
29 memset(&inarg, 0, sizeof(inarg));
Miklos Szeredi6ff958e2007-10-18 03:07:02 -070030 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
31 if (!fc->atomic_o_trunc)
32 inarg.flags &= ~O_TRUNC;
Miklos Szeredi70781872014-12-12 09:49:05 +010033 args.in.h.opcode = opcode;
34 args.in.h.nodeid = nodeid;
35 args.in.numargs = 1;
36 args.in.args[0].size = sizeof(inarg);
37 args.in.args[0].value = &inarg;
38 args.out.numargs = 1;
39 args.out.args[0].size = sizeof(*outargp);
40 args.out.args[0].value = outargp;
Miklos Szeredifd72faa2005-11-07 00:59:51 -080041
Miklos Szeredi70781872014-12-12 09:49:05 +010042 return fuse_simple_request(fc, &args);
Miklos Szeredifd72faa2005-11-07 00:59:51 -080043}
44
Tejun Heoacf99432008-11-26 12:03:55 +010045struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
Miklos Szeredifd72faa2005-11-07 00:59:51 -080046{
47 struct fuse_file *ff;
Tejun Heo6b2db282009-04-14 10:54:49 +090048
Miklos Szeredifd72faa2005-11-07 00:59:51 -080049 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
Tejun Heo6b2db282009-04-14 10:54:49 +090050 if (unlikely(!ff))
51 return NULL;
52
Miklos Szeredida5e4712009-04-28 16:56:36 +020053 ff->fc = fc;
Maxim Patlasov4250c062012-10-26 19:48:07 +040054 ff->reserved_req = fuse_request_alloc(0);
Tejun Heo6b2db282009-04-14 10:54:49 +090055 if (unlikely(!ff->reserved_req)) {
56 kfree(ff);
57 return NULL;
Miklos Szeredifd72faa2005-11-07 00:59:51 -080058 }
Tejun Heo6b2db282009-04-14 10:54:49 +090059
60 INIT_LIST_HEAD(&ff->write_entry);
61 atomic_set(&ff->count, 0);
62 RB_CLEAR_NODE(&ff->polled_node);
63 init_waitqueue_head(&ff->poll_wait);
64
65 spin_lock(&fc->lock);
66 ff->kh = ++fc->khctr;
67 spin_unlock(&fc->lock);
68
Miklos Szeredifd72faa2005-11-07 00:59:51 -080069 return ff;
70}
71
72void fuse_file_free(struct fuse_file *ff)
73{
Miklos Szeredi33649c92006-06-25 05:48:52 -070074 fuse_request_free(ff->reserved_req);
Miklos Szeredifd72faa2005-11-07 00:59:51 -080075 kfree(ff);
76}
77
Miklos Szeredic7b71432009-04-28 16:56:37 +020078struct fuse_file *fuse_file_get(struct fuse_file *ff)
Miklos Szeredic756e0a2007-10-16 23:31:00 -070079{
80 atomic_inc(&ff->count);
81 return ff;
82}
83
Miklos Szeredi5a18ec12011-02-25 14:44:58 +010084static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
85{
Miklos Szeredibaebccb2014-12-12 09:49:04 +010086 iput(req->misc.release.inode);
Miklos Szeredi5a18ec12011-02-25 14:44:58 +010087}
88
89static void fuse_file_put(struct fuse_file *ff, bool sync)
Miklos Szeredic756e0a2007-10-16 23:31:00 -070090{
91 if (atomic_dec_and_test(&ff->count)) {
92 struct fuse_req *req = ff->reserved_req;
Miklos Szeredi8b0797a2009-04-28 16:56:39 +020093
Andrew Gallagher7678ac52013-11-05 16:05:52 +010094 if (ff->fc->no_open) {
95 /*
96 * Drop the release request when client does not
97 * implement 'open'
98 */
Miklos Szeredi825d6d32015-07-01 16:25:58 +020099 __clear_bit(FR_BACKGROUND, &req->flags);
Miklos Szeredibaebccb2014-12-12 09:49:04 +0100100 iput(req->misc.release.inode);
Andrew Gallagher7678ac52013-11-05 16:05:52 +0100101 fuse_put_request(ff->fc, req);
102 } else if (sync) {
Miklos Szeredi825d6d32015-07-01 16:25:58 +0200103 __clear_bit(FR_BACKGROUND, &req->flags);
Miklos Szeredi5a18ec12011-02-25 14:44:58 +0100104 fuse_request_send(ff->fc, req);
Miklos Szeredibaebccb2014-12-12 09:49:04 +0100105 iput(req->misc.release.inode);
Miklos Szeredi5a18ec12011-02-25 14:44:58 +0100106 fuse_put_request(ff->fc, req);
107 } else {
108 req->end = fuse_release_end;
Miklos Szeredi825d6d32015-07-01 16:25:58 +0200109 __set_bit(FR_BACKGROUND, &req->flags);
Miklos Szeredi5a18ec12011-02-25 14:44:58 +0100110 fuse_request_send_background(ff->fc, req);
111 }
Miklos Szeredic756e0a2007-10-16 23:31:00 -0700112 kfree(ff);
113 }
114}
115
Tejun Heo08cbf542009-04-14 10:54:53 +0900116int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
117 bool isdir)
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200118{
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200119 struct fuse_file *ff;
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200120 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
121
122 ff = fuse_file_alloc(fc);
123 if (!ff)
124 return -ENOMEM;
125
Andrew Gallagher7678ac52013-11-05 16:05:52 +0100126 ff->fh = 0;
127 ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */
128 if (!fc->no_open || isdir) {
129 struct fuse_open_out outarg;
130 int err;
131
132 err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
133 if (!err) {
134 ff->fh = outarg.fh;
135 ff->open_flags = outarg.open_flags;
136
137 } else if (err != -ENOSYS || isdir) {
138 fuse_file_free(ff);
139 return err;
140 } else {
141 fc->no_open = 1;
142 }
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200143 }
144
145 if (isdir)
Andrew Gallagher7678ac52013-11-05 16:05:52 +0100146 ff->open_flags &= ~FOPEN_DIRECT_IO;
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200147
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200148 ff->nodeid = nodeid;
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200149 file->private_data = fuse_file_get(ff);
150
151 return 0;
152}
Tejun Heo08cbf542009-04-14 10:54:53 +0900153EXPORT_SYMBOL_GPL(fuse_do_open);
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200154
Pavel Emelyanov650b22b2013-10-10 17:10:04 +0400155static void fuse_link_write_file(struct file *file)
156{
157 struct inode *inode = file_inode(file);
158 struct fuse_conn *fc = get_fuse_conn(inode);
159 struct fuse_inode *fi = get_fuse_inode(inode);
160 struct fuse_file *ff = file->private_data;
161 /*
162 * file may be written through mmap, so chain it onto the
163 * inodes's write_file list
164 */
165 spin_lock(&fc->lock);
166 if (list_empty(&ff->write_entry))
167 list_add(&ff->write_entry, &fi->write_files);
168 spin_unlock(&fc->lock);
169}
170
Miklos Szeredic7b71432009-04-28 16:56:37 +0200171void fuse_finish_open(struct inode *inode, struct file *file)
Miklos Szeredifd72faa2005-11-07 00:59:51 -0800172{
Miklos Szeredic7b71432009-04-28 16:56:37 +0200173 struct fuse_file *ff = file->private_data;
Ken Sumralla0822c52010-11-24 12:57:00 -0800174 struct fuse_conn *fc = get_fuse_conn(inode);
Miklos Szeredic7b71432009-04-28 16:56:37 +0200175
176 if (ff->open_flags & FOPEN_DIRECT_IO)
Miklos Szeredifd72faa2005-11-07 00:59:51 -0800177 file->f_op = &fuse_direct_io_file_operations;
Miklos Szeredic7b71432009-04-28 16:56:37 +0200178 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
Miklos Szeredib1009972007-10-16 23:31:01 -0700179 invalidate_inode_pages2(inode->i_mapping);
Miklos Szeredic7b71432009-04-28 16:56:37 +0200180 if (ff->open_flags & FOPEN_NONSEEKABLE)
Tejun Heoa7c1b992008-10-16 16:08:57 +0200181 nonseekable_open(inode, file);
Ken Sumralla0822c52010-11-24 12:57:00 -0800182 if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
183 struct fuse_inode *fi = get_fuse_inode(inode);
184
185 spin_lock(&fc->lock);
186 fi->attr_version = ++fc->attr_version;
187 i_size_write(inode, 0);
188 spin_unlock(&fc->lock);
189 fuse_invalidate_attr(inode);
Maxim Patlasov75caeec2014-04-28 14:19:22 +0200190 if (fc->writeback_cache)
191 file_update_time(file);
Ken Sumralla0822c52010-11-24 12:57:00 -0800192 }
Pavel Emelyanov4d99ff82013-10-10 17:12:18 +0400193 if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache)
194 fuse_link_write_file(file);
Miklos Szeredifd72faa2005-11-07 00:59:51 -0800195}
196
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200197int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
Miklos Szeredifd72faa2005-11-07 00:59:51 -0800198{
Tejun Heoacf99432008-11-26 12:03:55 +0100199 struct fuse_conn *fc = get_fuse_conn(inode);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700200 int err;
Maxim Patlasov75caeec2014-04-28 14:19:22 +0200201 bool lock_inode = (file->f_flags & O_TRUNC) &&
202 fc->atomic_o_trunc &&
203 fc->writeback_cache;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700204
205 err = generic_file_open(inode, file);
206 if (err)
207 return err;
208
Maxim Patlasov75caeec2014-04-28 14:19:22 +0200209 if (lock_inode)
Al Viro59551022016-01-22 15:40:57 -0500210 inode_lock(inode);
Maxim Patlasov75caeec2014-04-28 14:19:22 +0200211
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200212 err = fuse_do_open(fc, get_node_id(inode), file, isdir);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700213
Maxim Patlasov75caeec2014-04-28 14:19:22 +0200214 if (!err)
215 fuse_finish_open(inode, file);
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200216
Maxim Patlasov75caeec2014-04-28 14:19:22 +0200217 if (lock_inode)
Al Viro59551022016-01-22 15:40:57 -0500218 inode_unlock(inode);
Maxim Patlasov75caeec2014-04-28 14:19:22 +0200219
220 return err;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700221}
222
Miklos Szeredi8b0797a2009-04-28 16:56:39 +0200223static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
Miklos Szeredi64c6d8e2006-01-16 22:14:42 -0800224{
Miklos Szeredi8b0797a2009-04-28 16:56:39 +0200225 struct fuse_conn *fc = ff->fc;
Miklos Szeredi33649c92006-06-25 05:48:52 -0700226 struct fuse_req *req = ff->reserved_req;
Miklos Szeredib57d4262008-02-06 01:38:39 -0800227 struct fuse_release_in *inarg = &req->misc.release.in;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700228
Miklos Szeredi8b0797a2009-04-28 16:56:39 +0200229 spin_lock(&fc->lock);
230 list_del(&ff->write_entry);
231 if (!RB_EMPTY_NODE(&ff->polled_node))
232 rb_erase(&ff->polled_node, &fc->polled_files);
233 spin_unlock(&fc->lock);
234
Bryan Green357ccf22011-03-01 16:43:52 -0800235 wake_up_interruptible_all(&ff->poll_wait);
Miklos Szeredi8b0797a2009-04-28 16:56:39 +0200236
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700237 inarg->fh = ff->fh;
Miklos Szeredifd72faa2005-11-07 00:59:51 -0800238 inarg->flags = flags;
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700239 req->in.h.opcode = opcode;
Miklos Szeredic7b71432009-04-28 16:56:37 +0200240 req->in.h.nodeid = ff->nodeid;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700241 req->in.numargs = 1;
242 req->in.args[0].size = sizeof(struct fuse_release_in);
243 req->in.args[0].value = inarg;
Miklos Szeredifd72faa2005-11-07 00:59:51 -0800244}
245
Miklos Szeredi8b0797a2009-04-28 16:56:39 +0200246void fuse_release_common(struct file *file, int opcode)
Miklos Szeredifd72faa2005-11-07 00:59:51 -0800247{
Tejun Heo6b2db282009-04-14 10:54:49 +0900248 struct fuse_file *ff;
249 struct fuse_req *req;
Miklos Szeredi93a8c3c2007-10-18 03:07:03 -0700250
Tejun Heo6b2db282009-04-14 10:54:49 +0900251 ff = file->private_data;
252 if (unlikely(!ff))
Miklos Szeredi8b0797a2009-04-28 16:56:39 +0200253 return;
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700254
Tejun Heo6b2db282009-04-14 10:54:49 +0900255 req = ff->reserved_req;
Miklos Szeredi8b0797a2009-04-28 16:56:39 +0200256 fuse_prepare_release(ff, file->f_flags, opcode);
Tejun Heo95668a62008-11-26 12:03:55 +0100257
Miklos Szeredi37fb3a32011-08-08 16:08:08 +0200258 if (ff->flock) {
259 struct fuse_release_in *inarg = &req->misc.release.in;
260 inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
261 inarg->lock_owner = fuse_lock_owner_id(ff->fc,
262 (fl_owner_t) file);
263 }
Miklos Szeredibaebccb2014-12-12 09:49:04 +0100264 /* Hold inode until release is finished */
265 req->misc.release.inode = igrab(file_inode(file));
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700266
Tejun Heo6b2db282009-04-14 10:54:49 +0900267 /*
268 * Normally this will send the RELEASE request, however if
269 * some asynchronous READ or WRITE requests are outstanding,
270 * the sending will be delayed.
Miklos Szeredi5a18ec12011-02-25 14:44:58 +0100271 *
272 * Make the release synchronous if this is a fuseblk mount,
273 * synchronous RELEASE is allowed (and desirable) in this case
274 * because the server can be trusted not to screw up.
Tejun Heo6b2db282009-04-14 10:54:49 +0900275 */
Miklos Szeredi5a18ec12011-02-25 14:44:58 +0100276 fuse_file_put(ff, ff->fc->destroy_req != NULL);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700277}
278
Miklos Szeredi04730fe2005-09-09 13:10:36 -0700279static int fuse_open(struct inode *inode, struct file *file)
280{
Miklos Szeredi91fe96b2009-04-28 16:56:37 +0200281 return fuse_open_common(inode, file, false);
Miklos Szeredi04730fe2005-09-09 13:10:36 -0700282}
283
284static int fuse_release(struct inode *inode, struct file *file)
285{
Pavel Emelyanove7cc133c2013-10-10 17:19:06 +0400286 struct fuse_conn *fc = get_fuse_conn(inode);
287
288 /* see fuse_vma_close() for !writeback_cache case */
289 if (fc->writeback_cache)
Miklos Szeredi1e18bda2014-04-28 14:19:23 +0200290 write_inode_now(inode, 1);
Maxim Patlasovb0aa7602013-12-26 19:51:11 +0400291
Miklos Szeredi8b0797a2009-04-28 16:56:39 +0200292 fuse_release_common(file, FUSE_RELEASE);
293
294 /* return value is ignored by VFS */
295 return 0;
296}
297
298void fuse_sync_release(struct fuse_file *ff, int flags)
299{
300 WARN_ON(atomic_read(&ff->count) > 1);
301 fuse_prepare_release(ff, flags, FUSE_RELEASE);
Miklos Szeredi825d6d32015-07-01 16:25:58 +0200302 __set_bit(FR_FORCE, &ff->reserved_req->flags);
303 __clear_bit(FR_BACKGROUND, &ff->reserved_req->flags);
Miklos Szeredi8b0797a2009-04-28 16:56:39 +0200304 fuse_request_send(ff->fc, ff->reserved_req);
305 fuse_put_request(ff->fc, ff->reserved_req);
306 kfree(ff);
Miklos Szeredi04730fe2005-09-09 13:10:36 -0700307}
Tejun Heo08cbf542009-04-14 10:54:53 +0900308EXPORT_SYMBOL_GPL(fuse_sync_release);
Miklos Szeredi04730fe2005-09-09 13:10:36 -0700309
Miklos Szeredi71421252006-06-25 05:48:52 -0700310/*
Miklos Szeredi9c8ef562006-06-25 05:48:55 -0700311 * Scramble the ID space with XTEA, so that the value of the files_struct
312 * pointer is not exposed to userspace.
Miklos Szeredi71421252006-06-25 05:48:52 -0700313 */
Miklos Szeredif3332112007-10-18 03:07:04 -0700314u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
Miklos Szeredi71421252006-06-25 05:48:52 -0700315{
Miklos Szeredi9c8ef562006-06-25 05:48:55 -0700316 u32 *k = fc->scramble_key;
317 u64 v = (unsigned long) id;
318 u32 v0 = v;
319 u32 v1 = v >> 32;
320 u32 sum = 0;
321 int i;
322
323 for (i = 0; i < 32; i++) {
324 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
325 sum += 0x9E3779B9;
326 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
327 }
328
329 return (u64) v0 + ((u64) v1 << 32);
Miklos Szeredi71421252006-06-25 05:48:52 -0700330}
331
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700332/*
Pavel Emelyanovea8cd332013-10-10 17:12:05 +0400333 * Check if any page in a range is under writeback
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700334 *
335 * This is currently done by walking the list of writepage requests
336 * for the inode, which can be pretty inefficient.
337 */
Pavel Emelyanovea8cd332013-10-10 17:12:05 +0400338static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
339 pgoff_t idx_to)
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700340{
341 struct fuse_conn *fc = get_fuse_conn(inode);
342 struct fuse_inode *fi = get_fuse_inode(inode);
343 struct fuse_req *req;
344 bool found = false;
345
346 spin_lock(&fc->lock);
347 list_for_each_entry(req, &fi->writepages, writepages_entry) {
348 pgoff_t curr_index;
349
350 BUG_ON(req->inode != inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300351 curr_index = req->misc.write.in.offset >> PAGE_SHIFT;
Pavel Emelyanovea8cd332013-10-10 17:12:05 +0400352 if (idx_from < curr_index + req->num_pages &&
353 curr_index <= idx_to) {
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700354 found = true;
355 break;
356 }
357 }
358 spin_unlock(&fc->lock);
359
360 return found;
361}
362
Pavel Emelyanovea8cd332013-10-10 17:12:05 +0400363static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
364{
365 return fuse_range_is_writeback(inode, index, index);
366}
367
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700368/*
369 * Wait for page writeback to be completed.
370 *
371 * Since fuse doesn't rely on the VM writeback tracking, this has to
372 * use some other means.
373 */
374static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
375{
376 struct fuse_inode *fi = get_fuse_inode(inode);
377
378 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
379 return 0;
380}
381
Maxim Patlasovfe38d7d2013-10-10 17:11:54 +0400382/*
383 * Wait for all pending writepages on the inode to finish.
384 *
385 * This is currently done by blocking further writes with FUSE_NOWRITE
386 * and waiting for all sent writes to complete.
387 *
388 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
389 * could conflict with truncation.
390 */
391static void fuse_sync_writes(struct inode *inode)
392{
393 fuse_set_nowrite(inode);
394 fuse_release_nowrite(inode);
395}
396
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700397static int fuse_flush(struct file *file, fl_owner_t id)
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700398{
Al Viro6131ffa2013-02-27 16:59:05 -0500399 struct inode *inode = file_inode(file);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700400 struct fuse_conn *fc = get_fuse_conn(inode);
401 struct fuse_file *ff = file->private_data;
402 struct fuse_req *req;
403 struct fuse_flush_in inarg;
404 int err;
405
Miklos Szeredi248d86e2006-01-06 00:19:39 -0800406 if (is_bad_inode(inode))
407 return -EIO;
408
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700409 if (fc->no_flush)
410 return 0;
411
Miklos Szeredi1e18bda2014-04-28 14:19:23 +0200412 err = write_inode_now(inode, 1);
Maxim Patlasovfe38d7d2013-10-10 17:11:54 +0400413 if (err)
414 return err;
415
Al Viro59551022016-01-22 15:40:57 -0500416 inode_lock(inode);
Maxim Patlasovfe38d7d2013-10-10 17:11:54 +0400417 fuse_sync_writes(inode);
Al Viro59551022016-01-22 15:40:57 -0500418 inode_unlock(inode);
Maxim Patlasovfe38d7d2013-10-10 17:11:54 +0400419
Maxim Patlasovb111c8c2012-10-26 19:48:30 +0400420 req = fuse_get_req_nofail_nopages(fc, file);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700421 memset(&inarg, 0, sizeof(inarg));
422 inarg.fh = ff->fh;
Miklos Szeredi9c8ef562006-06-25 05:48:55 -0700423 inarg.lock_owner = fuse_lock_owner_id(fc, id);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700424 req->in.h.opcode = FUSE_FLUSH;
425 req->in.h.nodeid = get_node_id(inode);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700426 req->in.numargs = 1;
427 req->in.args[0].size = sizeof(inarg);
428 req->in.args[0].value = &inarg;
Miklos Szeredi825d6d32015-07-01 16:25:58 +0200429 __set_bit(FR_FORCE, &req->flags);
Tejun Heob93f8582008-11-26 12:03:55 +0100430 fuse_request_send(fc, req);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700431 err = req->out.h.error;
432 fuse_put_request(fc, req);
433 if (err == -ENOSYS) {
434 fc->no_flush = 1;
435 err = 0;
436 }
437 return err;
438}
439
Josef Bacik02c24a82011-07-16 20:44:56 -0400440int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
441 int datasync, int isdir)
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700442{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200443 struct inode *inode = file->f_mapping->host;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700444 struct fuse_conn *fc = get_fuse_conn(inode);
445 struct fuse_file *ff = file->private_data;
Miklos Szeredi70781872014-12-12 09:49:05 +0100446 FUSE_ARGS(args);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700447 struct fuse_fsync_in inarg;
448 int err;
449
Miklos Szeredi248d86e2006-01-06 00:19:39 -0800450 if (is_bad_inode(inode))
451 return -EIO;
452
Al Viro59551022016-01-22 15:40:57 -0500453 inode_lock(inode);
Josef Bacik02c24a82011-07-16 20:44:56 -0400454
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700455 /*
456 * Start writeback against all dirty pages of the inode, then
457 * wait for all outstanding writes, before sending the FSYNC
458 * request.
459 */
Miklos Szeredi22401e72014-04-28 14:19:23 +0200460 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700461 if (err)
Josef Bacik02c24a82011-07-16 20:44:56 -0400462 goto out;
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700463
464 fuse_sync_writes(inode);
Miklos Szeredi1e18bda2014-04-28 14:19:23 +0200465 err = sync_inode_metadata(inode, 1);
466 if (err)
467 goto out;
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700468
Miklos Szeredi22401e72014-04-28 14:19:23 +0200469 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
470 goto out;
Maxim Patlasovb0aa7602013-12-26 19:51:11 +0400471
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700472 memset(&inarg, 0, sizeof(inarg));
473 inarg.fh = ff->fh;
474 inarg.fsync_flags = datasync ? 1 : 0;
Miklos Szeredi70781872014-12-12 09:49:05 +0100475 args.in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
476 args.in.h.nodeid = get_node_id(inode);
477 args.in.numargs = 1;
478 args.in.args[0].size = sizeof(inarg);
479 args.in.args[0].value = &inarg;
480 err = fuse_simple_request(fc, &args);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700481 if (err == -ENOSYS) {
Miklos Szeredi82547982005-09-09 13:10:38 -0700482 if (isdir)
483 fc->no_fsyncdir = 1;
484 else
485 fc->no_fsync = 1;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700486 err = 0;
487 }
Josef Bacik02c24a82011-07-16 20:44:56 -0400488out:
Al Viro59551022016-01-22 15:40:57 -0500489 inode_unlock(inode);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700490 return err;
491}
492
Josef Bacik02c24a82011-07-16 20:44:56 -0400493static int fuse_fsync(struct file *file, loff_t start, loff_t end,
494 int datasync)
Miklos Szeredi82547982005-09-09 13:10:38 -0700495{
Josef Bacik02c24a82011-07-16 20:44:56 -0400496 return fuse_fsync_common(file, start, end, datasync, 0);
Miklos Szeredi82547982005-09-09 13:10:38 -0700497}
498
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200499void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
500 size_t count, int opcode)
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700501{
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700502 struct fuse_read_in *inarg = &req->misc.read.in;
Miklos Szeredia6643092007-11-28 16:22:00 -0800503 struct fuse_file *ff = file->private_data;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700504
Miklos Szeredi361b1eb52006-01-16 22:14:45 -0800505 inarg->fh = ff->fh;
506 inarg->offset = pos;
507 inarg->size = count;
Miklos Szeredia6643092007-11-28 16:22:00 -0800508 inarg->flags = file->f_flags;
Miklos Szeredi361b1eb52006-01-16 22:14:45 -0800509 req->in.h.opcode = opcode;
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200510 req->in.h.nodeid = ff->nodeid;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700511 req->in.numargs = 1;
512 req->in.args[0].size = sizeof(struct fuse_read_in);
Miklos Szeredic1aa96a2006-01-16 22:14:46 -0800513 req->in.args[0].value = inarg;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700514 req->out.argvar = 1;
515 req->out.numargs = 1;
516 req->out.args[0].size = count;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700517}
518
Maxim Patlasov187c5c32012-12-14 19:20:25 +0400519static void fuse_release_user_pages(struct fuse_req *req, int write)
520{
521 unsigned i;
522
523 for (i = 0; i < req->num_pages; i++) {
524 struct page *page = req->pages[i];
525 if (write)
526 set_page_dirty_lock(page);
527 put_page(page);
528 }
529}
530
Seth Forshee744742d2016-03-11 10:35:34 -0600531static void fuse_io_release(struct kref *kref)
532{
533 kfree(container_of(kref, struct fuse_io_priv, refcnt));
534}
535
Christoph Hellwig9d5722b2015-02-02 14:59:43 +0100536static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io)
537{
538 if (io->err)
539 return io->err;
540
541 if (io->bytes >= 0 && io->write)
542 return -EIO;
543
544 return io->bytes < 0 ? io->size : io->bytes;
545}
546
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400547/**
548 * In case of short read, the caller sets 'pos' to the position of
549 * actual end of fuse request in IO request. Otherwise, if bytes_requested
550 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
551 *
552 * An example:
553 * User requested DIO read of 64K. It was splitted into two 32K fuse requests,
554 * both submitted asynchronously. The first of them was ACKed by userspace as
555 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
556 * second request was ACKed as short, e.g. only 1K was read, resulting in
557 * pos == 33K.
558 *
559 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
560 * will be equal to the length of the longest contiguous fragment of
561 * transferred data starting from the beginning of IO request.
562 */
563static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
564{
Christoph Hellwig9d5722b2015-02-02 14:59:43 +0100565 bool is_sync = is_sync_kiocb(io->iocb);
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400566 int left;
567
568 spin_lock(&io->lock);
569 if (err)
570 io->err = io->err ? : err;
571 else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes))
572 io->bytes = pos;
573
574 left = --io->reqs;
Christoph Hellwig9d5722b2015-02-02 14:59:43 +0100575 if (!left && is_sync)
576 complete(io->done);
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400577 spin_unlock(&io->lock);
578
Christoph Hellwig9d5722b2015-02-02 14:59:43 +0100579 if (!left && !is_sync) {
580 ssize_t res = fuse_get_res_by_io(io);
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400581
Christoph Hellwig9d5722b2015-02-02 14:59:43 +0100582 if (res >= 0) {
583 struct inode *inode = file_inode(io->iocb->ki_filp);
584 struct fuse_conn *fc = get_fuse_conn(inode);
585 struct fuse_inode *fi = get_fuse_inode(inode);
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400586
Christoph Hellwig9d5722b2015-02-02 14:59:43 +0100587 spin_lock(&fc->lock);
588 fi->attr_version = ++fc->attr_version;
589 spin_unlock(&fc->lock);
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400590 }
591
Christoph Hellwig04b2fa92015-02-02 14:49:06 +0100592 io->iocb->ki_complete(io->iocb, res, 0);
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400593 }
Seth Forshee744742d2016-03-11 10:35:34 -0600594
595 kref_put(&io->refcnt, fuse_io_release);
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400596}
597
598static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req)
599{
600 struct fuse_io_priv *io = req->io;
601 ssize_t pos = -1;
602
603 fuse_release_user_pages(req, !io->write);
604
605 if (io->write) {
606 if (req->misc.write.in.size != req->misc.write.out.size)
607 pos = req->misc.write.in.offset - io->offset +
608 req->misc.write.out.size;
609 } else {
610 if (req->misc.read.in.size != req->out.args[0].size)
611 pos = req->misc.read.in.offset - io->offset +
612 req->out.args[0].size;
613 }
614
615 fuse_aio_complete(io, req->out.h.error, pos);
616}
617
618static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req,
619 size_t num_bytes, struct fuse_io_priv *io)
620{
621 spin_lock(&io->lock);
Seth Forshee744742d2016-03-11 10:35:34 -0600622 kref_get(&io->refcnt);
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400623 io->size += num_bytes;
624 io->reqs++;
625 spin_unlock(&io->lock);
626
627 req->io = io;
628 req->end = fuse_aio_complete_req;
629
Maxim Patlasov36cf66e2012-12-14 19:20:51 +0400630 __fuse_get_request(req);
Maxim Patlasov01e9d112012-12-14 19:20:41 +0400631 fuse_request_send_background(fc, req);
632
633 return num_bytes;
634}
635
Maxim Patlasov36cf66e2012-12-14 19:20:51 +0400636static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io,
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200637 loff_t pos, size_t count, fl_owner_t owner)
Miklos Szeredi04730fe2005-09-09 13:10:36 -0700638{
Maxim Patlasov36cf66e2012-12-14 19:20:51 +0400639 struct file *file = io->file;
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200640 struct fuse_file *ff = file->private_data;
641 struct fuse_conn *fc = ff->fc;
Miklos Szeredif3332112007-10-18 03:07:04 -0700642
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200643 fuse_read_fill(req, file, pos, count, FUSE_READ);
Miklos Szeredif3332112007-10-18 03:07:04 -0700644 if (owner != NULL) {
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700645 struct fuse_read_in *inarg = &req->misc.read.in;
Miklos Szeredif3332112007-10-18 03:07:04 -0700646
647 inarg->read_flags |= FUSE_READ_LOCKOWNER;
648 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
649 }
Maxim Patlasov36cf66e2012-12-14 19:20:51 +0400650
651 if (io->async)
652 return fuse_async_req_send(fc, req, count, io);
653
Tejun Heob93f8582008-11-26 12:03:55 +0100654 fuse_request_send(fc, req);
Miklos Szeredi361b1eb52006-01-16 22:14:45 -0800655 return req->out.args[0].size;
Miklos Szeredi04730fe2005-09-09 13:10:36 -0700656}
657
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700658static void fuse_read_update_size(struct inode *inode, loff_t size,
659 u64 attr_ver)
660{
661 struct fuse_conn *fc = get_fuse_conn(inode);
662 struct fuse_inode *fi = get_fuse_inode(inode);
663
664 spin_lock(&fc->lock);
Maxim Patlasov06a7c3c2013-08-30 17:06:04 +0400665 if (attr_ver == fi->attr_version && size < inode->i_size &&
666 !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700667 fi->attr_version = ++fc->attr_version;
668 i_size_write(inode, size);
669 }
670 spin_unlock(&fc->lock);
671}
672
Pavel Emelyanova92adc82013-10-10 17:10:16 +0400673static void fuse_short_read(struct fuse_req *req, struct inode *inode,
674 u64 attr_ver)
675{
676 size_t num_read = req->out.args[0].size;
Pavel Emelyanov83732002013-10-10 17:10:46 +0400677 struct fuse_conn *fc = get_fuse_conn(inode);
Pavel Emelyanova92adc82013-10-10 17:10:16 +0400678
Pavel Emelyanov83732002013-10-10 17:10:46 +0400679 if (fc->writeback_cache) {
680 /*
681 * A hole in a file. Some data after the hole are in page cache,
682 * but have not reached the client fs yet. So, the hole is not
683 * present there.
684 */
685 int i;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300686 int start_idx = num_read >> PAGE_SHIFT;
687 size_t off = num_read & (PAGE_SIZE - 1);
Pavel Emelyanov83732002013-10-10 17:10:46 +0400688
689 for (i = start_idx; i < req->num_pages; i++) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300690 zero_user_segment(req->pages[i], off, PAGE_SIZE);
Pavel Emelyanov83732002013-10-10 17:10:46 +0400691 off = 0;
692 }
693 } else {
694 loff_t pos = page_offset(req->pages[0]) + num_read;
695 fuse_read_update_size(inode, pos, attr_ver);
696 }
Pavel Emelyanova92adc82013-10-10 17:10:16 +0400697}
698
Maxim Patlasov482fce52013-10-10 17:11:25 +0400699static int fuse_do_readpage(struct file *file, struct page *page)
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700700{
Seth Forshee744742d2016-03-11 10:35:34 -0600701 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(file);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700702 struct inode *inode = page->mapping->host;
703 struct fuse_conn *fc = get_fuse_conn(inode);
Miklos Szeredi248d86e2006-01-06 00:19:39 -0800704 struct fuse_req *req;
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700705 size_t num_read;
706 loff_t pos = page_offset(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300707 size_t count = PAGE_SIZE;
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700708 u64 attr_ver;
Miklos Szeredi248d86e2006-01-06 00:19:39 -0800709 int err;
710
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700711 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300712 * Page writeback can extend beyond the lifetime of the
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700713 * page-cache page, so make sure we read a properly synced
714 * page.
715 */
716 fuse_wait_on_page_writeback(inode, page->index);
717
Maxim Patlasovb111c8c2012-10-26 19:48:30 +0400718 req = fuse_get_req(fc, 1);
Miklos Szeredice1d5a42006-04-10 22:54:58 -0700719 if (IS_ERR(req))
Maxim Patlasov482fce52013-10-10 17:11:25 +0400720 return PTR_ERR(req);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700721
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700722 attr_ver = fuse_get_attr_version(fc);
723
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700724 req->out.page_zeroing = 1;
Miklos Szeredif4975c62009-04-02 14:25:34 +0200725 req->out.argpages = 1;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700726 req->num_pages = 1;
727 req->pages[0] = page;
Maxim Patlasov85f40ae2012-10-26 19:49:33 +0400728 req->page_descs[0].length = count;
Maxim Patlasov36cf66e2012-12-14 19:20:51 +0400729 num_read = fuse_send_read(req, &io, pos, count, NULL);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700730 err = req->out.h.error;
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700731
732 if (!err) {
733 /*
734 * Short read means EOF. If file size is larger, truncate it
735 */
736 if (num_read < count)
Pavel Emelyanova92adc82013-10-10 17:10:16 +0400737 fuse_short_read(req, inode, attr_ver);
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700738
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700739 SetPageUptodate(page);
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700740 }
741
Pavel Emelyanova92adc82013-10-10 17:10:16 +0400742 fuse_put_request(fc, req);
Maxim Patlasov482fce52013-10-10 17:11:25 +0400743
744 return err;
745}
746
747static int fuse_readpage(struct file *file, struct page *page)
748{
749 struct inode *inode = page->mapping->host;
750 int err;
751
752 err = -EIO;
753 if (is_bad_inode(inode))
754 goto out;
755
756 err = fuse_do_readpage(file, page);
Andrew Gallagher451418f2013-11-05 03:55:43 -0800757 fuse_invalidate_atime(inode);
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700758 out:
759 unlock_page(page);
760 return err;
761}
762
Miklos Szeredic1aa96a2006-01-16 22:14:46 -0800763static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredidb50b962005-09-09 13:10:33 -0700764{
Miklos Szeredic1aa96a2006-01-16 22:14:46 -0800765 int i;
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700766 size_t count = req->misc.read.in.size;
767 size_t num_read = req->out.args[0].size;
Miklos Szeredice534fb2010-05-25 15:06:07 +0200768 struct address_space *mapping = NULL;
Miklos Szeredic1aa96a2006-01-16 22:14:46 -0800769
Miklos Szeredice534fb2010-05-25 15:06:07 +0200770 for (i = 0; mapping == NULL && i < req->num_pages; i++)
771 mapping = req->pages[i]->mapping;
772
773 if (mapping) {
774 struct inode *inode = mapping->host;
775
776 /*
777 * Short read means EOF. If file size is larger, truncate it
778 */
Pavel Emelyanova92adc82013-10-10 17:10:16 +0400779 if (!req->out.h.error && num_read < count)
780 fuse_short_read(req, inode, req->misc.read.attr_ver);
Miklos Szeredice534fb2010-05-25 15:06:07 +0200781
Andrew Gallagher451418f2013-11-05 03:55:43 -0800782 fuse_invalidate_atime(inode);
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700783 }
784
Miklos Szeredidb50b962005-09-09 13:10:33 -0700785 for (i = 0; i < req->num_pages; i++) {
786 struct page *page = req->pages[i];
787 if (!req->out.h.error)
788 SetPageUptodate(page);
Miklos Szeredic1aa96a2006-01-16 22:14:46 -0800789 else
790 SetPageError(page);
Miklos Szeredidb50b962005-09-09 13:10:33 -0700791 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300792 put_page(page);
Miklos Szeredidb50b962005-09-09 13:10:33 -0700793 }
Miklos Szeredic756e0a2007-10-16 23:31:00 -0700794 if (req->ff)
Miklos Szeredi5a18ec12011-02-25 14:44:58 +0100795 fuse_file_put(req->ff, false);
Miklos Szeredic1aa96a2006-01-16 22:14:46 -0800796}
797
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200798static void fuse_send_readpages(struct fuse_req *req, struct file *file)
Miklos Szeredic1aa96a2006-01-16 22:14:46 -0800799{
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200800 struct fuse_file *ff = file->private_data;
801 struct fuse_conn *fc = ff->fc;
Miklos Szeredic1aa96a2006-01-16 22:14:46 -0800802 loff_t pos = page_offset(req->pages[0]);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300803 size_t count = req->num_pages << PAGE_SHIFT;
Miklos Szeredif4975c62009-04-02 14:25:34 +0200804
805 req->out.argpages = 1;
Miklos Szeredic1aa96a2006-01-16 22:14:46 -0800806 req->out.page_zeroing = 1;
Miklos Szeredice534fb2010-05-25 15:06:07 +0200807 req->out.page_replace = 1;
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200808 fuse_read_fill(req, file, pos, count, FUSE_READ);
Miklos Szeredi5c5c5e52008-04-30 00:54:43 -0700809 req->misc.read.attr_ver = fuse_get_attr_version(fc);
Miklos Szeredi9cd68452006-02-01 03:04:40 -0800810 if (fc->async_read) {
Miklos Szeredic756e0a2007-10-16 23:31:00 -0700811 req->ff = fuse_file_get(ff);
Miklos Szeredi9cd68452006-02-01 03:04:40 -0800812 req->end = fuse_readpages_end;
Tejun Heob93f8582008-11-26 12:03:55 +0100813 fuse_request_send_background(fc, req);
Miklos Szeredi9cd68452006-02-01 03:04:40 -0800814 } else {
Tejun Heob93f8582008-11-26 12:03:55 +0100815 fuse_request_send(fc, req);
Miklos Szeredi9cd68452006-02-01 03:04:40 -0800816 fuse_readpages_end(fc, req);
Tejun Heoe9bb09d2008-11-26 12:03:54 +0100817 fuse_put_request(fc, req);
Miklos Szeredi9cd68452006-02-01 03:04:40 -0800818 }
Miklos Szeredidb50b962005-09-09 13:10:33 -0700819}
820
Miklos Szeredic756e0a2007-10-16 23:31:00 -0700821struct fuse_fill_data {
Miklos Szeredidb50b962005-09-09 13:10:33 -0700822 struct fuse_req *req;
Miklos Szeredia6643092007-11-28 16:22:00 -0800823 struct file *file;
Miklos Szeredidb50b962005-09-09 13:10:33 -0700824 struct inode *inode;
Maxim Patlasovf8dbdf82012-10-26 19:48:51 +0400825 unsigned nr_pages;
Miklos Szeredidb50b962005-09-09 13:10:33 -0700826};
827
828static int fuse_readpages_fill(void *_data, struct page *page)
829{
Miklos Szeredic756e0a2007-10-16 23:31:00 -0700830 struct fuse_fill_data *data = _data;
Miklos Szeredidb50b962005-09-09 13:10:33 -0700831 struct fuse_req *req = data->req;
832 struct inode *inode = data->inode;
833 struct fuse_conn *fc = get_fuse_conn(inode);
834
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700835 fuse_wait_on_page_writeback(inode, page->index);
836
Miklos Szeredidb50b962005-09-09 13:10:33 -0700837 if (req->num_pages &&
838 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300839 (req->num_pages + 1) * PAGE_SIZE > fc->max_read ||
Miklos Szeredidb50b962005-09-09 13:10:33 -0700840 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
Maxim Patlasovf8dbdf82012-10-26 19:48:51 +0400841 int nr_alloc = min_t(unsigned, data->nr_pages,
842 FUSE_MAX_PAGES_PER_REQ);
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200843 fuse_send_readpages(req, data->file);
Maxim Patlasov8b41e672013-03-21 18:02:04 +0400844 if (fc->async_read)
845 req = fuse_get_req_for_background(fc, nr_alloc);
846 else
847 req = fuse_get_req(fc, nr_alloc);
848
849 data->req = req;
Miklos Szeredice1d5a42006-04-10 22:54:58 -0700850 if (IS_ERR(req)) {
Miklos Szeredidb50b962005-09-09 13:10:33 -0700851 unlock_page(page);
Miklos Szeredice1d5a42006-04-10 22:54:58 -0700852 return PTR_ERR(req);
Miklos Szeredidb50b962005-09-09 13:10:33 -0700853 }
Miklos Szeredidb50b962005-09-09 13:10:33 -0700854 }
Maxim Patlasovf8dbdf82012-10-26 19:48:51 +0400855
856 if (WARN_ON(req->num_pages >= req->max_pages)) {
857 fuse_put_request(fc, req);
858 return -EIO;
859 }
860
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300861 get_page(page);
Miklos Szeredidb50b962005-09-09 13:10:33 -0700862 req->pages[req->num_pages] = page;
Maxim Patlasov85f40ae2012-10-26 19:49:33 +0400863 req->page_descs[req->num_pages].length = PAGE_SIZE;
Miklos Szeredi1729a162008-11-26 12:03:54 +0100864 req->num_pages++;
Maxim Patlasovf8dbdf82012-10-26 19:48:51 +0400865 data->nr_pages--;
Miklos Szeredidb50b962005-09-09 13:10:33 -0700866 return 0;
867}
868
869static int fuse_readpages(struct file *file, struct address_space *mapping,
870 struct list_head *pages, unsigned nr_pages)
871{
872 struct inode *inode = mapping->host;
873 struct fuse_conn *fc = get_fuse_conn(inode);
Miklos Szeredic756e0a2007-10-16 23:31:00 -0700874 struct fuse_fill_data data;
Miklos Szeredidb50b962005-09-09 13:10:33 -0700875 int err;
Maxim Patlasovf8dbdf82012-10-26 19:48:51 +0400876 int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ);
Miklos Szeredi248d86e2006-01-06 00:19:39 -0800877
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700878 err = -EIO;
Miklos Szeredi248d86e2006-01-06 00:19:39 -0800879 if (is_bad_inode(inode))
OGAWA Hirofumi2e990022006-11-02 22:07:09 -0800880 goto out;
Miklos Szeredi248d86e2006-01-06 00:19:39 -0800881
Miklos Szeredia6643092007-11-28 16:22:00 -0800882 data.file = file;
Miklos Szeredidb50b962005-09-09 13:10:33 -0700883 data.inode = inode;
Maxim Patlasov8b41e672013-03-21 18:02:04 +0400884 if (fc->async_read)
885 data.req = fuse_get_req_for_background(fc, nr_alloc);
886 else
887 data.req = fuse_get_req(fc, nr_alloc);
Maxim Patlasovf8dbdf82012-10-26 19:48:51 +0400888 data.nr_pages = nr_pages;
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700889 err = PTR_ERR(data.req);
Miklos Szeredice1d5a42006-04-10 22:54:58 -0700890 if (IS_ERR(data.req))
OGAWA Hirofumi2e990022006-11-02 22:07:09 -0800891 goto out;
Miklos Szeredidb50b962005-09-09 13:10:33 -0700892
893 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
Miklos Szeredid3406ff2006-04-10 22:54:49 -0700894 if (!err) {
895 if (data.req->num_pages)
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200896 fuse_send_readpages(data.req, file);
Miklos Szeredid3406ff2006-04-10 22:54:49 -0700897 else
898 fuse_put_request(fc, data.req);
899 }
OGAWA Hirofumi2e990022006-11-02 22:07:09 -0800900out:
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700901 return err;
Miklos Szeredidb50b962005-09-09 13:10:33 -0700902}
903
Al Viro37c20f12014-04-02 14:47:09 -0400904static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Miklos Szeredibcb4be82007-11-28 16:21:59 -0800905{
906 struct inode *inode = iocb->ki_filp->f_mapping->host;
Brian Fostera8894272012-07-16 15:23:50 -0400907 struct fuse_conn *fc = get_fuse_conn(inode);
Miklos Szeredibcb4be82007-11-28 16:21:59 -0800908
Brian Fostera8894272012-07-16 15:23:50 -0400909 /*
910 * In auto invalidate mode, always update attributes on read.
911 * Otherwise, only update if we attempt to read past EOF (to ensure
912 * i_size is up to date).
913 */
914 if (fc->auto_inval_data ||
Al Viro37c20f12014-04-02 14:47:09 -0400915 (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) {
Miklos Szeredibcb4be82007-11-28 16:21:59 -0800916 int err;
Miklos Szeredibcb4be82007-11-28 16:21:59 -0800917 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
918 if (err)
919 return err;
920 }
921
Al Viro37c20f12014-04-02 14:47:09 -0400922 return generic_file_read_iter(iocb, to);
Miklos Szeredibcb4be82007-11-28 16:21:59 -0800923}
924
Miklos Szeredi2d698b02009-04-28 16:56:36 +0200925static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200926 loff_t pos, size_t count)
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700927{
Miklos Szeredib25e82e2007-10-18 03:07:03 -0700928 struct fuse_write_in *inarg = &req->misc.write.in;
929 struct fuse_write_out *outarg = &req->misc.write.out;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700930
Miklos Szeredib25e82e2007-10-18 03:07:03 -0700931 inarg->fh = ff->fh;
932 inarg->offset = pos;
933 inarg->size = count;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700934 req->in.h.opcode = FUSE_WRITE;
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200935 req->in.h.nodeid = ff->nodeid;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700936 req->in.numargs = 2;
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200937 if (ff->fc->minor < 9)
Miklos Szeredif3332112007-10-18 03:07:04 -0700938 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
939 else
940 req->in.args[0].size = sizeof(struct fuse_write_in);
Miklos Szeredib25e82e2007-10-18 03:07:03 -0700941 req->in.args[0].value = inarg;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700942 req->in.args[1].size = count;
943 req->out.numargs = 1;
944 req->out.args[0].size = sizeof(struct fuse_write_out);
Miklos Szeredib25e82e2007-10-18 03:07:03 -0700945 req->out.args[0].value = outarg;
946}
947
Maxim Patlasov36cf66e2012-12-14 19:20:51 +0400948static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io,
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200949 loff_t pos, size_t count, fl_owner_t owner)
Miklos Szeredib25e82e2007-10-18 03:07:03 -0700950{
Maxim Patlasov36cf66e2012-12-14 19:20:51 +0400951 struct file *file = io->file;
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200952 struct fuse_file *ff = file->private_data;
953 struct fuse_conn *fc = ff->fc;
Miklos Szeredi2d698b02009-04-28 16:56:36 +0200954 struct fuse_write_in *inarg = &req->misc.write.in;
955
Miklos Szeredi2106cb12009-04-28 16:56:37 +0200956 fuse_write_fill(req, ff, pos, count);
Miklos Szeredi2d698b02009-04-28 16:56:36 +0200957 inarg->flags = file->f_flags;
Miklos Szeredif3332112007-10-18 03:07:04 -0700958 if (owner != NULL) {
Miklos Szeredif3332112007-10-18 03:07:04 -0700959 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
960 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
961 }
Maxim Patlasov36cf66e2012-12-14 19:20:51 +0400962
963 if (io->async)
964 return fuse_async_req_send(fc, req, count, io);
965
Tejun Heob93f8582008-11-26 12:03:55 +0100966 fuse_request_send(fc, req);
Miklos Szeredib25e82e2007-10-18 03:07:03 -0700967 return req->misc.write.out.size;
Miklos Szeredib6aeade2005-09-09 13:10:30 -0700968}
969
Maxim Patlasovb0aa7602013-12-26 19:51:11 +0400970bool fuse_write_update_size(struct inode *inode, loff_t pos)
Miklos Szeredi854512e2008-04-30 00:54:41 -0700971{
972 struct fuse_conn *fc = get_fuse_conn(inode);
973 struct fuse_inode *fi = get_fuse_inode(inode);
Maxim Patlasovb0aa7602013-12-26 19:51:11 +0400974 bool ret = false;
Miklos Szeredi854512e2008-04-30 00:54:41 -0700975
976 spin_lock(&fc->lock);
977 fi->attr_version = ++fc->attr_version;
Maxim Patlasovb0aa7602013-12-26 19:51:11 +0400978 if (pos > inode->i_size) {
Miklos Szeredi854512e2008-04-30 00:54:41 -0700979 i_size_write(inode, pos);
Maxim Patlasovb0aa7602013-12-26 19:51:11 +0400980 ret = true;
981 }
Miklos Szeredi854512e2008-04-30 00:54:41 -0700982 spin_unlock(&fc->lock);
Maxim Patlasovb0aa7602013-12-26 19:51:11 +0400983
984 return ret;
Miklos Szeredi854512e2008-04-30 00:54:41 -0700985}
986
Nick Pigginea9b9902008-04-30 00:54:42 -0700987static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
988 struct inode *inode, loff_t pos,
989 size_t count)
990{
991 size_t res;
992 unsigned offset;
993 unsigned i;
Seth Forshee744742d2016-03-11 10:35:34 -0600994 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(file);
Nick Pigginea9b9902008-04-30 00:54:42 -0700995
996 for (i = 0; i < req->num_pages; i++)
997 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
998
Maxim Patlasov36cf66e2012-12-14 19:20:51 +0400999 res = fuse_send_write(req, &io, pos, count, NULL);
Nick Pigginea9b9902008-04-30 00:54:42 -07001000
Maxim Patlasovb2430d72012-10-26 19:49:24 +04001001 offset = req->page_descs[0].offset;
Nick Pigginea9b9902008-04-30 00:54:42 -07001002 count = res;
1003 for (i = 0; i < req->num_pages; i++) {
1004 struct page *page = req->pages[i];
1005
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001006 if (!req->out.h.error && !offset && count >= PAGE_SIZE)
Nick Pigginea9b9902008-04-30 00:54:42 -07001007 SetPageUptodate(page);
1008
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001009 if (count > PAGE_SIZE - offset)
1010 count -= PAGE_SIZE - offset;
Nick Pigginea9b9902008-04-30 00:54:42 -07001011 else
1012 count = 0;
1013 offset = 0;
1014
1015 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001016 put_page(page);
Nick Pigginea9b9902008-04-30 00:54:42 -07001017 }
1018
1019 return res;
1020}
1021
1022static ssize_t fuse_fill_write_pages(struct fuse_req *req,
1023 struct address_space *mapping,
1024 struct iov_iter *ii, loff_t pos)
1025{
1026 struct fuse_conn *fc = get_fuse_conn(mapping->host);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001027 unsigned offset = pos & (PAGE_SIZE - 1);
Nick Pigginea9b9902008-04-30 00:54:42 -07001028 size_t count = 0;
1029 int err;
1030
Miklos Szeredif4975c62009-04-02 14:25:34 +02001031 req->in.argpages = 1;
Maxim Patlasovb2430d72012-10-26 19:49:24 +04001032 req->page_descs[0].offset = offset;
Nick Pigginea9b9902008-04-30 00:54:42 -07001033
1034 do {
1035 size_t tmp;
1036 struct page *page;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001037 pgoff_t index = pos >> PAGE_SHIFT;
1038 size_t bytes = min_t(size_t, PAGE_SIZE - offset,
Nick Pigginea9b9902008-04-30 00:54:42 -07001039 iov_iter_count(ii));
1040
1041 bytes = min_t(size_t, bytes, fc->max_write - count);
1042
1043 again:
1044 err = -EFAULT;
1045 if (iov_iter_fault_in_readable(ii, bytes))
1046 break;
1047
1048 err = -ENOMEM;
Nick Piggin54566b22009-01-04 12:00:53 -08001049 page = grab_cache_page_write_begin(mapping, index, 0);
Nick Pigginea9b9902008-04-30 00:54:42 -07001050 if (!page)
1051 break;
1052
anfei zhou931e80e2010-02-02 13:44:02 -08001053 if (mapping_writably_mapped(mapping))
1054 flush_dcache_page(page);
1055
Nick Pigginea9b9902008-04-30 00:54:42 -07001056 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
Nick Pigginea9b9902008-04-30 00:54:42 -07001057 flush_dcache_page(page);
1058
Roman Gushchin3ca81382015-10-12 16:33:44 +03001059 iov_iter_advance(ii, tmp);
Nick Pigginea9b9902008-04-30 00:54:42 -07001060 if (!tmp) {
1061 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001062 put_page(page);
Nick Pigginea9b9902008-04-30 00:54:42 -07001063 bytes = min(bytes, iov_iter_single_seg_count(ii));
1064 goto again;
1065 }
1066
1067 err = 0;
1068 req->pages[req->num_pages] = page;
Maxim Patlasov85f40ae2012-10-26 19:49:33 +04001069 req->page_descs[req->num_pages].length = tmp;
Nick Pigginea9b9902008-04-30 00:54:42 -07001070 req->num_pages++;
1071
Nick Pigginea9b9902008-04-30 00:54:42 -07001072 count += tmp;
1073 pos += tmp;
1074 offset += tmp;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001075 if (offset == PAGE_SIZE)
Nick Pigginea9b9902008-04-30 00:54:42 -07001076 offset = 0;
1077
Miklos Szeredi78bb6cb2008-05-12 14:02:32 -07001078 if (!fc->big_writes)
1079 break;
Nick Pigginea9b9902008-04-30 00:54:42 -07001080 } while (iov_iter_count(ii) && count < fc->max_write &&
Maxim Patlasovd07f09f2012-10-26 19:49:00 +04001081 req->num_pages < req->max_pages && offset == 0);
Nick Pigginea9b9902008-04-30 00:54:42 -07001082
1083 return count > 0 ? count : err;
1084}
1085
Maxim Patlasovd07f09f2012-10-26 19:49:00 +04001086static inline unsigned fuse_wr_pages(loff_t pos, size_t len)
1087{
1088 return min_t(unsigned,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001089 ((pos + len - 1) >> PAGE_SHIFT) -
1090 (pos >> PAGE_SHIFT) + 1,
Maxim Patlasovd07f09f2012-10-26 19:49:00 +04001091 FUSE_MAX_PAGES_PER_REQ);
1092}
1093
Nick Pigginea9b9902008-04-30 00:54:42 -07001094static ssize_t fuse_perform_write(struct file *file,
1095 struct address_space *mapping,
1096 struct iov_iter *ii, loff_t pos)
1097{
1098 struct inode *inode = mapping->host;
1099 struct fuse_conn *fc = get_fuse_conn(inode);
Maxim Patlasov06a7c3c2013-08-30 17:06:04 +04001100 struct fuse_inode *fi = get_fuse_inode(inode);
Nick Pigginea9b9902008-04-30 00:54:42 -07001101 int err = 0;
1102 ssize_t res = 0;
1103
1104 if (is_bad_inode(inode))
1105 return -EIO;
1106
Maxim Patlasov06a7c3c2013-08-30 17:06:04 +04001107 if (inode->i_size < pos + iov_iter_count(ii))
1108 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1109
Nick Pigginea9b9902008-04-30 00:54:42 -07001110 do {
1111 struct fuse_req *req;
1112 ssize_t count;
Maxim Patlasovd07f09f2012-10-26 19:49:00 +04001113 unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii));
Nick Pigginea9b9902008-04-30 00:54:42 -07001114
Maxim Patlasovd07f09f2012-10-26 19:49:00 +04001115 req = fuse_get_req(fc, nr_pages);
Nick Pigginea9b9902008-04-30 00:54:42 -07001116 if (IS_ERR(req)) {
1117 err = PTR_ERR(req);
1118 break;
1119 }
1120
1121 count = fuse_fill_write_pages(req, mapping, ii, pos);
1122 if (count <= 0) {
1123 err = count;
1124 } else {
1125 size_t num_written;
1126
1127 num_written = fuse_send_write_pages(req, file, inode,
1128 pos, count);
1129 err = req->out.h.error;
1130 if (!err) {
1131 res += num_written;
1132 pos += num_written;
1133
1134 /* break out of the loop on short write */
1135 if (num_written != count)
1136 err = -EIO;
1137 }
1138 }
1139 fuse_put_request(fc, req);
1140 } while (!err && iov_iter_count(ii));
1141
1142 if (res > 0)
1143 fuse_write_update_size(inode, pos);
1144
Maxim Patlasov06a7c3c2013-08-30 17:06:04 +04001145 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
Nick Pigginea9b9902008-04-30 00:54:42 -07001146 fuse_invalidate_attr(inode);
1147
1148 return res > 0 ? res : err;
1149}
1150
Al Viro84c3d552014-04-03 14:33:23 -04001151static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Nick Pigginea9b9902008-04-30 00:54:42 -07001152{
1153 struct file *file = iocb->ki_filp;
1154 struct address_space *mapping = file->f_mapping;
Nick Pigginea9b9902008-04-30 00:54:42 -07001155 ssize_t written = 0;
Anand Avati4273b792012-02-17 12:46:25 -05001156 ssize_t written_buffered = 0;
Nick Pigginea9b9902008-04-30 00:54:42 -07001157 struct inode *inode = mapping->host;
1158 ssize_t err;
Anand Avati4273b792012-02-17 12:46:25 -05001159 loff_t endbyte = 0;
Nick Pigginea9b9902008-04-30 00:54:42 -07001160
Pavel Emelyanov4d99ff82013-10-10 17:12:18 +04001161 if (get_fuse_conn(inode)->writeback_cache) {
1162 /* Update size (EOF optimization) and mode (SUID clearing) */
1163 err = fuse_update_attributes(mapping->host, NULL, file, NULL);
1164 if (err)
1165 return err;
1166
Al Viro84c3d552014-04-03 14:33:23 -04001167 return generic_file_write_iter(iocb, from);
Pavel Emelyanov4d99ff82013-10-10 17:12:18 +04001168 }
1169
Al Viro59551022016-01-22 15:40:57 -05001170 inode_lock(inode);
Nick Pigginea9b9902008-04-30 00:54:42 -07001171
1172 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01001173 current->backing_dev_info = inode_to_bdi(inode);
Nick Pigginea9b9902008-04-30 00:54:42 -07001174
Al Viro3309dd02015-04-09 12:55:47 -04001175 err = generic_write_checks(iocb, from);
1176 if (err <= 0)
Nick Pigginea9b9902008-04-30 00:54:42 -07001177 goto out;
1178
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001179 err = file_remove_privs(file);
Nick Pigginea9b9902008-04-30 00:54:42 -07001180 if (err)
1181 goto out;
1182
Josef Bacikc3b2da32012-03-26 09:59:21 -04001183 err = file_update_time(file);
1184 if (err)
1185 goto out;
Nick Pigginea9b9902008-04-30 00:54:42 -07001186
Al Viro2ba48ce2015-04-09 13:52:01 -04001187 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro3309dd02015-04-09 12:55:47 -04001188 loff_t pos = iocb->ki_pos;
Al Viro84c3d552014-04-03 14:33:23 -04001189 written = generic_file_direct_write(iocb, from, pos);
1190 if (written < 0 || !iov_iter_count(from))
Anand Avati4273b792012-02-17 12:46:25 -05001191 goto out;
Nick Pigginea9b9902008-04-30 00:54:42 -07001192
Anand Avati4273b792012-02-17 12:46:25 -05001193 pos += written;
Anand Avati4273b792012-02-17 12:46:25 -05001194
Al Viro84c3d552014-04-03 14:33:23 -04001195 written_buffered = fuse_perform_write(file, mapping, from, pos);
Anand Avati4273b792012-02-17 12:46:25 -05001196 if (written_buffered < 0) {
1197 err = written_buffered;
1198 goto out;
1199 }
1200 endbyte = pos + written_buffered - 1;
1201
1202 err = filemap_write_and_wait_range(file->f_mapping, pos,
1203 endbyte);
1204 if (err)
1205 goto out;
1206
1207 invalidate_mapping_pages(file->f_mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001208 pos >> PAGE_SHIFT,
1209 endbyte >> PAGE_SHIFT);
Anand Avati4273b792012-02-17 12:46:25 -05001210
1211 written += written_buffered;
1212 iocb->ki_pos = pos + written_buffered;
1213 } else {
Al Viro3309dd02015-04-09 12:55:47 -04001214 written = fuse_perform_write(file, mapping, from, iocb->ki_pos);
Anand Avati4273b792012-02-17 12:46:25 -05001215 if (written >= 0)
Al Viro3309dd02015-04-09 12:55:47 -04001216 iocb->ki_pos += written;
Anand Avati4273b792012-02-17 12:46:25 -05001217 }
Nick Pigginea9b9902008-04-30 00:54:42 -07001218out:
1219 current->backing_dev_info = NULL;
Al Viro59551022016-01-22 15:40:57 -05001220 inode_unlock(inode);
Nick Pigginea9b9902008-04-30 00:54:42 -07001221
1222 return written ? written : err;
1223}
1224
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001225static inline void fuse_page_descs_length_init(struct fuse_req *req,
1226 unsigned index, unsigned nr_pages)
Maxim Patlasov85f40ae2012-10-26 19:49:33 +04001227{
1228 int i;
1229
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001230 for (i = index; i < index + nr_pages; i++)
Maxim Patlasov85f40ae2012-10-26 19:49:33 +04001231 req->page_descs[i].length = PAGE_SIZE -
1232 req->page_descs[i].offset;
1233}
1234
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001235static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
1236{
1237 return (unsigned long)ii->iov->iov_base + ii->iov_offset;
1238}
1239
1240static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
1241 size_t max_size)
1242{
1243 return min(iov_iter_single_seg_count(ii), max_size);
1244}
1245
Maxim Patlasovb98d0232012-10-26 19:50:15 +04001246static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii,
Miklos Szeredice60a2f2009-04-09 17:37:52 +02001247 size_t *nbytesp, int write)
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001248{
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001249 size_t nbytes = 0; /* # bytes already packed in req */
Ashish Samant742f9922016-03-14 21:57:35 -07001250 ssize_t ret = 0;
Maxim Patlasovb98d0232012-10-26 19:50:15 +04001251
Miklos Szeredif4975c62009-04-02 14:25:34 +02001252 /* Special case for kernel I/O: can copy directly into the buffer */
Al Viro62a80672014-04-04 23:12:29 -04001253 if (ii->type & ITER_KVEC) {
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001254 unsigned long user_addr = fuse_get_user_addr(ii);
1255 size_t frag_size = fuse_get_frag_size(ii, *nbytesp);
1256
Miklos Szeredif4975c62009-04-02 14:25:34 +02001257 if (write)
1258 req->in.args[1].value = (void *) user_addr;
1259 else
1260 req->out.args[0].value = (void *) user_addr;
1261
Maxim Patlasovb98d0232012-10-26 19:50:15 +04001262 iov_iter_advance(ii, frag_size);
1263 *nbytesp = frag_size;
Miklos Szeredif4975c62009-04-02 14:25:34 +02001264 return 0;
1265 }
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001266
Maxim Patlasov5565a9d2012-10-26 19:50:36 +04001267 while (nbytes < *nbytesp && req->num_pages < req->max_pages) {
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001268 unsigned npages;
Al Virof67da302014-03-19 01:16:16 -04001269 size_t start;
Ashish Samant742f9922016-03-14 21:57:35 -07001270 ret = iov_iter_get_pages(ii, &req->pages[req->num_pages],
Miklos Szeredi2c809292014-09-24 17:09:11 +02001271 *nbytesp - nbytes,
Al Viroc7f38882014-06-18 20:34:33 -04001272 req->max_pages - req->num_pages,
1273 &start);
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001274 if (ret < 0)
Ashish Samant742f9922016-03-14 21:57:35 -07001275 break;
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001276
Al Viroc9c37e22014-03-16 16:08:30 -04001277 iov_iter_advance(ii, ret);
1278 nbytes += ret;
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001279
Al Viroc9c37e22014-03-16 16:08:30 -04001280 ret += start;
1281 npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE;
1282
1283 req->page_descs[req->num_pages].offset = start;
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001284 fuse_page_descs_length_init(req, req->num_pages, npages);
1285
1286 req->num_pages += npages;
1287 req->page_descs[req->num_pages - 1].length -=
Al Viroc9c37e22014-03-16 16:08:30 -04001288 (PAGE_SIZE - ret) & (PAGE_SIZE - 1);
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001289 }
Miklos Szeredif4975c62009-04-02 14:25:34 +02001290
1291 if (write)
1292 req->in.argpages = 1;
1293 else
1294 req->out.argpages = 1;
1295
Maxim Patlasov7c190c82012-10-26 19:50:29 +04001296 *nbytesp = nbytes;
Miklos Szeredif4975c62009-04-02 14:25:34 +02001297
Ashish Samant2c932d42016-03-25 10:53:41 -07001298 return ret < 0 ? ret : 0;
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001299}
1300
Maxim Patlasov5565a9d2012-10-26 19:50:36 +04001301static inline int fuse_iter_npages(const struct iov_iter *ii_p)
1302{
Al Virof67da302014-03-19 01:16:16 -04001303 return iov_iter_npages(ii_p, FUSE_MAX_PAGES_PER_REQ);
Maxim Patlasov5565a9d2012-10-26 19:50:36 +04001304}
1305
Al Virod22a9432014-03-16 15:50:47 -04001306ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1307 loff_t *ppos, int flags)
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001308{
Pavel Emelyanovea8cd332013-10-10 17:12:05 +04001309 int write = flags & FUSE_DIO_WRITE;
1310 int cuse = flags & FUSE_DIO_CUSE;
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04001311 struct file *file = io->file;
Pavel Emelyanovea8cd332013-10-10 17:12:05 +04001312 struct inode *inode = file->f_mapping->host;
Miklos Szeredi2106cb12009-04-28 16:56:37 +02001313 struct fuse_file *ff = file->private_data;
1314 struct fuse_conn *fc = ff->fc;
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001315 size_t nmax = write ? fc->max_write : fc->max_read;
1316 loff_t pos = *ppos;
Al Virod22a9432014-03-16 15:50:47 -04001317 size_t count = iov_iter_count(iter);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001318 pgoff_t idx_from = pos >> PAGE_SHIFT;
1319 pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT;
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001320 ssize_t res = 0;
Miklos Szeredi248d86e2006-01-06 00:19:39 -08001321 struct fuse_req *req;
Ashish Samant742f9922016-03-14 21:57:35 -07001322 int err = 0;
Miklos Szeredi248d86e2006-01-06 00:19:39 -08001323
Brian Fosterde82b922013-05-14 11:25:39 -04001324 if (io->async)
Al Virod22a9432014-03-16 15:50:47 -04001325 req = fuse_get_req_for_background(fc, fuse_iter_npages(iter));
Brian Fosterde82b922013-05-14 11:25:39 -04001326 else
Al Virod22a9432014-03-16 15:50:47 -04001327 req = fuse_get_req(fc, fuse_iter_npages(iter));
Miklos Szeredice1d5a42006-04-10 22:54:58 -07001328 if (IS_ERR(req))
1329 return PTR_ERR(req);
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001330
Pavel Emelyanovea8cd332013-10-10 17:12:05 +04001331 if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) {
1332 if (!write)
Al Viro59551022016-01-22 15:40:57 -05001333 inode_lock(inode);
Pavel Emelyanovea8cd332013-10-10 17:12:05 +04001334 fuse_sync_writes(inode);
1335 if (!write)
Al Viro59551022016-01-22 15:40:57 -05001336 inode_unlock(inode);
Pavel Emelyanovea8cd332013-10-10 17:12:05 +04001337 }
1338
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001339 while (count) {
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001340 size_t nres;
Miklos Szeredi2106cb12009-04-28 16:56:37 +02001341 fl_owner_t owner = current->files;
Miklos Szeredif4975c62009-04-02 14:25:34 +02001342 size_t nbytes = min(count, nmax);
Ashish Samant742f9922016-03-14 21:57:35 -07001343 err = fuse_get_user_pages(req, iter, &nbytes, write);
1344 if (err && !nbytes)
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001345 break;
Miklos Szeredif4975c62009-04-02 14:25:34 +02001346
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001347 if (write)
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04001348 nres = fuse_send_write(req, io, pos, nbytes, owner);
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001349 else
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04001350 nres = fuse_send_read(req, io, pos, nbytes, owner);
Miklos Szeredi2106cb12009-04-28 16:56:37 +02001351
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04001352 if (!io->async)
1353 fuse_release_user_pages(req, !write);
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001354 if (req->out.h.error) {
Ashish Samant742f9922016-03-14 21:57:35 -07001355 err = req->out.h.error;
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001356 break;
1357 } else if (nres > nbytes) {
Ashish Samant742f9922016-03-14 21:57:35 -07001358 res = 0;
1359 err = -EIO;
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001360 break;
1361 }
1362 count -= nres;
1363 res += nres;
1364 pos += nres;
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001365 if (nres != nbytes)
1366 break;
Miklos Szeredi56cf34f2006-04-11 21:16:51 +02001367 if (count) {
1368 fuse_put_request(fc, req);
Brian Fosterde82b922013-05-14 11:25:39 -04001369 if (io->async)
1370 req = fuse_get_req_for_background(fc,
Al Virod22a9432014-03-16 15:50:47 -04001371 fuse_iter_npages(iter));
Brian Fosterde82b922013-05-14 11:25:39 -04001372 else
Al Virod22a9432014-03-16 15:50:47 -04001373 req = fuse_get_req(fc, fuse_iter_npages(iter));
Miklos Szeredi56cf34f2006-04-11 21:16:51 +02001374 if (IS_ERR(req))
1375 break;
1376 }
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001377 }
Anand V. Avatif60311d2009-10-22 06:24:52 -07001378 if (!IS_ERR(req))
1379 fuse_put_request(fc, req);
Miklos Szeredid09cb9d2009-04-28 16:56:36 +02001380 if (res > 0)
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001381 *ppos = pos;
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001382
Ashish Samant742f9922016-03-14 21:57:35 -07001383 return res > 0 ? res : err;
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001384}
Tejun Heo08cbf542009-04-14 10:54:53 +09001385EXPORT_SYMBOL_GPL(fuse_direct_io);
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001386
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04001387static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
Al Virod22a9432014-03-16 15:50:47 -04001388 struct iov_iter *iter,
1389 loff_t *ppos)
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001390{
Miklos Szeredid09cb9d2009-04-28 16:56:36 +02001391 ssize_t res;
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04001392 struct file *file = io->file;
Al Viro6131ffa2013-02-27 16:59:05 -05001393 struct inode *inode = file_inode(file);
Miklos Szeredid09cb9d2009-04-28 16:56:36 +02001394
1395 if (is_bad_inode(inode))
1396 return -EIO;
1397
Al Virod22a9432014-03-16 15:50:47 -04001398 res = fuse_direct_io(io, iter, ppos, 0);
Miklos Szeredid09cb9d2009-04-28 16:56:36 +02001399
1400 fuse_invalidate_attr(inode);
1401
1402 return res;
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001403}
1404
Al Viro15316262015-03-30 22:08:36 -04001405static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to)
Maxim Patlasovb98d0232012-10-26 19:50:15 +04001406{
Seth Forshee744742d2016-03-11 10:35:34 -06001407 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb->ki_filp);
Al Viro15316262015-03-30 22:08:36 -04001408 return __fuse_direct_read(&io, to, &iocb->ki_pos);
Maxim Patlasovb98d0232012-10-26 19:50:15 +04001409}
1410
Al Viro15316262015-03-30 22:08:36 -04001411static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001412{
Al Viro15316262015-03-30 22:08:36 -04001413 struct file *file = iocb->ki_filp;
Al Viro6131ffa2013-02-27 16:59:05 -05001414 struct inode *inode = file_inode(file);
Seth Forshee744742d2016-03-11 10:35:34 -06001415 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(file);
Al Viro15316262015-03-30 22:08:36 -04001416 ssize_t res;
Miklos Szeredid09cb9d2009-04-28 16:56:36 +02001417
1418 if (is_bad_inode(inode))
1419 return -EIO;
1420
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001421 /* Don't allow parallel writes to the same file */
Al Viro59551022016-01-22 15:40:57 -05001422 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04001423 res = generic_write_checks(iocb, from);
1424 if (res > 0)
Al Viro812408f2015-03-30 22:15:58 -04001425 res = fuse_direct_io(&io, from, &iocb->ki_pos, FUSE_DIO_WRITE);
Al Viro812408f2015-03-30 22:15:58 -04001426 fuse_invalidate_attr(inode);
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04001427 if (res > 0)
Al Viro15316262015-03-30 22:08:36 -04001428 fuse_write_update_size(inode, iocb->ki_pos);
Al Viro59551022016-01-22 15:40:57 -05001429 inode_unlock(inode);
Miklos Szeredid09cb9d2009-04-28 16:56:36 +02001430
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07001431 return res;
1432}
1433
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001434static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredib6aeade2005-09-09 13:10:30 -07001435{
Pavel Emelyanov385b1262013-06-29 21:42:48 +04001436 int i;
1437
1438 for (i = 0; i < req->num_pages; i++)
1439 __free_page(req->pages[i]);
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001440
1441 if (req->ff)
1442 fuse_file_put(req->ff, false);
Miklos Szeredib6aeade2005-09-09 13:10:30 -07001443}
1444
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001445static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredib6aeade2005-09-09 13:10:30 -07001446{
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001447 struct inode *inode = req->inode;
1448 struct fuse_inode *fi = get_fuse_inode(inode);
Christoph Hellwigde1414a2015-01-14 10:42:36 +01001449 struct backing_dev_info *bdi = inode_to_bdi(inode);
Pavel Emelyanov385b1262013-06-29 21:42:48 +04001450 int i;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001451
1452 list_del(&req->writepages_entry);
Pavel Emelyanov385b1262013-06-29 21:42:48 +04001453 for (i = 0; i < req->num_pages; i++) {
Tejun Heo93f78d82015-05-22 17:13:27 -04001454 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
Pavel Emelyanov385b1262013-06-29 21:42:48 +04001455 dec_zone_page_state(req->pages[i], NR_WRITEBACK_TEMP);
Tejun Heo93f78d82015-05-22 17:13:27 -04001456 wb_writeout_inc(&bdi->wb);
Pavel Emelyanov385b1262013-06-29 21:42:48 +04001457 }
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001458 wake_up(&fi->page_waitq);
1459}
1460
1461/* Called under fc->lock, may release and reacquire it */
Maxim Patlasov6eaf4782013-10-02 21:38:32 +04001462static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req,
1463 loff_t size)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +02001464__releases(fc->lock)
1465__acquires(fc->lock)
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001466{
1467 struct fuse_inode *fi = get_fuse_inode(req->inode);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001468 struct fuse_write_in *inarg = &req->misc.write.in;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001469 __u64 data_size = req->num_pages * PAGE_SIZE;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001470
1471 if (!fc->connected)
1472 goto out_free;
1473
Pavel Emelyanov385b1262013-06-29 21:42:48 +04001474 if (inarg->offset + data_size <= size) {
1475 inarg->size = data_size;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001476 } else if (inarg->offset < size) {
Pavel Emelyanov385b1262013-06-29 21:42:48 +04001477 inarg->size = size - inarg->offset;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001478 } else {
1479 /* Got truncated off completely */
1480 goto out_free;
1481 }
1482
1483 req->in.args[1].size = inarg->size;
1484 fi->writectr++;
Tejun Heob93f8582008-11-26 12:03:55 +01001485 fuse_request_send_background_locked(fc, req);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001486 return;
1487
1488 out_free:
1489 fuse_writepage_finish(fc, req);
1490 spin_unlock(&fc->lock);
1491 fuse_writepage_free(fc, req);
Tejun Heoe9bb09d2008-11-26 12:03:54 +01001492 fuse_put_request(fc, req);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001493 spin_lock(&fc->lock);
1494}
1495
1496/*
1497 * If fi->writectr is positive (no truncate or fsync going on) send
1498 * all queued writepage requests.
1499 *
1500 * Called with fc->lock
1501 */
1502void fuse_flush_writepages(struct inode *inode)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +02001503__releases(fc->lock)
1504__acquires(fc->lock)
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001505{
1506 struct fuse_conn *fc = get_fuse_conn(inode);
1507 struct fuse_inode *fi = get_fuse_inode(inode);
Maxim Patlasov6eaf4782013-10-02 21:38:32 +04001508 size_t crop = i_size_read(inode);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001509 struct fuse_req *req;
1510
1511 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1512 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1513 list_del_init(&req->list);
Maxim Patlasov6eaf4782013-10-02 21:38:32 +04001514 fuse_send_writepage(fc, req, crop);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001515 }
1516}
1517
1518static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1519{
1520 struct inode *inode = req->inode;
1521 struct fuse_inode *fi = get_fuse_inode(inode);
1522
1523 mapping_set_error(inode->i_mapping, req->out.h.error);
1524 spin_lock(&fc->lock);
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001525 while (req->misc.write.next) {
Maxim Patlasov6eaf4782013-10-02 21:38:32 +04001526 struct fuse_conn *fc = get_fuse_conn(inode);
1527 struct fuse_write_in *inarg = &req->misc.write.in;
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001528 struct fuse_req *next = req->misc.write.next;
1529 req->misc.write.next = next->misc.write.next;
1530 next->misc.write.next = NULL;
Maxim Patlasovce128de2013-10-02 21:38:54 +04001531 next->ff = fuse_file_get(req->ff);
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001532 list_add(&next->writepages_entry, &fi->writepages);
Maxim Patlasov6eaf4782013-10-02 21:38:32 +04001533
1534 /*
1535 * Skip fuse_flush_writepages() to make it easy to crop requests
1536 * based on primary request size.
1537 *
1538 * 1st case (trivial): there are no concurrent activities using
1539 * fuse_set/release_nowrite. Then we're on safe side because
1540 * fuse_flush_writepages() would call fuse_send_writepage()
1541 * anyway.
1542 *
1543 * 2nd case: someone called fuse_set_nowrite and it is waiting
1544 * now for completion of all in-flight requests. This happens
1545 * rarely and no more than once per page, so this should be
1546 * okay.
1547 *
1548 * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle
1549 * of fuse_set_nowrite..fuse_release_nowrite section. The fact
1550 * that fuse_set_nowrite returned implies that all in-flight
1551 * requests were completed along with all of their secondary
1552 * requests. Further primary requests are blocked by negative
1553 * writectr. Hence there cannot be any in-flight requests and
1554 * no invocations of fuse_writepage_end() while we're in
1555 * fuse_set_nowrite..fuse_release_nowrite section.
1556 */
1557 fuse_send_writepage(fc, next, inarg->offset + inarg->size);
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001558 }
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001559 fi->writectr--;
1560 fuse_writepage_finish(fc, req);
1561 spin_unlock(&fc->lock);
1562 fuse_writepage_free(fc, req);
1563}
1564
Miklos Szeredi1e18bda2014-04-28 14:19:23 +02001565static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc,
1566 struct fuse_inode *fi)
Pavel Emelyanovadcadfa2013-06-29 21:42:20 +04001567{
Miklos Szeredi72523422013-10-01 16:44:52 +02001568 struct fuse_file *ff = NULL;
Pavel Emelyanovadcadfa2013-06-29 21:42:20 +04001569
1570 spin_lock(&fc->lock);
Miklos Szeredi1e18bda2014-04-28 14:19:23 +02001571 if (!list_empty(&fi->write_files)) {
Miklos Szeredi72523422013-10-01 16:44:52 +02001572 ff = list_entry(fi->write_files.next, struct fuse_file,
1573 write_entry);
1574 fuse_file_get(ff);
1575 }
Pavel Emelyanovadcadfa2013-06-29 21:42:20 +04001576 spin_unlock(&fc->lock);
1577
1578 return ff;
1579}
1580
Miklos Szeredi1e18bda2014-04-28 14:19:23 +02001581static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc,
1582 struct fuse_inode *fi)
1583{
1584 struct fuse_file *ff = __fuse_write_file_get(fc, fi);
1585 WARN_ON(!ff);
1586 return ff;
1587}
1588
1589int fuse_write_inode(struct inode *inode, struct writeback_control *wbc)
1590{
1591 struct fuse_conn *fc = get_fuse_conn(inode);
1592 struct fuse_inode *fi = get_fuse_inode(inode);
1593 struct fuse_file *ff;
1594 int err;
1595
1596 ff = __fuse_write_file_get(fc, fi);
Maxim Patlasovab9e13f2014-04-28 14:19:24 +02001597 err = fuse_flush_times(inode, ff);
Miklos Szeredi1e18bda2014-04-28 14:19:23 +02001598 if (ff)
1599 fuse_file_put(ff, 0);
1600
1601 return err;
1602}
1603
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001604static int fuse_writepage_locked(struct page *page)
1605{
1606 struct address_space *mapping = page->mapping;
1607 struct inode *inode = mapping->host;
1608 struct fuse_conn *fc = get_fuse_conn(inode);
1609 struct fuse_inode *fi = get_fuse_inode(inode);
1610 struct fuse_req *req;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001611 struct page *tmp_page;
Miklos Szeredi72523422013-10-01 16:44:52 +02001612 int error = -ENOMEM;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001613
1614 set_page_writeback(page);
1615
Maxim Patlasov4250c062012-10-26 19:48:07 +04001616 req = fuse_request_alloc_nofs(1);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001617 if (!req)
1618 goto err;
1619
Miklos Szeredi825d6d32015-07-01 16:25:58 +02001620 /* writeback always goes to bg_queue */
1621 __set_bit(FR_BACKGROUND, &req->flags);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001622 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1623 if (!tmp_page)
1624 goto err_free;
1625
Miklos Szeredi72523422013-10-01 16:44:52 +02001626 error = -EIO;
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001627 req->ff = fuse_write_file_get(fc, fi);
Miklos Szeredi72523422013-10-01 16:44:52 +02001628 if (!req->ff)
Maxim Patlasov27f1b362014-07-10 15:32:43 +04001629 goto err_nofile;
Miklos Szeredi72523422013-10-01 16:44:52 +02001630
Pavel Emelyanovadcadfa2013-06-29 21:42:20 +04001631 fuse_write_fill(req, req->ff, page_offset(page), 0);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001632
1633 copy_highpage(tmp_page, page);
Miklos Szeredi2d698b02009-04-28 16:56:36 +02001634 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001635 req->misc.write.next = NULL;
Miklos Szeredif4975c62009-04-02 14:25:34 +02001636 req->in.argpages = 1;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001637 req->num_pages = 1;
1638 req->pages[0] = tmp_page;
Maxim Patlasovb2430d72012-10-26 19:49:24 +04001639 req->page_descs[0].offset = 0;
Maxim Patlasov85f40ae2012-10-26 19:49:33 +04001640 req->page_descs[0].length = PAGE_SIZE;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001641 req->end = fuse_writepage_end;
1642 req->inode = inode;
1643
Tejun Heo93f78d82015-05-22 17:13:27 -04001644 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001645 inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001646
1647 spin_lock(&fc->lock);
1648 list_add(&req->writepages_entry, &fi->writepages);
1649 list_add_tail(&req->list, &fi->queued_writes);
1650 fuse_flush_writepages(inode);
1651 spin_unlock(&fc->lock);
1652
Maxim Patlasov4a4ac4e2013-08-12 20:39:30 +04001653 end_page_writeback(page);
1654
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001655 return 0;
1656
Maxim Patlasov27f1b362014-07-10 15:32:43 +04001657err_nofile:
1658 __free_page(tmp_page);
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001659err_free:
1660 fuse_request_free(req);
1661err:
1662 end_page_writeback(page);
Miklos Szeredi72523422013-10-01 16:44:52 +02001663 return error;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001664}
1665
1666static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1667{
1668 int err;
1669
Miklos Szerediff17be02013-10-01 16:44:53 +02001670 if (fuse_page_is_writeback(page->mapping->host, page->index)) {
1671 /*
1672 * ->writepages() should be called for sync() and friends. We
1673 * should only get here on direct reclaim and then we are
1674 * allowed to skip a page which is already in flight
1675 */
1676 WARN_ON(wbc->sync_mode == WB_SYNC_ALL);
1677
1678 redirty_page_for_writepage(wbc, page);
1679 return 0;
1680 }
1681
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001682 err = fuse_writepage_locked(page);
1683 unlock_page(page);
1684
1685 return err;
1686}
1687
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001688struct fuse_fill_wb_data {
1689 struct fuse_req *req;
1690 struct fuse_file *ff;
1691 struct inode *inode;
Maxim Patlasov2d033ea2013-08-16 15:51:41 +04001692 struct page **orig_pages;
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001693};
1694
1695static void fuse_writepages_send(struct fuse_fill_wb_data *data)
1696{
1697 struct fuse_req *req = data->req;
1698 struct inode *inode = data->inode;
1699 struct fuse_conn *fc = get_fuse_conn(inode);
1700 struct fuse_inode *fi = get_fuse_inode(inode);
Maxim Patlasov2d033ea2013-08-16 15:51:41 +04001701 int num_pages = req->num_pages;
1702 int i;
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001703
1704 req->ff = fuse_file_get(data->ff);
1705 spin_lock(&fc->lock);
1706 list_add_tail(&req->list, &fi->queued_writes);
1707 fuse_flush_writepages(inode);
1708 spin_unlock(&fc->lock);
Maxim Patlasov2d033ea2013-08-16 15:51:41 +04001709
1710 for (i = 0; i < num_pages; i++)
1711 end_page_writeback(data->orig_pages[i]);
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001712}
1713
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001714static bool fuse_writepage_in_flight(struct fuse_req *new_req,
1715 struct page *page)
1716{
1717 struct fuse_conn *fc = get_fuse_conn(new_req->inode);
1718 struct fuse_inode *fi = get_fuse_inode(new_req->inode);
1719 struct fuse_req *tmp;
1720 struct fuse_req *old_req;
1721 bool found = false;
1722 pgoff_t curr_index;
1723
1724 BUG_ON(new_req->num_pages != 0);
1725
1726 spin_lock(&fc->lock);
1727 list_del(&new_req->writepages_entry);
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001728 list_for_each_entry(old_req, &fi->writepages, writepages_entry) {
1729 BUG_ON(old_req->inode != new_req->inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001730 curr_index = old_req->misc.write.in.offset >> PAGE_SHIFT;
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001731 if (curr_index <= page->index &&
1732 page->index < curr_index + old_req->num_pages) {
1733 found = true;
1734 break;
1735 }
1736 }
Maxim Patlasovf6011082013-10-02 15:01:07 +04001737 if (!found) {
1738 list_add(&new_req->writepages_entry, &fi->writepages);
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001739 goto out_unlock;
Maxim Patlasovf6011082013-10-02 15:01:07 +04001740 }
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001741
Maxim Patlasovf6011082013-10-02 15:01:07 +04001742 new_req->num_pages = 1;
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001743 for (tmp = old_req; tmp != NULL; tmp = tmp->misc.write.next) {
1744 BUG_ON(tmp->inode != new_req->inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001745 curr_index = tmp->misc.write.in.offset >> PAGE_SHIFT;
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001746 if (tmp->num_pages == 1 &&
1747 curr_index == page->index) {
1748 old_req = tmp;
1749 }
1750 }
1751
Miklos Szeredi33e14b42015-07-01 16:26:01 +02001752 if (old_req->num_pages == 1 && test_bit(FR_PENDING, &old_req->flags)) {
Christoph Hellwigde1414a2015-01-14 10:42:36 +01001753 struct backing_dev_info *bdi = inode_to_bdi(page->mapping->host);
Maxim Patlasov41b6e412013-10-02 21:38:43 +04001754
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001755 copy_highpage(old_req->pages[0], page);
1756 spin_unlock(&fc->lock);
1757
Tejun Heo93f78d82015-05-22 17:13:27 -04001758 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001759 dec_zone_page_state(page, NR_WRITEBACK_TEMP);
Tejun Heo93f78d82015-05-22 17:13:27 -04001760 wb_writeout_inc(&bdi->wb);
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001761 fuse_writepage_free(fc, new_req);
1762 fuse_request_free(new_req);
1763 goto out;
1764 } else {
1765 new_req->misc.write.next = old_req->misc.write.next;
1766 old_req->misc.write.next = new_req;
1767 }
1768out_unlock:
1769 spin_unlock(&fc->lock);
1770out:
1771 return found;
1772}
1773
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001774static int fuse_writepages_fill(struct page *page,
1775 struct writeback_control *wbc, void *_data)
1776{
1777 struct fuse_fill_wb_data *data = _data;
1778 struct fuse_req *req = data->req;
1779 struct inode *inode = data->inode;
1780 struct fuse_conn *fc = get_fuse_conn(inode);
1781 struct page *tmp_page;
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001782 bool is_writeback;
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001783 int err;
1784
1785 if (!data->ff) {
1786 err = -EIO;
1787 data->ff = fuse_write_file_get(fc, get_fuse_inode(inode));
1788 if (!data->ff)
1789 goto out_unlock;
1790 }
1791
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001792 /*
1793 * Being under writeback is unlikely but possible. For example direct
1794 * read to an mmaped fuse file will set the page dirty twice; once when
1795 * the pages are faulted with get_user_pages(), and then after the read
1796 * completed.
1797 */
1798 is_writeback = fuse_page_is_writeback(inode, page->index);
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001799
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001800 if (req && req->num_pages &&
1801 (is_writeback || req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001802 (req->num_pages + 1) * PAGE_SIZE > fc->max_write ||
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001803 data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) {
1804 fuse_writepages_send(data);
1805 data->req = NULL;
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001806 }
1807 err = -ENOMEM;
1808 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1809 if (!tmp_page)
1810 goto out_unlock;
1811
1812 /*
1813 * The page must not be redirtied until the writeout is completed
1814 * (i.e. userspace has sent a reply to the write request). Otherwise
1815 * there could be more than one temporary page instance for each real
1816 * page.
1817 *
1818 * This is ensured by holding the page lock in page_mkwrite() while
1819 * checking fuse_page_is_writeback(). We already hold the page lock
1820 * since clear_page_dirty_for_io() and keep it held until we add the
1821 * request to the fi->writepages list and increment req->num_pages.
1822 * After this fuse_page_is_writeback() will indicate that the page is
1823 * under writeback, so we can release the page lock.
1824 */
1825 if (data->req == NULL) {
1826 struct fuse_inode *fi = get_fuse_inode(inode);
1827
1828 err = -ENOMEM;
1829 req = fuse_request_alloc_nofs(FUSE_MAX_PAGES_PER_REQ);
1830 if (!req) {
1831 __free_page(tmp_page);
1832 goto out_unlock;
1833 }
1834
1835 fuse_write_fill(req, data->ff, page_offset(page), 0);
1836 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001837 req->misc.write.next = NULL;
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001838 req->in.argpages = 1;
Miklos Szeredi825d6d32015-07-01 16:25:58 +02001839 __set_bit(FR_BACKGROUND, &req->flags);
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001840 req->num_pages = 0;
1841 req->end = fuse_writepage_end;
1842 req->inode = inode;
1843
1844 spin_lock(&fc->lock);
1845 list_add(&req->writepages_entry, &fi->writepages);
1846 spin_unlock(&fc->lock);
1847
1848 data->req = req;
1849 }
1850 set_page_writeback(page);
1851
1852 copy_highpage(tmp_page, page);
1853 req->pages[req->num_pages] = tmp_page;
1854 req->page_descs[req->num_pages].offset = 0;
1855 req->page_descs[req->num_pages].length = PAGE_SIZE;
1856
Tejun Heo93f78d82015-05-22 17:13:27 -04001857 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001858 inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
Miklos Szeredi8b284dc2013-10-01 16:44:53 +02001859
1860 err = 0;
1861 if (is_writeback && fuse_writepage_in_flight(req, page)) {
1862 end_page_writeback(page);
1863 data->req = NULL;
1864 goto out_unlock;
1865 }
Maxim Patlasov2d033ea2013-08-16 15:51:41 +04001866 data->orig_pages[req->num_pages] = page;
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001867
1868 /*
1869 * Protected by fc->lock against concurrent access by
1870 * fuse_page_is_writeback().
1871 */
1872 spin_lock(&fc->lock);
1873 req->num_pages++;
1874 spin_unlock(&fc->lock);
1875
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001876out_unlock:
1877 unlock_page(page);
1878
1879 return err;
1880}
1881
1882static int fuse_writepages(struct address_space *mapping,
1883 struct writeback_control *wbc)
1884{
1885 struct inode *inode = mapping->host;
1886 struct fuse_fill_wb_data data;
1887 int err;
1888
1889 err = -EIO;
1890 if (is_bad_inode(inode))
1891 goto out;
1892
1893 data.inode = inode;
1894 data.req = NULL;
1895 data.ff = NULL;
1896
Maxim Patlasov2d033ea2013-08-16 15:51:41 +04001897 err = -ENOMEM;
Fabian Frederickf2b34552014-06-23 18:35:15 +02001898 data.orig_pages = kcalloc(FUSE_MAX_PAGES_PER_REQ,
1899 sizeof(struct page *),
Maxim Patlasov2d033ea2013-08-16 15:51:41 +04001900 GFP_NOFS);
1901 if (!data.orig_pages)
1902 goto out;
1903
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001904 err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data);
1905 if (data.req) {
1906 /* Ignore errors if we can write at least one page */
1907 BUG_ON(!data.req->num_pages);
1908 fuse_writepages_send(&data);
1909 err = 0;
1910 }
1911 if (data.ff)
1912 fuse_file_put(data.ff, false);
Maxim Patlasov2d033ea2013-08-16 15:51:41 +04001913
1914 kfree(data.orig_pages);
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04001915out:
1916 return err;
1917}
1918
Pavel Emelyanov6b12c1b2013-10-10 17:11:43 +04001919/*
1920 * It's worthy to make sure that space is reserved on disk for the write,
1921 * but how to implement it without killing performance need more thinking.
1922 */
1923static int fuse_write_begin(struct file *file, struct address_space *mapping,
1924 loff_t pos, unsigned len, unsigned flags,
1925 struct page **pagep, void **fsdata)
1926{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001927 pgoff_t index = pos >> PAGE_SHIFT;
Al Viroa4555892014-10-21 20:11:25 -04001928 struct fuse_conn *fc = get_fuse_conn(file_inode(file));
Pavel Emelyanov6b12c1b2013-10-10 17:11:43 +04001929 struct page *page;
1930 loff_t fsize;
1931 int err = -ENOMEM;
1932
1933 WARN_ON(!fc->writeback_cache);
1934
1935 page = grab_cache_page_write_begin(mapping, index, flags);
1936 if (!page)
1937 goto error;
1938
1939 fuse_wait_on_page_writeback(mapping->host, page->index);
1940
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001941 if (PageUptodate(page) || len == PAGE_SIZE)
Pavel Emelyanov6b12c1b2013-10-10 17:11:43 +04001942 goto success;
1943 /*
1944 * Check if the start this page comes after the end of file, in which
1945 * case the readpage can be optimized away.
1946 */
1947 fsize = i_size_read(mapping->host);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001948 if (fsize <= (pos & PAGE_MASK)) {
1949 size_t off = pos & ~PAGE_MASK;
Pavel Emelyanov6b12c1b2013-10-10 17:11:43 +04001950 if (off)
1951 zero_user_segment(page, 0, off);
1952 goto success;
1953 }
1954 err = fuse_do_readpage(file, page);
1955 if (err)
1956 goto cleanup;
1957success:
1958 *pagep = page;
1959 return 0;
1960
1961cleanup:
1962 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001963 put_page(page);
Pavel Emelyanov6b12c1b2013-10-10 17:11:43 +04001964error:
1965 return err;
1966}
1967
1968static int fuse_write_end(struct file *file, struct address_space *mapping,
1969 loff_t pos, unsigned len, unsigned copied,
1970 struct page *page, void *fsdata)
1971{
1972 struct inode *inode = page->mapping->host;
1973
1974 if (!PageUptodate(page)) {
1975 /* Zero any unwritten bytes at the end of the page */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001976 size_t endoff = (pos + copied) & ~PAGE_MASK;
Pavel Emelyanov6b12c1b2013-10-10 17:11:43 +04001977 if (endoff)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001978 zero_user_segment(page, endoff, PAGE_SIZE);
Pavel Emelyanov6b12c1b2013-10-10 17:11:43 +04001979 SetPageUptodate(page);
1980 }
1981
1982 fuse_write_update_size(inode, pos + copied);
1983 set_page_dirty(page);
1984 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001985 put_page(page);
Pavel Emelyanov6b12c1b2013-10-10 17:11:43 +04001986
1987 return copied;
1988}
1989
Miklos Szeredi3be5a522008-04-30 00:54:41 -07001990static int fuse_launder_page(struct page *page)
1991{
1992 int err = 0;
1993 if (clear_page_dirty_for_io(page)) {
1994 struct inode *inode = page->mapping->host;
1995 err = fuse_writepage_locked(page);
1996 if (!err)
1997 fuse_wait_on_page_writeback(inode, page->index);
1998 }
1999 return err;
2000}
2001
2002/*
2003 * Write back dirty pages now, because there may not be any suitable
2004 * open files later
2005 */
2006static void fuse_vma_close(struct vm_area_struct *vma)
2007{
2008 filemap_write_and_wait(vma->vm_file->f_mapping);
2009}
2010
2011/*
2012 * Wait for writeback against this page to complete before allowing it
2013 * to be marked dirty again, and hence written back again, possibly
2014 * before the previous writepage completed.
2015 *
2016 * Block here, instead of in ->writepage(), so that the userspace fs
2017 * can only block processes actually operating on the filesystem.
2018 *
2019 * Otherwise unprivileged userspace fs would be able to block
2020 * unrelated:
2021 *
2022 * - page migration
2023 * - sync(2)
2024 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
2025 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07002026static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Miklos Szeredi3be5a522008-04-30 00:54:41 -07002027{
Nick Pigginc2ec1752009-03-31 15:23:21 -07002028 struct page *page = vmf->page;
Miklos Szeredicca24372013-10-01 16:44:51 +02002029 struct inode *inode = file_inode(vma->vm_file);
2030
2031 file_update_time(vma->vm_file);
2032 lock_page(page);
2033 if (page->mapping != inode->i_mapping) {
2034 unlock_page(page);
2035 return VM_FAULT_NOPAGE;
2036 }
Miklos Szeredi3be5a522008-04-30 00:54:41 -07002037
2038 fuse_wait_on_page_writeback(inode, page->index);
Miklos Szeredicca24372013-10-01 16:44:51 +02002039 return VM_FAULT_LOCKED;
Miklos Szeredi3be5a522008-04-30 00:54:41 -07002040}
2041
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002042static const struct vm_operations_struct fuse_file_vm_ops = {
Miklos Szeredi3be5a522008-04-30 00:54:41 -07002043 .close = fuse_vma_close,
2044 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002045 .map_pages = filemap_map_pages,
Miklos Szeredi3be5a522008-04-30 00:54:41 -07002046 .page_mkwrite = fuse_page_mkwrite,
2047};
2048
2049static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
2050{
Pavel Emelyanov650b22b2013-10-10 17:10:04 +04002051 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2052 fuse_link_write_file(file);
2053
Miklos Szeredi3be5a522008-04-30 00:54:41 -07002054 file_accessed(file);
2055 vma->vm_ops = &fuse_file_vm_ops;
Miklos Szeredib6aeade2005-09-09 13:10:30 -07002056 return 0;
2057}
2058
Miklos Szeredifc280c92009-04-02 14:25:35 +02002059static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
2060{
2061 /* Can't provide the coherency needed for MAP_SHARED */
2062 if (vma->vm_flags & VM_MAYSHARE)
2063 return -ENODEV;
2064
Miklos Szeredi3121bfe2009-04-09 17:37:53 +02002065 invalidate_inode_pages2(file->f_mapping);
2066
Miklos Szeredifc280c92009-04-02 14:25:35 +02002067 return generic_file_mmap(file, vma);
2068}
2069
Miklos Szeredi71421252006-06-25 05:48:52 -07002070static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
2071 struct file_lock *fl)
2072{
2073 switch (ffl->type) {
2074 case F_UNLCK:
2075 break;
2076
2077 case F_RDLCK:
2078 case F_WRLCK:
2079 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
2080 ffl->end < ffl->start)
2081 return -EIO;
2082
2083 fl->fl_start = ffl->start;
2084 fl->fl_end = ffl->end;
2085 fl->fl_pid = ffl->pid;
2086 break;
2087
2088 default:
2089 return -EIO;
2090 }
2091 fl->fl_type = ffl->type;
2092 return 0;
2093}
2094
Miklos Szeredi70781872014-12-12 09:49:05 +01002095static void fuse_lk_fill(struct fuse_args *args, struct file *file,
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07002096 const struct file_lock *fl, int opcode, pid_t pid,
Miklos Szeredi70781872014-12-12 09:49:05 +01002097 int flock, struct fuse_lk_in *inarg)
Miklos Szeredi71421252006-06-25 05:48:52 -07002098{
Al Viro6131ffa2013-02-27 16:59:05 -05002099 struct inode *inode = file_inode(file);
Miklos Szeredi9c8ef562006-06-25 05:48:55 -07002100 struct fuse_conn *fc = get_fuse_conn(inode);
Miklos Szeredi71421252006-06-25 05:48:52 -07002101 struct fuse_file *ff = file->private_data;
Miklos Szeredi71421252006-06-25 05:48:52 -07002102
Miklos Szeredi70781872014-12-12 09:49:05 +01002103 memset(inarg, 0, sizeof(*inarg));
2104 inarg->fh = ff->fh;
2105 inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
2106 inarg->lk.start = fl->fl_start;
2107 inarg->lk.end = fl->fl_end;
2108 inarg->lk.type = fl->fl_type;
2109 inarg->lk.pid = pid;
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07002110 if (flock)
Miklos Szeredi70781872014-12-12 09:49:05 +01002111 inarg->lk_flags |= FUSE_LK_FLOCK;
2112 args->in.h.opcode = opcode;
2113 args->in.h.nodeid = get_node_id(inode);
2114 args->in.numargs = 1;
2115 args->in.args[0].size = sizeof(*inarg);
2116 args->in.args[0].value = inarg;
Miklos Szeredi71421252006-06-25 05:48:52 -07002117}
2118
2119static int fuse_getlk(struct file *file, struct file_lock *fl)
2120{
Al Viro6131ffa2013-02-27 16:59:05 -05002121 struct inode *inode = file_inode(file);
Miklos Szeredi71421252006-06-25 05:48:52 -07002122 struct fuse_conn *fc = get_fuse_conn(inode);
Miklos Szeredi70781872014-12-12 09:49:05 +01002123 FUSE_ARGS(args);
2124 struct fuse_lk_in inarg;
Miklos Szeredi71421252006-06-25 05:48:52 -07002125 struct fuse_lk_out outarg;
2126 int err;
2127
Miklos Szeredi70781872014-12-12 09:49:05 +01002128 fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg);
2129 args.out.numargs = 1;
2130 args.out.args[0].size = sizeof(outarg);
2131 args.out.args[0].value = &outarg;
2132 err = fuse_simple_request(fc, &args);
Miklos Szeredi71421252006-06-25 05:48:52 -07002133 if (!err)
2134 err = convert_fuse_file_lock(&outarg.lk, fl);
2135
2136 return err;
2137}
2138
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07002139static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
Miklos Szeredi71421252006-06-25 05:48:52 -07002140{
Al Viro6131ffa2013-02-27 16:59:05 -05002141 struct inode *inode = file_inode(file);
Miklos Szeredi71421252006-06-25 05:48:52 -07002142 struct fuse_conn *fc = get_fuse_conn(inode);
Miklos Szeredi70781872014-12-12 09:49:05 +01002143 FUSE_ARGS(args);
2144 struct fuse_lk_in inarg;
Miklos Szeredi71421252006-06-25 05:48:52 -07002145 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
2146 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
2147 int err;
2148
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002149 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
Miklos Szeredi48e90762008-07-25 01:49:02 -07002150 /* NLM needs asynchronous locks, which we don't support yet */
2151 return -ENOLCK;
2152 }
2153
Miklos Szeredi71421252006-06-25 05:48:52 -07002154 /* Unlock on close is handled by the flush method */
2155 if (fl->fl_flags & FL_CLOSE)
2156 return 0;
2157
Miklos Szeredi70781872014-12-12 09:49:05 +01002158 fuse_lk_fill(&args, file, fl, opcode, pid, flock, &inarg);
2159 err = fuse_simple_request(fc, &args);
Miklos Szeredi71421252006-06-25 05:48:52 -07002160
Miklos Szeredia4d27e72006-06-25 05:48:54 -07002161 /* locking is restartable */
2162 if (err == -EINTR)
2163 err = -ERESTARTSYS;
Miklos Szeredi70781872014-12-12 09:49:05 +01002164
Miklos Szeredi71421252006-06-25 05:48:52 -07002165 return err;
2166}
2167
2168static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
2169{
Al Viro6131ffa2013-02-27 16:59:05 -05002170 struct inode *inode = file_inode(file);
Miklos Szeredi71421252006-06-25 05:48:52 -07002171 struct fuse_conn *fc = get_fuse_conn(inode);
2172 int err;
2173
Miklos Szeredi48e90762008-07-25 01:49:02 -07002174 if (cmd == F_CANCELLK) {
2175 err = 0;
2176 } else if (cmd == F_GETLK) {
Miklos Szeredi71421252006-06-25 05:48:52 -07002177 if (fc->no_lock) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002178 posix_test_lock(file, fl);
Miklos Szeredi71421252006-06-25 05:48:52 -07002179 err = 0;
2180 } else
2181 err = fuse_getlk(file, fl);
2182 } else {
2183 if (fc->no_lock)
Miklos Szeredi48e90762008-07-25 01:49:02 -07002184 err = posix_lock_file(file, fl, NULL);
Miklos Szeredi71421252006-06-25 05:48:52 -07002185 else
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07002186 err = fuse_setlk(file, fl, 0);
Miklos Szeredi71421252006-06-25 05:48:52 -07002187 }
2188 return err;
2189}
2190
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07002191static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
2192{
Al Viro6131ffa2013-02-27 16:59:05 -05002193 struct inode *inode = file_inode(file);
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07002194 struct fuse_conn *fc = get_fuse_conn(inode);
2195 int err;
2196
Miklos Szeredi37fb3a32011-08-08 16:08:08 +02002197 if (fc->no_flock) {
Benjamin Coddington4f656362015-10-22 13:38:14 -04002198 err = locks_lock_file_wait(file, fl);
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07002199 } else {
Miklos Szeredi37fb3a32011-08-08 16:08:08 +02002200 struct fuse_file *ff = file->private_data;
2201
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07002202 /* emulate flock with POSIX locks */
Miklos Szeredi37fb3a32011-08-08 16:08:08 +02002203 ff->flock = true;
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07002204 err = fuse_setlk(file, fl, 1);
2205 }
2206
2207 return err;
2208}
2209
Miklos Szeredib2d22722006-12-06 20:35:51 -08002210static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
2211{
2212 struct inode *inode = mapping->host;
2213 struct fuse_conn *fc = get_fuse_conn(inode);
Miklos Szeredi70781872014-12-12 09:49:05 +01002214 FUSE_ARGS(args);
Miklos Szeredib2d22722006-12-06 20:35:51 -08002215 struct fuse_bmap_in inarg;
2216 struct fuse_bmap_out outarg;
2217 int err;
2218
2219 if (!inode->i_sb->s_bdev || fc->no_bmap)
2220 return 0;
2221
Miklos Szeredib2d22722006-12-06 20:35:51 -08002222 memset(&inarg, 0, sizeof(inarg));
2223 inarg.block = block;
2224 inarg.blocksize = inode->i_sb->s_blocksize;
Miklos Szeredi70781872014-12-12 09:49:05 +01002225 args.in.h.opcode = FUSE_BMAP;
2226 args.in.h.nodeid = get_node_id(inode);
2227 args.in.numargs = 1;
2228 args.in.args[0].size = sizeof(inarg);
2229 args.in.args[0].value = &inarg;
2230 args.out.numargs = 1;
2231 args.out.args[0].size = sizeof(outarg);
2232 args.out.args[0].value = &outarg;
2233 err = fuse_simple_request(fc, &args);
Miklos Szeredib2d22722006-12-06 20:35:51 -08002234 if (err == -ENOSYS)
2235 fc->no_bmap = 1;
2236
2237 return err ? 0 : outarg.block;
2238}
2239
Ravishankar N0b5da8d2015-06-30 23:40:22 +05302240static loff_t fuse_lseek(struct file *file, loff_t offset, int whence)
2241{
2242 struct inode *inode = file->f_mapping->host;
2243 struct fuse_conn *fc = get_fuse_conn(inode);
2244 struct fuse_file *ff = file->private_data;
2245 FUSE_ARGS(args);
2246 struct fuse_lseek_in inarg = {
2247 .fh = ff->fh,
2248 .offset = offset,
2249 .whence = whence
2250 };
2251 struct fuse_lseek_out outarg;
2252 int err;
2253
2254 if (fc->no_lseek)
2255 goto fallback;
2256
2257 args.in.h.opcode = FUSE_LSEEK;
2258 args.in.h.nodeid = ff->nodeid;
2259 args.in.numargs = 1;
2260 args.in.args[0].size = sizeof(inarg);
2261 args.in.args[0].value = &inarg;
2262 args.out.numargs = 1;
2263 args.out.args[0].size = sizeof(outarg);
2264 args.out.args[0].value = &outarg;
2265 err = fuse_simple_request(fc, &args);
2266 if (err) {
2267 if (err == -ENOSYS) {
2268 fc->no_lseek = 1;
2269 goto fallback;
2270 }
2271 return err;
2272 }
2273
2274 return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes);
2275
2276fallback:
2277 err = fuse_update_attributes(inode, NULL, file, NULL);
2278 if (!err)
2279 return generic_file_llseek(file, offset, whence);
2280 else
2281 return err;
2282}
2283
Andrew Morton965c8e52012-12-17 15:59:39 -08002284static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
Miklos Szeredi5559b8f2008-04-30 00:54:45 -07002285{
2286 loff_t retval;
Al Viro6131ffa2013-02-27 16:59:05 -05002287 struct inode *inode = file_inode(file);
Miklos Szeredi5559b8f2008-04-30 00:54:45 -07002288
Ravishankar N0b5da8d2015-06-30 23:40:22 +05302289 switch (whence) {
2290 case SEEK_SET:
2291 case SEEK_CUR:
2292 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
Andrew Morton965c8e52012-12-17 15:59:39 -08002293 retval = generic_file_llseek(file, offset, whence);
Ravishankar N0b5da8d2015-06-30 23:40:22 +05302294 break;
2295 case SEEK_END:
Al Viro59551022016-01-22 15:40:57 -05002296 inode_lock(inode);
Ravishankar N0b5da8d2015-06-30 23:40:22 +05302297 retval = fuse_update_attributes(inode, NULL, file, NULL);
2298 if (!retval)
2299 retval = generic_file_llseek(file, offset, whence);
Al Viro59551022016-01-22 15:40:57 -05002300 inode_unlock(inode);
Ravishankar N0b5da8d2015-06-30 23:40:22 +05302301 break;
2302 case SEEK_HOLE:
2303 case SEEK_DATA:
Al Viro59551022016-01-22 15:40:57 -05002304 inode_lock(inode);
Ravishankar N0b5da8d2015-06-30 23:40:22 +05302305 retval = fuse_lseek(file, offset, whence);
Al Viro59551022016-01-22 15:40:57 -05002306 inode_unlock(inode);
Ravishankar N0b5da8d2015-06-30 23:40:22 +05302307 break;
2308 default:
2309 retval = -EINVAL;
2310 }
Miklos Szeredic07c3d12011-12-13 11:58:48 +01002311
Miklos Szeredi5559b8f2008-04-30 00:54:45 -07002312 return retval;
2313}
2314
Tejun Heo59efec72008-11-26 12:03:55 +01002315static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov,
2316 unsigned int nr_segs, size_t bytes, bool to_user)
2317{
2318 struct iov_iter ii;
2319 int page_idx = 0;
2320
2321 if (!bytes)
2322 return 0;
2323
Al Viro71d8e532014-03-05 19:28:09 -05002324 iov_iter_init(&ii, to_user ? READ : WRITE, iov, nr_segs, bytes);
Tejun Heo59efec72008-11-26 12:03:55 +01002325
2326 while (iov_iter_count(&ii)) {
2327 struct page *page = pages[page_idx++];
2328 size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii));
Dan Carpenter4aa0edd2010-05-07 10:28:17 +02002329 void *kaddr;
Tejun Heo59efec72008-11-26 12:03:55 +01002330
Dan Carpenter4aa0edd2010-05-07 10:28:17 +02002331 kaddr = kmap(page);
Tejun Heo59efec72008-11-26 12:03:55 +01002332
2333 while (todo) {
2334 char __user *uaddr = ii.iov->iov_base + ii.iov_offset;
2335 size_t iov_len = ii.iov->iov_len - ii.iov_offset;
2336 size_t copy = min(todo, iov_len);
2337 size_t left;
2338
2339 if (!to_user)
2340 left = copy_from_user(kaddr, uaddr, copy);
2341 else
2342 left = copy_to_user(uaddr, kaddr, copy);
2343
2344 if (unlikely(left))
2345 return -EFAULT;
2346
2347 iov_iter_advance(&ii, copy);
2348 todo -= copy;
2349 kaddr += copy;
2350 }
2351
Jens Axboe0bd87182009-11-03 11:40:44 +01002352 kunmap(page);
Tejun Heo59efec72008-11-26 12:03:55 +01002353 }
2354
2355 return 0;
2356}
2357
2358/*
Miklos Szeredid9d318d2010-11-30 16:39:27 +01002359 * CUSE servers compiled on 32bit broke on 64bit kernels because the
2360 * ABI was defined to be 'struct iovec' which is different on 32bit
2361 * and 64bit. Fortunately we can determine which structure the server
2362 * used from the size of the reply.
2363 */
Miklos Szeredi1baa26b2010-12-07 20:16:56 +01002364static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
2365 size_t transferred, unsigned count,
2366 bool is_compat)
Miklos Szeredid9d318d2010-11-30 16:39:27 +01002367{
2368#ifdef CONFIG_COMPAT
2369 if (count * sizeof(struct compat_iovec) == transferred) {
2370 struct compat_iovec *ciov = src;
2371 unsigned i;
2372
2373 /*
2374 * With this interface a 32bit server cannot support
2375 * non-compat (i.e. ones coming from 64bit apps) ioctl
2376 * requests
2377 */
2378 if (!is_compat)
2379 return -EINVAL;
2380
2381 for (i = 0; i < count; i++) {
2382 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
2383 dst[i].iov_len = ciov[i].iov_len;
2384 }
2385 return 0;
2386 }
2387#endif
2388
2389 if (count * sizeof(struct iovec) != transferred)
2390 return -EIO;
2391
2392 memcpy(dst, src, transferred);
2393 return 0;
2394}
2395
Miklos Szeredi75727772010-11-30 16:39:27 +01002396/* Make sure iov_length() won't overflow */
2397static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count)
2398{
2399 size_t n;
2400 u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT;
2401
Zach Brownfb6ccff2012-07-24 12:10:11 -07002402 for (n = 0; n < count; n++, iov++) {
Miklos Szeredi75727772010-11-30 16:39:27 +01002403 if (iov->iov_len > (size_t) max)
2404 return -ENOMEM;
2405 max -= iov->iov_len;
2406 }
2407 return 0;
2408}
2409
Miklos Szeredi1baa26b2010-12-07 20:16:56 +01002410static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
2411 void *src, size_t transferred, unsigned count,
2412 bool is_compat)
2413{
2414 unsigned i;
2415 struct fuse_ioctl_iovec *fiov = src;
2416
2417 if (fc->minor < 16) {
2418 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
2419 count, is_compat);
2420 }
2421
2422 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
2423 return -EIO;
2424
2425 for (i = 0; i < count; i++) {
2426 /* Did the server supply an inappropriate value? */
2427 if (fiov[i].base != (unsigned long) fiov[i].base ||
2428 fiov[i].len != (unsigned long) fiov[i].len)
2429 return -EIO;
2430
2431 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
2432 dst[i].iov_len = (size_t) fiov[i].len;
2433
2434#ifdef CONFIG_COMPAT
2435 if (is_compat &&
2436 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
2437 (compat_size_t) dst[i].iov_len != fiov[i].len))
2438 return -EIO;
2439#endif
2440 }
2441
2442 return 0;
2443}
2444
2445
Miklos Szeredid9d318d2010-11-30 16:39:27 +01002446/*
Tejun Heo59efec72008-11-26 12:03:55 +01002447 * For ioctls, there is no generic way to determine how much memory
2448 * needs to be read and/or written. Furthermore, ioctls are allowed
2449 * to dereference the passed pointer, so the parameter requires deep
2450 * copying but FUSE has no idea whatsoever about what to copy in or
2451 * out.
2452 *
2453 * This is solved by allowing FUSE server to retry ioctl with
2454 * necessary in/out iovecs. Let's assume the ioctl implementation
2455 * needs to read in the following structure.
2456 *
2457 * struct a {
2458 * char *buf;
2459 * size_t buflen;
2460 * }
2461 *
2462 * On the first callout to FUSE server, inarg->in_size and
2463 * inarg->out_size will be NULL; then, the server completes the ioctl
2464 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
2465 * the actual iov array to
2466 *
2467 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
2468 *
2469 * which tells FUSE to copy in the requested area and retry the ioctl.
2470 * On the second round, the server has access to the structure and
2471 * from that it can tell what to look for next, so on the invocation,
2472 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
2473 *
2474 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
2475 * { .iov_base = a.buf, .iov_len = a.buflen } }
2476 *
2477 * FUSE will copy both struct a and the pointed buffer from the
2478 * process doing the ioctl and retry ioctl with both struct a and the
2479 * buffer.
2480 *
2481 * This time, FUSE server has everything it needs and completes ioctl
2482 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
2483 *
2484 * Copying data out works the same way.
2485 *
2486 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
2487 * automatically initializes in and out iovs by decoding @cmd with
2488 * _IOC_* macros and the server is not allowed to request RETRY. This
2489 * limits ioctl data transfers to well-formed ioctls and is the forced
2490 * behavior for all FUSE servers.
2491 */
Tejun Heo08cbf542009-04-14 10:54:53 +09002492long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
2493 unsigned int flags)
Tejun Heo59efec72008-11-26 12:03:55 +01002494{
Tejun Heo59efec72008-11-26 12:03:55 +01002495 struct fuse_file *ff = file->private_data;
Miklos Szeredid36f2482009-04-28 16:56:39 +02002496 struct fuse_conn *fc = ff->fc;
Tejun Heo59efec72008-11-26 12:03:55 +01002497 struct fuse_ioctl_in inarg = {
2498 .fh = ff->fh,
2499 .cmd = cmd,
2500 .arg = arg,
2501 .flags = flags
2502 };
2503 struct fuse_ioctl_out outarg;
2504 struct fuse_req *req = NULL;
2505 struct page **pages = NULL;
Miklos Szeredi8ac83502010-12-07 20:16:56 +01002506 struct iovec *iov_page = NULL;
Tejun Heo59efec72008-11-26 12:03:55 +01002507 struct iovec *in_iov = NULL, *out_iov = NULL;
2508 unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
2509 size_t in_size, out_size, transferred;
2510 int err;
2511
Miklos Szeredi1baa26b2010-12-07 20:16:56 +01002512#if BITS_PER_LONG == 32
2513 inarg.flags |= FUSE_IOCTL_32BIT;
2514#else
2515 if (flags & FUSE_IOCTL_COMPAT)
2516 inarg.flags |= FUSE_IOCTL_32BIT;
2517#endif
2518
Tejun Heo59efec72008-11-26 12:03:55 +01002519 /* assume all the iovs returned by client always fits in a page */
Miklos Szeredi1baa26b2010-12-07 20:16:56 +01002520 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
Tejun Heo59efec72008-11-26 12:03:55 +01002521
Tejun Heo59efec72008-11-26 12:03:55 +01002522 err = -ENOMEM;
Thomas Meyerc411cc82011-11-29 22:08:00 +01002523 pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL);
Miklos Szeredi8ac83502010-12-07 20:16:56 +01002524 iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
Tejun Heo59efec72008-11-26 12:03:55 +01002525 if (!pages || !iov_page)
2526 goto out;
2527
2528 /*
2529 * If restricted, initialize IO parameters as encoded in @cmd.
2530 * RETRY from server is not allowed.
2531 */
2532 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
Miklos Szeredi8ac83502010-12-07 20:16:56 +01002533 struct iovec *iov = iov_page;
Tejun Heo59efec72008-11-26 12:03:55 +01002534
Miklos Szeredic9f0523d2008-12-02 14:49:42 +01002535 iov->iov_base = (void __user *)arg;
Tejun Heo59efec72008-11-26 12:03:55 +01002536 iov->iov_len = _IOC_SIZE(cmd);
2537
2538 if (_IOC_DIR(cmd) & _IOC_WRITE) {
2539 in_iov = iov;
2540 in_iovs = 1;
2541 }
2542
2543 if (_IOC_DIR(cmd) & _IOC_READ) {
2544 out_iov = iov;
2545 out_iovs = 1;
2546 }
2547 }
2548
2549 retry:
2550 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
2551 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
2552
2553 /*
2554 * Out data can be used either for actual out data or iovs,
2555 * make sure there always is at least one page.
2556 */
2557 out_size = max_t(size_t, out_size, PAGE_SIZE);
2558 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
2559
2560 /* make sure there are enough buffer pages and init request with them */
2561 err = -ENOMEM;
2562 if (max_pages > FUSE_MAX_PAGES_PER_REQ)
2563 goto out;
2564 while (num_pages < max_pages) {
2565 pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2566 if (!pages[num_pages])
2567 goto out;
2568 num_pages++;
2569 }
2570
Maxim Patlasov54b96672012-10-26 19:49:13 +04002571 req = fuse_get_req(fc, num_pages);
Tejun Heo59efec72008-11-26 12:03:55 +01002572 if (IS_ERR(req)) {
2573 err = PTR_ERR(req);
2574 req = NULL;
2575 goto out;
2576 }
2577 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
2578 req->num_pages = num_pages;
Maxim Patlasov7c190c82012-10-26 19:50:29 +04002579 fuse_page_descs_length_init(req, 0, req->num_pages);
Tejun Heo59efec72008-11-26 12:03:55 +01002580
2581 /* okay, let's send it to the client */
2582 req->in.h.opcode = FUSE_IOCTL;
Miklos Szeredid36f2482009-04-28 16:56:39 +02002583 req->in.h.nodeid = ff->nodeid;
Tejun Heo59efec72008-11-26 12:03:55 +01002584 req->in.numargs = 1;
2585 req->in.args[0].size = sizeof(inarg);
2586 req->in.args[0].value = &inarg;
2587 if (in_size) {
2588 req->in.numargs++;
2589 req->in.args[1].size = in_size;
2590 req->in.argpages = 1;
2591
2592 err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size,
2593 false);
2594 if (err)
2595 goto out;
2596 }
2597
2598 req->out.numargs = 2;
2599 req->out.args[0].size = sizeof(outarg);
2600 req->out.args[0].value = &outarg;
2601 req->out.args[1].size = out_size;
2602 req->out.argpages = 1;
2603 req->out.argvar = 1;
2604
Tejun Heob93f8582008-11-26 12:03:55 +01002605 fuse_request_send(fc, req);
Tejun Heo59efec72008-11-26 12:03:55 +01002606 err = req->out.h.error;
2607 transferred = req->out.args[1].size;
2608 fuse_put_request(fc, req);
2609 req = NULL;
2610 if (err)
2611 goto out;
2612
2613 /* did it ask for retry? */
2614 if (outarg.flags & FUSE_IOCTL_RETRY) {
Miklos Szeredi8ac83502010-12-07 20:16:56 +01002615 void *vaddr;
Tejun Heo59efec72008-11-26 12:03:55 +01002616
2617 /* no retry if in restricted mode */
2618 err = -EIO;
2619 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
2620 goto out;
2621
2622 in_iovs = outarg.in_iovs;
2623 out_iovs = outarg.out_iovs;
2624
2625 /*
2626 * Make sure things are in boundary, separate checks
2627 * are to protect against overflow.
2628 */
2629 err = -ENOMEM;
2630 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
2631 out_iovs > FUSE_IOCTL_MAX_IOV ||
2632 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
2633 goto out;
2634
Cong Wang2408f6e2011-11-25 23:14:30 +08002635 vaddr = kmap_atomic(pages[0]);
Miklos Szeredi1baa26b2010-12-07 20:16:56 +01002636 err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
Miklos Szeredid9d318d2010-11-30 16:39:27 +01002637 transferred, in_iovs + out_iovs,
2638 (flags & FUSE_IOCTL_COMPAT) != 0);
Cong Wang2408f6e2011-11-25 23:14:30 +08002639 kunmap_atomic(vaddr);
Miklos Szeredid9d318d2010-11-30 16:39:27 +01002640 if (err)
2641 goto out;
Tejun Heo59efec72008-11-26 12:03:55 +01002642
Miklos Szeredi8ac83502010-12-07 20:16:56 +01002643 in_iov = iov_page;
Tejun Heo59efec72008-11-26 12:03:55 +01002644 out_iov = in_iov + in_iovs;
2645
Miklos Szeredi75727772010-11-30 16:39:27 +01002646 err = fuse_verify_ioctl_iov(in_iov, in_iovs);
2647 if (err)
2648 goto out;
2649
2650 err = fuse_verify_ioctl_iov(out_iov, out_iovs);
2651 if (err)
2652 goto out;
2653
Tejun Heo59efec72008-11-26 12:03:55 +01002654 goto retry;
2655 }
2656
2657 err = -EIO;
2658 if (transferred > inarg.out_size)
2659 goto out;
2660
2661 err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true);
2662 out:
2663 if (req)
2664 fuse_put_request(fc, req);
Miklos Szeredi8ac83502010-12-07 20:16:56 +01002665 free_page((unsigned long) iov_page);
Tejun Heo59efec72008-11-26 12:03:55 +01002666 while (num_pages)
2667 __free_page(pages[--num_pages]);
2668 kfree(pages);
2669
2670 return err ? err : outarg.result;
2671}
Tejun Heo08cbf542009-04-14 10:54:53 +09002672EXPORT_SYMBOL_GPL(fuse_do_ioctl);
Tejun Heo59efec72008-11-26 12:03:55 +01002673
Miklos Szeredib18da0c2011-12-13 11:58:49 +01002674long fuse_ioctl_common(struct file *file, unsigned int cmd,
2675 unsigned long arg, unsigned int flags)
Miklos Szeredid36f2482009-04-28 16:56:39 +02002676{
Al Viro6131ffa2013-02-27 16:59:05 -05002677 struct inode *inode = file_inode(file);
Miklos Szeredid36f2482009-04-28 16:56:39 +02002678 struct fuse_conn *fc = get_fuse_conn(inode);
2679
Anatol Pomozovc2132c12013-01-14 22:30:00 -08002680 if (!fuse_allow_current_process(fc))
Miklos Szeredid36f2482009-04-28 16:56:39 +02002681 return -EACCES;
2682
2683 if (is_bad_inode(inode))
2684 return -EIO;
2685
2686 return fuse_do_ioctl(file, cmd, arg, flags);
2687}
2688
Tejun Heo59efec72008-11-26 12:03:55 +01002689static long fuse_file_ioctl(struct file *file, unsigned int cmd,
2690 unsigned long arg)
2691{
Miklos Szeredib18da0c2011-12-13 11:58:49 +01002692 return fuse_ioctl_common(file, cmd, arg, 0);
Tejun Heo59efec72008-11-26 12:03:55 +01002693}
2694
2695static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
2696 unsigned long arg)
2697{
Miklos Szeredib18da0c2011-12-13 11:58:49 +01002698 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
Tejun Heo59efec72008-11-26 12:03:55 +01002699}
2700
Tejun Heo95668a62008-11-26 12:03:55 +01002701/*
2702 * All files which have been polled are linked to RB tree
2703 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2704 * find the matching one.
2705 */
2706static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2707 struct rb_node **parent_out)
2708{
2709 struct rb_node **link = &fc->polled_files.rb_node;
2710 struct rb_node *last = NULL;
2711
2712 while (*link) {
2713 struct fuse_file *ff;
2714
2715 last = *link;
2716 ff = rb_entry(last, struct fuse_file, polled_node);
2717
2718 if (kh < ff->kh)
2719 link = &last->rb_left;
2720 else if (kh > ff->kh)
2721 link = &last->rb_right;
2722 else
2723 return link;
2724 }
2725
2726 if (parent_out)
2727 *parent_out = last;
2728 return link;
2729}
2730
2731/*
2732 * The file is about to be polled. Make sure it's on the polled_files
2733 * RB tree. Note that files once added to the polled_files tree are
2734 * not removed before the file is released. This is because a file
2735 * polled once is likely to be polled again.
2736 */
2737static void fuse_register_polled_file(struct fuse_conn *fc,
2738 struct fuse_file *ff)
2739{
2740 spin_lock(&fc->lock);
2741 if (RB_EMPTY_NODE(&ff->polled_node)) {
Rajat Jainf3846262014-02-05 15:24:57 -08002742 struct rb_node **link, *uninitialized_var(parent);
Tejun Heo95668a62008-11-26 12:03:55 +01002743
2744 link = fuse_find_polled_node(fc, ff->kh, &parent);
2745 BUG_ON(*link);
2746 rb_link_node(&ff->polled_node, parent, link);
2747 rb_insert_color(&ff->polled_node, &fc->polled_files);
2748 }
2749 spin_unlock(&fc->lock);
2750}
2751
Tejun Heo08cbf542009-04-14 10:54:53 +09002752unsigned fuse_file_poll(struct file *file, poll_table *wait)
Tejun Heo95668a62008-11-26 12:03:55 +01002753{
Tejun Heo95668a62008-11-26 12:03:55 +01002754 struct fuse_file *ff = file->private_data;
Miklos Szeredi797759a2009-04-28 16:56:41 +02002755 struct fuse_conn *fc = ff->fc;
Tejun Heo95668a62008-11-26 12:03:55 +01002756 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
2757 struct fuse_poll_out outarg;
Miklos Szeredi70781872014-12-12 09:49:05 +01002758 FUSE_ARGS(args);
Tejun Heo95668a62008-11-26 12:03:55 +01002759 int err;
2760
2761 if (fc->no_poll)
2762 return DEFAULT_POLLMASK;
2763
2764 poll_wait(file, &ff->poll_wait, wait);
Enke Chen0415d292013-02-04 16:14:32 +01002765 inarg.events = (__u32)poll_requested_events(wait);
Tejun Heo95668a62008-11-26 12:03:55 +01002766
2767 /*
2768 * Ask for notification iff there's someone waiting for it.
2769 * The client may ignore the flag and always notify.
2770 */
2771 if (waitqueue_active(&ff->poll_wait)) {
2772 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
2773 fuse_register_polled_file(fc, ff);
2774 }
2775
Miklos Szeredi70781872014-12-12 09:49:05 +01002776 args.in.h.opcode = FUSE_POLL;
2777 args.in.h.nodeid = ff->nodeid;
2778 args.in.numargs = 1;
2779 args.in.args[0].size = sizeof(inarg);
2780 args.in.args[0].value = &inarg;
2781 args.out.numargs = 1;
2782 args.out.args[0].size = sizeof(outarg);
2783 args.out.args[0].value = &outarg;
2784 err = fuse_simple_request(fc, &args);
Tejun Heo95668a62008-11-26 12:03:55 +01002785
2786 if (!err)
2787 return outarg.revents;
2788 if (err == -ENOSYS) {
2789 fc->no_poll = 1;
2790 return DEFAULT_POLLMASK;
2791 }
2792 return POLLERR;
2793}
Tejun Heo08cbf542009-04-14 10:54:53 +09002794EXPORT_SYMBOL_GPL(fuse_file_poll);
Tejun Heo95668a62008-11-26 12:03:55 +01002795
2796/*
2797 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2798 * wakes up the poll waiters.
2799 */
2800int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2801 struct fuse_notify_poll_wakeup_out *outarg)
2802{
2803 u64 kh = outarg->kh;
2804 struct rb_node **link;
2805
2806 spin_lock(&fc->lock);
2807
2808 link = fuse_find_polled_node(fc, kh, NULL);
2809 if (*link) {
2810 struct fuse_file *ff;
2811
2812 ff = rb_entry(*link, struct fuse_file, polled_node);
2813 wake_up_interruptible_sync(&ff->poll_wait);
2814 }
2815
2816 spin_unlock(&fc->lock);
2817 return 0;
2818}
2819
Maxim Patlasovefb9fa92012-12-18 14:05:08 +04002820static void fuse_do_truncate(struct file *file)
2821{
2822 struct inode *inode = file->f_mapping->host;
2823 struct iattr attr;
2824
2825 attr.ia_valid = ATTR_SIZE;
2826 attr.ia_size = i_size_read(inode);
2827
2828 attr.ia_file = file;
2829 attr.ia_valid |= ATTR_FILE;
2830
2831 fuse_do_setattr(inode, &attr, file);
2832}
2833
Maxim Patlasove5c5f052013-05-30 16:41:34 +04002834static inline loff_t fuse_round_up(loff_t off)
2835{
2836 return round_up(off, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
2837}
2838
Anand Avati4273b792012-02-17 12:46:25 -05002839static ssize_t
Omar Sandoval22c61862015-03-16 04:33:53 -07002840fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
Anand Avati4273b792012-02-17 12:46:25 -05002841{
Christoph Hellwig9d5722b2015-02-02 14:59:43 +01002842 DECLARE_COMPLETION_ONSTACK(wait);
Anand Avati4273b792012-02-17 12:46:25 -05002843 ssize_t ret = 0;
Miklos Szeredi60b9df72013-05-01 14:37:21 +02002844 struct file *file = iocb->ki_filp;
2845 struct fuse_file *ff = file->private_data;
Maxim Patlasove5c5f052013-05-30 16:41:34 +04002846 bool async_dio = ff->fc->async_dio;
Anand Avati4273b792012-02-17 12:46:25 -05002847 loff_t pos = 0;
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002848 struct inode *inode;
2849 loff_t i_size;
Al Viroa6cbcd42014-03-04 22:38:00 -05002850 size_t count = iov_iter_count(iter);
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04002851 struct fuse_io_priv *io;
Robert Doebbelin7cabc612016-03-07 09:50:56 +01002852 bool is_sync = is_sync_kiocb(iocb);
Anand Avati4273b792012-02-17 12:46:25 -05002853
Anand Avati4273b792012-02-17 12:46:25 -05002854 pos = offset;
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002855 inode = file->f_mapping->host;
2856 i_size = i_size_read(inode);
Anand Avati4273b792012-02-17 12:46:25 -05002857
Omar Sandoval6f673762015-03-16 04:33:52 -07002858 if ((iov_iter_rw(iter) == READ) && (offset > i_size))
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002859 return 0;
2860
Maxim Patlasov439ee5f2012-12-14 19:21:26 +04002861 /* optimization for short read */
Omar Sandoval6f673762015-03-16 04:33:52 -07002862 if (async_dio && iov_iter_rw(iter) != WRITE && offset + count > i_size) {
Maxim Patlasov439ee5f2012-12-14 19:21:26 +04002863 if (offset >= i_size)
2864 return 0;
Al Viro6b775b12015-04-07 15:06:19 -04002865 iov_iter_truncate(iter, fuse_round_up(i_size - offset));
2866 count = iov_iter_count(iter);
Maxim Patlasov439ee5f2012-12-14 19:21:26 +04002867 }
2868
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002869 io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL);
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04002870 if (!io)
2871 return -ENOMEM;
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002872 spin_lock_init(&io->lock);
Seth Forshee744742d2016-03-11 10:35:34 -06002873 kref_init(&io->refcnt);
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002874 io->reqs = 1;
2875 io->bytes = -1;
2876 io->size = 0;
2877 io->offset = offset;
Omar Sandoval6f673762015-03-16 04:33:52 -07002878 io->write = (iov_iter_rw(iter) == WRITE);
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002879 io->err = 0;
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04002880 io->file = file;
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002881 /*
2882 * By default, we want to optimize all I/Os with async request
Miklos Szeredi60b9df72013-05-01 14:37:21 +02002883 * submission to the client filesystem if supported.
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002884 */
Maxim Patlasove5c5f052013-05-30 16:41:34 +04002885 io->async = async_dio;
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002886 io->iocb = iocb;
2887
2888 /*
2889 * We cannot asynchronously extend the size of a file. We have no method
2890 * to wait on real async I/O requests, so we must submit this request
2891 * synchronously.
2892 */
Robert Doebbelin7cabc612016-03-07 09:50:56 +01002893 if (!is_sync && (offset + count > i_size) &&
Omar Sandoval6f673762015-03-16 04:33:52 -07002894 iov_iter_rw(iter) == WRITE)
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002895 io->async = false;
Anand Avati4273b792012-02-17 12:46:25 -05002896
Seth Forshee744742d2016-03-11 10:35:34 -06002897 if (io->async && is_sync) {
2898 /*
2899 * Additional reference to keep io around after
2900 * calling fuse_aio_complete()
2901 */
2902 kref_get(&io->refcnt);
Christoph Hellwig9d5722b2015-02-02 14:59:43 +01002903 io->done = &wait;
Seth Forshee744742d2016-03-11 10:35:34 -06002904 }
Christoph Hellwig9d5722b2015-02-02 14:59:43 +01002905
Omar Sandoval6f673762015-03-16 04:33:52 -07002906 if (iov_iter_rw(iter) == WRITE) {
Al Viro6b775b12015-04-07 15:06:19 -04002907 ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE);
Al Viro812408f2015-03-30 22:15:58 -04002908 fuse_invalidate_attr(inode);
2909 } else {
Al Virod22a9432014-03-16 15:50:47 -04002910 ret = __fuse_direct_read(io, iter, &pos);
Al Viro812408f2015-03-30 22:15:58 -04002911 }
Maxim Patlasov36cf66e2012-12-14 19:20:51 +04002912
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002913 if (io->async) {
2914 fuse_aio_complete(io, ret < 0 ? ret : 0, -1);
2915
2916 /* we have a non-extending, async request, so return */
Robert Doebbelin7cabc612016-03-07 09:50:56 +01002917 if (!is_sync)
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002918 return -EIOCBQUEUED;
2919
Christoph Hellwig9d5722b2015-02-02 14:59:43 +01002920 wait_for_completion(&wait);
2921 ret = fuse_get_res_by_io(io);
Maxim Patlasovbcba24c2012-12-14 19:21:08 +04002922 }
2923
Seth Forshee744742d2016-03-11 10:35:34 -06002924 kref_put(&io->refcnt, fuse_io_release);
Christoph Hellwig9d5722b2015-02-02 14:59:43 +01002925
Omar Sandoval6f673762015-03-16 04:33:52 -07002926 if (iov_iter_rw(iter) == WRITE) {
Maxim Patlasovefb9fa92012-12-18 14:05:08 +04002927 if (ret > 0)
2928 fuse_write_update_size(inode, pos);
2929 else if (ret < 0 && offset + count > i_size)
2930 fuse_do_truncate(file);
2931 }
Anand Avati4273b792012-02-17 12:46:25 -05002932
2933 return ret;
2934}
2935
Miklos Szeredicdadb112012-11-10 16:55:56 +01002936static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
2937 loff_t length)
Anatol Pomozov05ba1f02012-04-22 18:45:24 -07002938{
2939 struct fuse_file *ff = file->private_data;
Miklos Szeredi1c682712014-12-12 10:04:51 +01002940 struct inode *inode = file_inode(file);
Maxim Patlasov0ab08f52013-09-13 19:20:16 +04002941 struct fuse_inode *fi = get_fuse_inode(inode);
Anatol Pomozov05ba1f02012-04-22 18:45:24 -07002942 struct fuse_conn *fc = ff->fc;
Miklos Szeredi70781872014-12-12 09:49:05 +01002943 FUSE_ARGS(args);
Anatol Pomozov05ba1f02012-04-22 18:45:24 -07002944 struct fuse_fallocate_in inarg = {
2945 .fh = ff->fh,
2946 .offset = offset,
2947 .length = length,
2948 .mode = mode
2949 };
2950 int err;
Maxim Patlasov14c14412013-06-13 12:16:39 +04002951 bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) ||
2952 (mode & FALLOC_FL_PUNCH_HOLE);
Anatol Pomozov05ba1f02012-04-22 18:45:24 -07002953
Miklos Szeredi4adb8302014-04-28 14:19:21 +02002954 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2955 return -EOPNOTSUPP;
2956
Miklos Szeredi519c6042012-04-26 10:56:36 +02002957 if (fc->no_fallocate)
2958 return -EOPNOTSUPP;
2959
Maxim Patlasov14c14412013-06-13 12:16:39 +04002960 if (lock_inode) {
Al Viro59551022016-01-22 15:40:57 -05002961 inode_lock(inode);
Maxim Patlasovbde52782013-09-13 19:19:54 +04002962 if (mode & FALLOC_FL_PUNCH_HOLE) {
2963 loff_t endbyte = offset + length - 1;
2964 err = filemap_write_and_wait_range(inode->i_mapping,
2965 offset, endbyte);
2966 if (err)
2967 goto out;
2968
2969 fuse_sync_writes(inode);
2970 }
Brian Foster3634a632013-05-17 09:30:32 -04002971 }
2972
Maxim Patlasov0ab08f52013-09-13 19:20:16 +04002973 if (!(mode & FALLOC_FL_KEEP_SIZE))
2974 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
2975
Miklos Szeredi70781872014-12-12 09:49:05 +01002976 args.in.h.opcode = FUSE_FALLOCATE;
2977 args.in.h.nodeid = ff->nodeid;
2978 args.in.numargs = 1;
2979 args.in.args[0].size = sizeof(inarg);
2980 args.in.args[0].value = &inarg;
2981 err = fuse_simple_request(fc, &args);
Miklos Szeredi519c6042012-04-26 10:56:36 +02002982 if (err == -ENOSYS) {
2983 fc->no_fallocate = 1;
2984 err = -EOPNOTSUPP;
2985 }
Brian Fosterbee6c302013-05-17 15:27:34 -04002986 if (err)
2987 goto out;
2988
2989 /* we could have extended the file */
Maxim Patlasovb0aa7602013-12-26 19:51:11 +04002990 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
2991 bool changed = fuse_write_update_size(inode, offset + length);
2992
Miklos Szeredi93d22692014-04-28 14:19:22 +02002993 if (changed && fc->writeback_cache)
2994 file_update_time(file);
Maxim Patlasovb0aa7602013-12-26 19:51:11 +04002995 }
Brian Fosterbee6c302013-05-17 15:27:34 -04002996
2997 if (mode & FALLOC_FL_PUNCH_HOLE)
2998 truncate_pagecache_range(inode, offset, offset + length - 1);
2999
3000 fuse_invalidate_attr(inode);
3001
Brian Foster3634a632013-05-17 09:30:32 -04003002out:
Maxim Patlasov0ab08f52013-09-13 19:20:16 +04003003 if (!(mode & FALLOC_FL_KEEP_SIZE))
3004 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
3005
Maxim Patlasovbde52782013-09-13 19:19:54 +04003006 if (lock_inode)
Al Viro59551022016-01-22 15:40:57 -05003007 inode_unlock(inode);
Brian Foster3634a632013-05-17 09:30:32 -04003008
Anatol Pomozov05ba1f02012-04-22 18:45:24 -07003009 return err;
3010}
Anatol Pomozov05ba1f02012-04-22 18:45:24 -07003011
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08003012static const struct file_operations fuse_file_operations = {
Miklos Szeredi5559b8f2008-04-30 00:54:45 -07003013 .llseek = fuse_file_llseek,
Al Viro37c20f12014-04-02 14:47:09 -04003014 .read_iter = fuse_file_read_iter,
Al Viro84c3d552014-04-03 14:33:23 -04003015 .write_iter = fuse_file_write_iter,
Miklos Szeredib6aeade2005-09-09 13:10:30 -07003016 .mmap = fuse_file_mmap,
3017 .open = fuse_open,
3018 .flush = fuse_flush,
3019 .release = fuse_release,
3020 .fsync = fuse_fsync,
Miklos Szeredi71421252006-06-25 05:48:52 -07003021 .lock = fuse_file_lock,
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07003022 .flock = fuse_file_flock,
Jens Axboe5ffc4ef2007-06-01 11:49:19 +02003023 .splice_read = generic_file_splice_read,
Tejun Heo59efec72008-11-26 12:03:55 +01003024 .unlocked_ioctl = fuse_file_ioctl,
3025 .compat_ioctl = fuse_file_compat_ioctl,
Tejun Heo95668a62008-11-26 12:03:55 +01003026 .poll = fuse_file_poll,
Anatol Pomozov05ba1f02012-04-22 18:45:24 -07003027 .fallocate = fuse_file_fallocate,
Miklos Szeredib6aeade2005-09-09 13:10:30 -07003028};
3029
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08003030static const struct file_operations fuse_direct_io_file_operations = {
Miklos Szeredi5559b8f2008-04-30 00:54:45 -07003031 .llseek = fuse_file_llseek,
Al Viro15316262015-03-30 22:08:36 -04003032 .read_iter = fuse_direct_read_iter,
Al Viro15316262015-03-30 22:08:36 -04003033 .write_iter = fuse_direct_write_iter,
Miklos Szeredifc280c92009-04-02 14:25:35 +02003034 .mmap = fuse_direct_mmap,
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07003035 .open = fuse_open,
3036 .flush = fuse_flush,
3037 .release = fuse_release,
3038 .fsync = fuse_fsync,
Miklos Szeredi71421252006-06-25 05:48:52 -07003039 .lock = fuse_file_lock,
Miklos Szeredia9ff4f82007-10-18 03:07:02 -07003040 .flock = fuse_file_flock,
Tejun Heo59efec72008-11-26 12:03:55 +01003041 .unlocked_ioctl = fuse_file_ioctl,
3042 .compat_ioctl = fuse_file_compat_ioctl,
Tejun Heo95668a62008-11-26 12:03:55 +01003043 .poll = fuse_file_poll,
Anatol Pomozov05ba1f02012-04-22 18:45:24 -07003044 .fallocate = fuse_file_fallocate,
Miklos Szeredifc280c92009-04-02 14:25:35 +02003045 /* no splice_read */
Miklos Szeredi413ef8c2005-09-09 13:10:35 -07003046};
3047
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003048static const struct address_space_operations fuse_file_aops = {
Miklos Szeredib6aeade2005-09-09 13:10:30 -07003049 .readpage = fuse_readpage,
Miklos Szeredi3be5a522008-04-30 00:54:41 -07003050 .writepage = fuse_writepage,
Pavel Emelyanov26d614d2013-06-29 21:45:29 +04003051 .writepages = fuse_writepages,
Miklos Szeredi3be5a522008-04-30 00:54:41 -07003052 .launder_page = fuse_launder_page,
Miklos Szeredidb50b962005-09-09 13:10:33 -07003053 .readpages = fuse_readpages,
Miklos Szeredi3be5a522008-04-30 00:54:41 -07003054 .set_page_dirty = __set_page_dirty_nobuffers,
Miklos Szeredib2d22722006-12-06 20:35:51 -08003055 .bmap = fuse_bmap,
Anand Avati4273b792012-02-17 12:46:25 -05003056 .direct_IO = fuse_direct_IO,
Pavel Emelyanov6b12c1b2013-10-10 17:11:43 +04003057 .write_begin = fuse_write_begin,
3058 .write_end = fuse_write_end,
Miklos Szeredib6aeade2005-09-09 13:10:30 -07003059};
3060
3061void fuse_init_file_inode(struct inode *inode)
3062{
Miklos Szeredi45323fb2005-09-09 13:10:37 -07003063 inode->i_fop = &fuse_file_operations;
3064 inode->i_data.a_ops = &fuse_file_aops;
Miklos Szeredib6aeade2005-09-09 13:10:30 -07003065}