Commit | Line | Data |
---|---|---|
a62a8ef9 SH |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * virtio-fs: Virtio Filesystem | |
4 | * Copyright (C) 2018 Red Hat, Inc. | |
5 | */ | |
6 | ||
7 | #include <linux/fs.h> | |
22f3787e SH |
8 | #include <linux/dax.h> |
9 | #include <linux/pci.h> | |
529395d2 PJG |
10 | #include <linux/interrupt.h> |
11 | #include <linux/group_cpus.h> | |
22f3787e | 12 | #include <linux/pfn_t.h> |
dc90f084 | 13 | #include <linux/memremap.h> |
a62a8ef9 SH |
14 | #include <linux/module.h> |
15 | #include <linux/virtio.h> | |
16 | #include <linux/virtio_fs.h> | |
17 | #include <linux/delay.h> | |
18 | #include <linux/fs_context.h> | |
1dd53957 | 19 | #include <linux/fs_parser.h> |
a62a8ef9 | 20 | #include <linux/highmem.h> |
562ce828 | 21 | #include <linux/cleanup.h> |
22f3787e | 22 | #include <linux/uio.h> |
a62a8ef9 SH |
23 | #include "fuse_i.h" |
24 | ||
a7f0d7aa CK |
25 | /* Used to help calculate the FUSE connection's max_pages limit for a request's |
26 | * size. Parts of the struct fuse_req are sliced into scattergather lists in | |
27 | * addition to the pages used, so this can help account for that overhead. | |
28 | */ | |
29 | #define FUSE_HEADER_OVERHEAD 4 | |
30 | ||
a62a8ef9 SH |
31 | /* List of virtio-fs device instances and a lock for the list. Also provides |
32 | * mutual exclusion in device removal and mounting path | |
33 | */ | |
34 | static DEFINE_MUTEX(virtio_fs_mutex); | |
35 | static LIST_HEAD(virtio_fs_instances); | |
36 | ||
a8f62f50 SH |
37 | /* The /sys/fs/virtio_fs/ kset */ |
38 | static struct kset *virtio_fs_kset; | |
39 | ||
a62a8ef9 SH |
40 | enum { |
41 | VQ_HIPRIO, | |
42 | VQ_REQUEST | |
43 | }; | |
44 | ||
b43b7e81 VG |
45 | #define VQ_NAME_LEN 24 |
46 | ||
a62a8ef9 SH |
47 | /* Per-virtqueue state */ |
48 | struct virtio_fs_vq { | |
49 | spinlock_t lock; | |
50 | struct virtqueue *vq; /* protected by ->lock */ | |
51 | struct work_struct done_work; | |
52 | struct list_head queued_reqs; | |
51fecdd2 | 53 | struct list_head end_reqs; /* End these requests */ |
a62a8ef9 SH |
54 | struct delayed_work dispatch_work; |
55 | struct fuse_dev *fud; | |
56 | bool connected; | |
57 | long in_flight; | |
724c15a4 | 58 | struct completion in_flight_zero; /* No inflight requests */ |
b43b7e81 | 59 | char name[VQ_NAME_LEN]; |
a62a8ef9 SH |
60 | } ____cacheline_aligned_in_smp; |
61 | ||
62 | /* A virtio-fs device instance */ | |
63 | struct virtio_fs { | |
a8f62f50 | 64 | struct kobject kobj; |
a62a8ef9 SH |
65 | struct list_head list; /* on virtio_fs_instances */ |
66 | char *tag; | |
67 | struct virtio_fs_vq *vqs; | |
68 | unsigned int nvqs; /* number of virtqueues */ | |
69 | unsigned int num_request_queues; /* number of request queues */ | |
22f3787e SH |
70 | struct dax_device *dax_dev; |
71 | ||
529395d2 PJG |
72 | unsigned int *mq_map; /* index = cpu id, value = request vq id */ |
73 | ||
22f3787e SH |
74 | /* DAX memory window where file contents are mapped */ |
75 | void *window_kaddr; | |
76 | phys_addr_t window_phys_addr; | |
77 | size_t window_len; | |
a62a8ef9 SH |
78 | }; |
79 | ||
1efcf39e | 80 | struct virtio_fs_forget_req { |
a62a8ef9 SH |
81 | struct fuse_in_header ih; |
82 | struct fuse_forget_in arg; | |
1efcf39e VG |
83 | }; |
84 | ||
85 | struct virtio_fs_forget { | |
a62a8ef9 SH |
86 | /* This request can be temporarily queued on virt queue */ |
87 | struct list_head list; | |
1efcf39e | 88 | struct virtio_fs_forget_req req; |
a62a8ef9 SH |
89 | }; |
90 | ||
bb737bbe VG |
91 | struct virtio_fs_req_work { |
92 | struct fuse_req *req; | |
93 | struct virtio_fs_vq *fsvq; | |
94 | struct work_struct done_work; | |
95 | }; | |
96 | ||
a9bfd9dd VG |
97 | static int virtio_fs_enqueue_req(struct virtio_fs_vq *fsvq, |
98 | struct fuse_req *req, bool in_flight); | |
99 | ||
780b1b95 JX |
100 | static const struct constant_table dax_param_enums[] = { |
101 | {"always", FUSE_DAX_ALWAYS }, | |
102 | {"never", FUSE_DAX_NEVER }, | |
103 | {"inode", FUSE_DAX_INODE_USER }, | |
104 | {} | |
105 | }; | |
106 | ||
1dd53957 VG |
107 | enum { |
108 | OPT_DAX, | |
780b1b95 | 109 | OPT_DAX_ENUM, |
1dd53957 VG |
110 | }; |
111 | ||
112 | static const struct fs_parameter_spec virtio_fs_parameters[] = { | |
113 | fsparam_flag("dax", OPT_DAX), | |
780b1b95 | 114 | fsparam_enum("dax", OPT_DAX_ENUM, dax_param_enums), |
1dd53957 VG |
115 | {} |
116 | }; | |
117 | ||
84c21507 | 118 | static int virtio_fs_parse_param(struct fs_context *fsc, |
1dd53957 VG |
119 | struct fs_parameter *param) |
120 | { | |
121 | struct fs_parse_result result; | |
84c21507 | 122 | struct fuse_fs_context *ctx = fsc->fs_private; |
1dd53957 VG |
123 | int opt; |
124 | ||
84c21507 | 125 | opt = fs_parse(fsc, virtio_fs_parameters, param, &result); |
1dd53957 VG |
126 | if (opt < 0) |
127 | return opt; | |
128 | ||
129 | switch (opt) { | |
130 | case OPT_DAX: | |
780b1b95 JX |
131 | ctx->dax_mode = FUSE_DAX_ALWAYS; |
132 | break; | |
133 | case OPT_DAX_ENUM: | |
134 | ctx->dax_mode = result.uint_32; | |
1dd53957 VG |
135 | break; |
136 | default: | |
137 | return -EINVAL; | |
138 | } | |
139 | ||
140 | return 0; | |
141 | } | |
142 | ||
84c21507 | 143 | static void virtio_fs_free_fsc(struct fs_context *fsc) |
1dd53957 | 144 | { |
84c21507 | 145 | struct fuse_fs_context *ctx = fsc->fs_private; |
1dd53957 VG |
146 | |
147 | kfree(ctx); | |
148 | } | |
149 | ||
a62a8ef9 SH |
150 | static inline struct virtio_fs_vq *vq_to_fsvq(struct virtqueue *vq) |
151 | { | |
152 | struct virtio_fs *fs = vq->vdev->priv; | |
153 | ||
154 | return &fs->vqs[vq->index]; | |
155 | } | |
156 | ||
c17ea009 VG |
157 | /* Should be called with fsvq->lock held. */ |
158 | static inline void inc_in_flight_req(struct virtio_fs_vq *fsvq) | |
159 | { | |
160 | fsvq->in_flight++; | |
161 | } | |
162 | ||
163 | /* Should be called with fsvq->lock held. */ | |
164 | static inline void dec_in_flight_req(struct virtio_fs_vq *fsvq) | |
165 | { | |
166 | WARN_ON(fsvq->in_flight <= 0); | |
167 | fsvq->in_flight--; | |
724c15a4 VG |
168 | if (!fsvq->in_flight) |
169 | complete(&fsvq->in_flight_zero); | |
c17ea009 VG |
170 | } |
171 | ||
a8f62f50 SH |
172 | static ssize_t tag_show(struct kobject *kobj, |
173 | struct kobj_attribute *attr, char *buf) | |
174 | { | |
175 | struct virtio_fs *fs = container_of(kobj, struct virtio_fs, kobj); | |
176 | ||
96d88f65 | 177 | return sysfs_emit(buf, "%s\n", fs->tag); |
a8f62f50 SH |
178 | } |
179 | ||
180 | static struct kobj_attribute virtio_fs_tag_attr = __ATTR_RO(tag); | |
181 | ||
182 | static struct attribute *virtio_fs_attrs[] = { | |
183 | &virtio_fs_tag_attr.attr, | |
184 | NULL | |
185 | }; | |
186 | ATTRIBUTE_GROUPS(virtio_fs); | |
187 | ||
188 | static void virtio_fs_ktype_release(struct kobject *kobj) | |
a62a8ef9 | 189 | { |
a8f62f50 | 190 | struct virtio_fs *vfs = container_of(kobj, struct virtio_fs, kobj); |
a62a8ef9 | 191 | |
529395d2 | 192 | kfree(vfs->mq_map); |
a62a8ef9 SH |
193 | kfree(vfs->vqs); |
194 | kfree(vfs); | |
195 | } | |
196 | ||
a8f62f50 SH |
197 | static const struct kobj_type virtio_fs_ktype = { |
198 | .release = virtio_fs_ktype_release, | |
199 | .sysfs_ops = &kobj_sysfs_ops, | |
200 | .default_groups = virtio_fs_groups, | |
201 | }; | |
202 | ||
a62a8ef9 SH |
203 | /* Make sure virtiofs_mutex is held */ |
204 | static void virtio_fs_put(struct virtio_fs *fs) | |
205 | { | |
a8f62f50 | 206 | kobject_put(&fs->kobj); |
a62a8ef9 SH |
207 | } |
208 | ||
209 | static void virtio_fs_fiq_release(struct fuse_iqueue *fiq) | |
210 | { | |
211 | struct virtio_fs *vfs = fiq->priv; | |
212 | ||
213 | mutex_lock(&virtio_fs_mutex); | |
214 | virtio_fs_put(vfs); | |
215 | mutex_unlock(&virtio_fs_mutex); | |
216 | } | |
217 | ||
218 | static void virtio_fs_drain_queue(struct virtio_fs_vq *fsvq) | |
219 | { | |
220 | WARN_ON(fsvq->in_flight < 0); | |
221 | ||
222 | /* Wait for in flight requests to finish.*/ | |
724c15a4 VG |
223 | spin_lock(&fsvq->lock); |
224 | if (fsvq->in_flight) { | |
225 | /* We are holding virtio_fs_mutex. There should not be any | |
226 | * waiters waiting for completion. | |
227 | */ | |
228 | reinit_completion(&fsvq->in_flight_zero); | |
229 | spin_unlock(&fsvq->lock); | |
230 | wait_for_completion(&fsvq->in_flight_zero); | |
231 | } else { | |
a62a8ef9 | 232 | spin_unlock(&fsvq->lock); |
a62a8ef9 SH |
233 | } |
234 | ||
235 | flush_work(&fsvq->done_work); | |
236 | flush_delayed_work(&fsvq->dispatch_work); | |
237 | } | |
238 | ||
724c15a4 | 239 | static void virtio_fs_drain_all_queues_locked(struct virtio_fs *fs) |
a62a8ef9 SH |
240 | { |
241 | struct virtio_fs_vq *fsvq; | |
242 | int i; | |
243 | ||
244 | for (i = 0; i < fs->nvqs; i++) { | |
245 | fsvq = &fs->vqs[i]; | |
a62a8ef9 SH |
246 | virtio_fs_drain_queue(fsvq); |
247 | } | |
248 | } | |
249 | ||
724c15a4 VG |
250 | static void virtio_fs_drain_all_queues(struct virtio_fs *fs) |
251 | { | |
252 | /* Provides mutual exclusion between ->remove and ->kill_sb | |
253 | * paths. We don't want both of these draining queue at the | |
254 | * same time. Current completion logic reinits completion | |
255 | * and that means there should not be any other thread | |
256 | * doing reinit or waiting for completion already. | |
257 | */ | |
258 | mutex_lock(&virtio_fs_mutex); | |
259 | virtio_fs_drain_all_queues_locked(fs); | |
260 | mutex_unlock(&virtio_fs_mutex); | |
261 | } | |
262 | ||
a62a8ef9 SH |
263 | static void virtio_fs_start_all_queues(struct virtio_fs *fs) |
264 | { | |
265 | struct virtio_fs_vq *fsvq; | |
266 | int i; | |
267 | ||
268 | for (i = 0; i < fs->nvqs; i++) { | |
269 | fsvq = &fs->vqs[i]; | |
270 | spin_lock(&fsvq->lock); | |
271 | fsvq->connected = true; | |
272 | spin_unlock(&fsvq->lock); | |
273 | } | |
274 | } | |
275 | ||
276 | /* Add a new instance to the list or return -EEXIST if tag name exists*/ | |
a8f62f50 SH |
277 | static int virtio_fs_add_instance(struct virtio_device *vdev, |
278 | struct virtio_fs *fs) | |
a62a8ef9 SH |
279 | { |
280 | struct virtio_fs *fs2; | |
a8f62f50 | 281 | int ret; |
a62a8ef9 SH |
282 | |
283 | mutex_lock(&virtio_fs_mutex); | |
284 | ||
285 | list_for_each_entry(fs2, &virtio_fs_instances, list) { | |
a8f62f50 SH |
286 | if (strcmp(fs->tag, fs2->tag) == 0) { |
287 | mutex_unlock(&virtio_fs_mutex); | |
288 | return -EEXIST; | |
289 | } | |
a62a8ef9 SH |
290 | } |
291 | ||
a8f62f50 SH |
292 | /* Use the virtio_device's index as a unique identifier, there is no |
293 | * need to allocate our own identifiers because the virtio_fs instance | |
294 | * is only visible to userspace as long as the underlying virtio_device | |
295 | * exists. | |
296 | */ | |
297 | fs->kobj.kset = virtio_fs_kset; | |
298 | ret = kobject_add(&fs->kobj, NULL, "%d", vdev->index); | |
299 | if (ret < 0) { | |
300 | mutex_unlock(&virtio_fs_mutex); | |
301 | return ret; | |
302 | } | |
303 | ||
304 | ret = sysfs_create_link(&fs->kobj, &vdev->dev.kobj, "device"); | |
305 | if (ret < 0) { | |
306 | kobject_del(&fs->kobj); | |
307 | mutex_unlock(&virtio_fs_mutex); | |
308 | return ret; | |
a62a8ef9 SH |
309 | } |
310 | ||
a8f62f50 | 311 | list_add_tail(&fs->list, &virtio_fs_instances); |
a62a8ef9 SH |
312 | |
313 | mutex_unlock(&virtio_fs_mutex); | |
314 | ||
9086b2d9 SH |
315 | kobject_uevent(&fs->kobj, KOBJ_ADD); |
316 | ||
a62a8ef9 SH |
317 | return 0; |
318 | } | |
319 | ||
320 | /* Return the virtio_fs with a given tag, or NULL */ | |
321 | static struct virtio_fs *virtio_fs_find_instance(const char *tag) | |
322 | { | |
323 | struct virtio_fs *fs; | |
324 | ||
325 | mutex_lock(&virtio_fs_mutex); | |
326 | ||
327 | list_for_each_entry(fs, &virtio_fs_instances, list) { | |
328 | if (strcmp(fs->tag, tag) == 0) { | |
a8f62f50 | 329 | kobject_get(&fs->kobj); |
a62a8ef9 SH |
330 | goto found; |
331 | } | |
332 | } | |
333 | ||
334 | fs = NULL; /* not found */ | |
335 | ||
336 | found: | |
337 | mutex_unlock(&virtio_fs_mutex); | |
338 | ||
339 | return fs; | |
340 | } | |
341 | ||
342 | static void virtio_fs_free_devs(struct virtio_fs *fs) | |
343 | { | |
344 | unsigned int i; | |
345 | ||
346 | for (i = 0; i < fs->nvqs; i++) { | |
347 | struct virtio_fs_vq *fsvq = &fs->vqs[i]; | |
348 | ||
349 | if (!fsvq->fud) | |
350 | continue; | |
351 | ||
352 | fuse_dev_free(fsvq->fud); | |
353 | fsvq->fud = NULL; | |
354 | } | |
355 | } | |
356 | ||
357 | /* Read filesystem name from virtio config into fs->tag (must kfree()). */ | |
358 | static int virtio_fs_read_tag(struct virtio_device *vdev, struct virtio_fs *fs) | |
359 | { | |
360 | char tag_buf[sizeof_field(struct virtio_fs_config, tag)]; | |
361 | char *end; | |
362 | size_t len; | |
363 | ||
364 | virtio_cread_bytes(vdev, offsetof(struct virtio_fs_config, tag), | |
365 | &tag_buf, sizeof(tag_buf)); | |
366 | end = memchr(tag_buf, '\0', sizeof(tag_buf)); | |
367 | if (end == tag_buf) | |
368 | return -EINVAL; /* empty tag */ | |
369 | if (!end) | |
370 | end = &tag_buf[sizeof(tag_buf)]; | |
371 | ||
372 | len = end - tag_buf; | |
373 | fs->tag = devm_kmalloc(&vdev->dev, len + 1, GFP_KERNEL); | |
374 | if (!fs->tag) | |
375 | return -ENOMEM; | |
376 | memcpy(fs->tag, tag_buf, len); | |
377 | fs->tag[len] = '\0'; | |
40488cc1 SH |
378 | |
379 | /* While the VIRTIO specification allows any character, newlines are | |
380 | * awkward on mount(8) command-lines and cause problems in the sysfs | |
381 | * "tag" attr and uevent TAG= properties. Forbid them. | |
382 | */ | |
383 | if (strchr(fs->tag, '\n')) { | |
384 | dev_dbg(&vdev->dev, "refusing virtiofs tag with newline character\n"); | |
385 | return -EINVAL; | |
386 | } | |
387 | ||
a62a8ef9 SH |
388 | return 0; |
389 | } | |
390 | ||
391 | /* Work function for hiprio completion */ | |
392 | static void virtio_fs_hiprio_done_work(struct work_struct *work) | |
393 | { | |
394 | struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq, | |
395 | done_work); | |
396 | struct virtqueue *vq = fsvq->vq; | |
397 | ||
398 | /* Free completed FUSE_FORGET requests */ | |
399 | spin_lock(&fsvq->lock); | |
400 | do { | |
401 | unsigned int len; | |
402 | void *req; | |
403 | ||
404 | virtqueue_disable_cb(vq); | |
405 | ||
406 | while ((req = virtqueue_get_buf(vq, &len)) != NULL) { | |
407 | kfree(req); | |
c17ea009 | 408 | dec_in_flight_req(fsvq); |
a62a8ef9 | 409 | } |
f9c29137 | 410 | } while (!virtqueue_enable_cb(vq)); |
a62a8ef9 SH |
411 | spin_unlock(&fsvq->lock); |
412 | } | |
413 | ||
51fecdd2 | 414 | static void virtio_fs_request_dispatch_work(struct work_struct *work) |
a62a8ef9 | 415 | { |
51fecdd2 VG |
416 | struct fuse_req *req; |
417 | struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq, | |
418 | dispatch_work.work); | |
a9bfd9dd | 419 | int ret; |
51fecdd2 VG |
420 | |
421 | pr_debug("virtio-fs: worker %s called.\n", __func__); | |
422 | while (1) { | |
423 | spin_lock(&fsvq->lock); | |
424 | req = list_first_entry_or_null(&fsvq->end_reqs, struct fuse_req, | |
425 | list); | |
426 | if (!req) { | |
427 | spin_unlock(&fsvq->lock); | |
a9bfd9dd | 428 | break; |
51fecdd2 VG |
429 | } |
430 | ||
431 | list_del_init(&req->list); | |
432 | spin_unlock(&fsvq->lock); | |
8f622e94 | 433 | fuse_request_end(req); |
51fecdd2 | 434 | } |
a9bfd9dd VG |
435 | |
436 | /* Dispatch pending requests */ | |
437 | while (1) { | |
438 | spin_lock(&fsvq->lock); | |
439 | req = list_first_entry_or_null(&fsvq->queued_reqs, | |
440 | struct fuse_req, list); | |
441 | if (!req) { | |
442 | spin_unlock(&fsvq->lock); | |
443 | return; | |
444 | } | |
445 | list_del_init(&req->list); | |
446 | spin_unlock(&fsvq->lock); | |
447 | ||
448 | ret = virtio_fs_enqueue_req(fsvq, req, true); | |
449 | if (ret < 0) { | |
450 | if (ret == -ENOMEM || ret == -ENOSPC) { | |
451 | spin_lock(&fsvq->lock); | |
452 | list_add_tail(&req->list, &fsvq->queued_reqs); | |
453 | schedule_delayed_work(&fsvq->dispatch_work, | |
454 | msecs_to_jiffies(1)); | |
455 | spin_unlock(&fsvq->lock); | |
456 | return; | |
457 | } | |
458 | req->out.h.error = ret; | |
459 | spin_lock(&fsvq->lock); | |
460 | dec_in_flight_req(fsvq); | |
461 | spin_unlock(&fsvq->lock); | |
462 | pr_err("virtio-fs: virtio_fs_enqueue_req() failed %d\n", | |
463 | ret); | |
8f622e94 | 464 | fuse_request_end(req); |
a9bfd9dd VG |
465 | } |
466 | } | |
a62a8ef9 SH |
467 | } |
468 | ||
58ada94f VG |
469 | /* |
470 | * Returns 1 if queue is full and sender should wait a bit before sending | |
471 | * next request, 0 otherwise. | |
472 | */ | |
473 | static int send_forget_request(struct virtio_fs_vq *fsvq, | |
474 | struct virtio_fs_forget *forget, | |
475 | bool in_flight) | |
476 | { | |
477 | struct scatterlist sg; | |
478 | struct virtqueue *vq; | |
479 | int ret = 0; | |
480 | bool notify; | |
1efcf39e | 481 | struct virtio_fs_forget_req *req = &forget->req; |
58ada94f VG |
482 | |
483 | spin_lock(&fsvq->lock); | |
484 | if (!fsvq->connected) { | |
485 | if (in_flight) | |
486 | dec_in_flight_req(fsvq); | |
487 | kfree(forget); | |
488 | goto out; | |
489 | } | |
490 | ||
1efcf39e | 491 | sg_init_one(&sg, req, sizeof(*req)); |
58ada94f VG |
492 | vq = fsvq->vq; |
493 | dev_dbg(&vq->vdev->dev, "%s\n", __func__); | |
494 | ||
495 | ret = virtqueue_add_outbuf(vq, &sg, 1, forget, GFP_ATOMIC); | |
496 | if (ret < 0) { | |
497 | if (ret == -ENOMEM || ret == -ENOSPC) { | |
498 | pr_debug("virtio-fs: Could not queue FORGET: err=%d. Will try later\n", | |
499 | ret); | |
500 | list_add_tail(&forget->list, &fsvq->queued_reqs); | |
501 | schedule_delayed_work(&fsvq->dispatch_work, | |
502 | msecs_to_jiffies(1)); | |
503 | if (!in_flight) | |
504 | inc_in_flight_req(fsvq); | |
505 | /* Queue is full */ | |
506 | ret = 1; | |
507 | } else { | |
508 | pr_debug("virtio-fs: Could not queue FORGET: err=%d. Dropping it.\n", | |
509 | ret); | |
510 | kfree(forget); | |
511 | if (in_flight) | |
512 | dec_in_flight_req(fsvq); | |
513 | } | |
514 | goto out; | |
515 | } | |
516 | ||
517 | if (!in_flight) | |
518 | inc_in_flight_req(fsvq); | |
519 | notify = virtqueue_kick_prepare(vq); | |
520 | spin_unlock(&fsvq->lock); | |
521 | ||
522 | if (notify) | |
523 | virtqueue_notify(vq); | |
524 | return ret; | |
525 | out: | |
526 | spin_unlock(&fsvq->lock); | |
527 | return ret; | |
528 | } | |
529 | ||
a62a8ef9 SH |
530 | static void virtio_fs_hiprio_dispatch_work(struct work_struct *work) |
531 | { | |
532 | struct virtio_fs_forget *forget; | |
533 | struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq, | |
534 | dispatch_work.work); | |
a62a8ef9 SH |
535 | pr_debug("virtio-fs: worker %s called.\n", __func__); |
536 | while (1) { | |
537 | spin_lock(&fsvq->lock); | |
538 | forget = list_first_entry_or_null(&fsvq->queued_reqs, | |
539 | struct virtio_fs_forget, list); | |
540 | if (!forget) { | |
541 | spin_unlock(&fsvq->lock); | |
542 | return; | |
543 | } | |
544 | ||
545 | list_del(&forget->list); | |
a62a8ef9 | 546 | spin_unlock(&fsvq->lock); |
58ada94f VG |
547 | if (send_forget_request(fsvq, forget, true)) |
548 | return; | |
a62a8ef9 SH |
549 | } |
550 | } | |
551 | ||
552 | /* Allocate and copy args into req->argbuf */ | |
553 | static int copy_args_to_argbuf(struct fuse_req *req) | |
554 | { | |
555 | struct fuse_args *args = req->args; | |
556 | unsigned int offset = 0; | |
557 | unsigned int num_in; | |
558 | unsigned int num_out; | |
559 | unsigned int len; | |
560 | unsigned int i; | |
561 | ||
562 | num_in = args->in_numargs - args->in_pages; | |
563 | num_out = args->out_numargs - args->out_pages; | |
564 | len = fuse_len_args(num_in, (struct fuse_arg *) args->in_args) + | |
565 | fuse_len_args(num_out, args->out_args); | |
566 | ||
567 | req->argbuf = kmalloc(len, GFP_ATOMIC); | |
568 | if (!req->argbuf) | |
569 | return -ENOMEM; | |
570 | ||
571 | for (i = 0; i < num_in; i++) { | |
572 | memcpy(req->argbuf + offset, | |
573 | args->in_args[i].value, | |
574 | args->in_args[i].size); | |
575 | offset += args->in_args[i].size; | |
576 | } | |
577 | ||
578 | return 0; | |
579 | } | |
580 | ||
581 | /* Copy args out of and free req->argbuf */ | |
582 | static void copy_args_from_argbuf(struct fuse_args *args, struct fuse_req *req) | |
583 | { | |
584 | unsigned int remaining; | |
585 | unsigned int offset; | |
586 | unsigned int num_in; | |
587 | unsigned int num_out; | |
588 | unsigned int i; | |
589 | ||
590 | remaining = req->out.h.len - sizeof(req->out.h); | |
591 | num_in = args->in_numargs - args->in_pages; | |
592 | num_out = args->out_numargs - args->out_pages; | |
593 | offset = fuse_len_args(num_in, (struct fuse_arg *)args->in_args); | |
594 | ||
595 | for (i = 0; i < num_out; i++) { | |
596 | unsigned int argsize = args->out_args[i].size; | |
597 | ||
598 | if (args->out_argvar && | |
599 | i == args->out_numargs - 1 && | |
600 | argsize > remaining) { | |
601 | argsize = remaining; | |
602 | } | |
603 | ||
604 | memcpy(args->out_args[i].value, req->argbuf + offset, argsize); | |
605 | offset += argsize; | |
606 | ||
607 | if (i != args->out_numargs - 1) | |
608 | remaining -= argsize; | |
609 | } | |
610 | ||
611 | /* Store the actual size of the variable-length arg */ | |
612 | if (args->out_argvar) | |
613 | args->out_args[args->out_numargs - 1].size = remaining; | |
614 | ||
615 | kfree(req->argbuf); | |
616 | req->argbuf = NULL; | |
617 | } | |
618 | ||
619 | /* Work function for request completion */ | |
bb737bbe VG |
620 | static void virtio_fs_request_complete(struct fuse_req *req, |
621 | struct virtio_fs_vq *fsvq) | |
622 | { | |
623 | struct fuse_pqueue *fpq = &fsvq->fud->pq; | |
bb737bbe VG |
624 | struct fuse_args *args; |
625 | struct fuse_args_pages *ap; | |
626 | unsigned int len, i, thislen; | |
627 | struct page *page; | |
628 | ||
629 | /* | |
630 | * TODO verify that server properly follows FUSE protocol | |
631 | * (oh.uniq, oh.len) | |
632 | */ | |
633 | args = req->args; | |
634 | copy_args_from_argbuf(args, req); | |
635 | ||
636 | if (args->out_pages && args->page_zeroing) { | |
637 | len = args->out_args[args->out_numargs - 1].size; | |
638 | ap = container_of(args, typeof(*ap), args); | |
639 | for (i = 0; i < ap->num_pages; i++) { | |
640 | thislen = ap->descs[i].length; | |
641 | if (len < thislen) { | |
642 | WARN_ON(ap->descs[i].offset); | |
643 | page = ap->pages[i]; | |
644 | zero_user_segment(page, len, thislen); | |
645 | len = 0; | |
646 | } else { | |
647 | len -= thislen; | |
648 | } | |
649 | } | |
650 | } | |
651 | ||
652 | spin_lock(&fpq->lock); | |
653 | clear_bit(FR_SENT, &req->flags); | |
654 | spin_unlock(&fpq->lock); | |
655 | ||
8f622e94 | 656 | fuse_request_end(req); |
bb737bbe VG |
657 | spin_lock(&fsvq->lock); |
658 | dec_in_flight_req(fsvq); | |
659 | spin_unlock(&fsvq->lock); | |
660 | } | |
661 | ||
662 | static void virtio_fs_complete_req_work(struct work_struct *work) | |
663 | { | |
664 | struct virtio_fs_req_work *w = | |
665 | container_of(work, typeof(*w), done_work); | |
666 | ||
667 | virtio_fs_request_complete(w->req, w->fsvq); | |
668 | kfree(w); | |
669 | } | |
670 | ||
a62a8ef9 SH |
671 | static void virtio_fs_requests_done_work(struct work_struct *work) |
672 | { | |
673 | struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq, | |
674 | done_work); | |
675 | struct fuse_pqueue *fpq = &fsvq->fud->pq; | |
a62a8ef9 SH |
676 | struct virtqueue *vq = fsvq->vq; |
677 | struct fuse_req *req; | |
a62a8ef9 | 678 | struct fuse_req *next; |
bb737bbe | 679 | unsigned int len; |
a62a8ef9 SH |
680 | LIST_HEAD(reqs); |
681 | ||
682 | /* Collect completed requests off the virtqueue */ | |
683 | spin_lock(&fsvq->lock); | |
684 | do { | |
685 | virtqueue_disable_cb(vq); | |
686 | ||
687 | while ((req = virtqueue_get_buf(vq, &len)) != NULL) { | |
688 | spin_lock(&fpq->lock); | |
689 | list_move_tail(&req->list, &reqs); | |
690 | spin_unlock(&fpq->lock); | |
691 | } | |
f9c29137 | 692 | } while (!virtqueue_enable_cb(vq)); |
a62a8ef9 SH |
693 | spin_unlock(&fsvq->lock); |
694 | ||
695 | /* End requests */ | |
696 | list_for_each_entry_safe(req, next, &reqs, list) { | |
a62a8ef9 | 697 | list_del_init(&req->list); |
a62a8ef9 | 698 | |
bb737bbe VG |
699 | /* blocking async request completes in a worker context */ |
700 | if (req->args->may_block) { | |
701 | struct virtio_fs_req_work *w; | |
702 | ||
703 | w = kzalloc(sizeof(*w), GFP_NOFS | __GFP_NOFAIL); | |
704 | INIT_WORK(&w->done_work, virtio_fs_complete_req_work); | |
705 | w->fsvq = fsvq; | |
706 | w->req = req; | |
707 | schedule_work(&w->done_work); | |
708 | } else { | |
709 | virtio_fs_request_complete(req, fsvq); | |
710 | } | |
a62a8ef9 SH |
711 | } |
712 | } | |
713 | ||
529395d2 PJG |
714 | static void virtio_fs_map_queues(struct virtio_device *vdev, struct virtio_fs *fs) |
715 | { | |
716 | const struct cpumask *mask, *masks; | |
717 | unsigned int q, cpu; | |
718 | ||
719 | /* First attempt to map using existing transport layer affinities | |
720 | * e.g. PCIe MSI-X | |
721 | */ | |
722 | if (!vdev->config->get_vq_affinity) | |
723 | goto fallback; | |
724 | ||
725 | for (q = 0; q < fs->num_request_queues; q++) { | |
726 | mask = vdev->config->get_vq_affinity(vdev, VQ_REQUEST + q); | |
727 | if (!mask) | |
728 | goto fallback; | |
729 | ||
730 | for_each_cpu(cpu, mask) | |
731 | fs->mq_map[cpu] = q; | |
732 | } | |
733 | ||
734 | return; | |
735 | fallback: | |
736 | /* Attempt to map evenly in groups over the CPUs */ | |
737 | masks = group_cpus_evenly(fs->num_request_queues); | |
738 | /* If even this fails we default to all CPUs use queue zero */ | |
739 | if (!masks) { | |
740 | for_each_possible_cpu(cpu) | |
741 | fs->mq_map[cpu] = 0; | |
742 | return; | |
743 | } | |
744 | ||
745 | for (q = 0; q < fs->num_request_queues; q++) { | |
746 | for_each_cpu(cpu, &masks[q]) | |
747 | fs->mq_map[cpu] = q; | |
748 | } | |
749 | kfree(masks); | |
750 | } | |
751 | ||
a62a8ef9 SH |
752 | /* Virtqueue interrupt handler */ |
753 | static void virtio_fs_vq_done(struct virtqueue *vq) | |
754 | { | |
755 | struct virtio_fs_vq *fsvq = vq_to_fsvq(vq); | |
756 | ||
757 | dev_dbg(&vq->vdev->dev, "%s %s\n", __func__, fsvq->name); | |
758 | ||
759 | schedule_work(&fsvq->done_work); | |
760 | } | |
761 | ||
b43b7e81 VG |
762 | static void virtio_fs_init_vq(struct virtio_fs_vq *fsvq, char *name, |
763 | int vq_type) | |
764 | { | |
7c594bbd | 765 | strscpy(fsvq->name, name, VQ_NAME_LEN); |
b43b7e81 VG |
766 | spin_lock_init(&fsvq->lock); |
767 | INIT_LIST_HEAD(&fsvq->queued_reqs); | |
768 | INIT_LIST_HEAD(&fsvq->end_reqs); | |
769 | init_completion(&fsvq->in_flight_zero); | |
770 | ||
771 | if (vq_type == VQ_REQUEST) { | |
772 | INIT_WORK(&fsvq->done_work, virtio_fs_requests_done_work); | |
773 | INIT_DELAYED_WORK(&fsvq->dispatch_work, | |
774 | virtio_fs_request_dispatch_work); | |
775 | } else { | |
776 | INIT_WORK(&fsvq->done_work, virtio_fs_hiprio_done_work); | |
777 | INIT_DELAYED_WORK(&fsvq->dispatch_work, | |
778 | virtio_fs_hiprio_dispatch_work); | |
779 | } | |
780 | } | |
781 | ||
a62a8ef9 SH |
782 | /* Initialize virtqueues */ |
783 | static int virtio_fs_setup_vqs(struct virtio_device *vdev, | |
784 | struct virtio_fs *fs) | |
785 | { | |
786 | struct virtqueue **vqs; | |
787 | vq_callback_t **callbacks; | |
529395d2 PJG |
788 | /* Specify pre_vectors to ensure that the queues before the |
789 | * request queues (e.g. hiprio) don't claim any of the CPUs in | |
790 | * the multi-queue mapping and interrupt affinities | |
791 | */ | |
792 | struct irq_affinity desc = { .pre_vectors = VQ_REQUEST }; | |
a62a8ef9 SH |
793 | const char **names; |
794 | unsigned int i; | |
795 | int ret = 0; | |
796 | ||
2c0349ec MT |
797 | virtio_cread_le(vdev, struct virtio_fs_config, num_request_queues, |
798 | &fs->num_request_queues); | |
a62a8ef9 SH |
799 | if (fs->num_request_queues == 0) |
800 | return -EINVAL; | |
801 | ||
103c2de1 PJG |
802 | /* Truncate nr of request queues to nr_cpu_id */ |
803 | fs->num_request_queues = min_t(unsigned int, fs->num_request_queues, | |
804 | nr_cpu_ids); | |
b43b7e81 | 805 | fs->nvqs = VQ_REQUEST + fs->num_request_queues; |
a62a8ef9 SH |
806 | fs->vqs = kcalloc(fs->nvqs, sizeof(fs->vqs[VQ_HIPRIO]), GFP_KERNEL); |
807 | if (!fs->vqs) | |
808 | return -ENOMEM; | |
809 | ||
810 | vqs = kmalloc_array(fs->nvqs, sizeof(vqs[VQ_HIPRIO]), GFP_KERNEL); | |
811 | callbacks = kmalloc_array(fs->nvqs, sizeof(callbacks[VQ_HIPRIO]), | |
812 | GFP_KERNEL); | |
813 | names = kmalloc_array(fs->nvqs, sizeof(names[VQ_HIPRIO]), GFP_KERNEL); | |
529395d2 PJG |
814 | fs->mq_map = kcalloc_node(nr_cpu_ids, sizeof(*fs->mq_map), GFP_KERNEL, |
815 | dev_to_node(&vdev->dev)); | |
816 | if (!vqs || !callbacks || !names || !fs->mq_map) { | |
a62a8ef9 SH |
817 | ret = -ENOMEM; |
818 | goto out; | |
819 | } | |
820 | ||
b43b7e81 | 821 | /* Initialize the hiprio/forget request virtqueue */ |
a62a8ef9 | 822 | callbacks[VQ_HIPRIO] = virtio_fs_vq_done; |
b43b7e81 | 823 | virtio_fs_init_vq(&fs->vqs[VQ_HIPRIO], "hiprio", VQ_HIPRIO); |
a62a8ef9 | 824 | names[VQ_HIPRIO] = fs->vqs[VQ_HIPRIO].name; |
a62a8ef9 SH |
825 | |
826 | /* Initialize the requests virtqueues */ | |
827 | for (i = VQ_REQUEST; i < fs->nvqs; i++) { | |
b43b7e81 VG |
828 | char vq_name[VQ_NAME_LEN]; |
829 | ||
830 | snprintf(vq_name, VQ_NAME_LEN, "requests.%u", i - VQ_REQUEST); | |
831 | virtio_fs_init_vq(&fs->vqs[i], vq_name, VQ_REQUEST); | |
a62a8ef9 SH |
832 | callbacks[i] = virtio_fs_vq_done; |
833 | names[i] = fs->vqs[i].name; | |
834 | } | |
835 | ||
529395d2 | 836 | ret = virtio_find_vqs(vdev, fs->nvqs, vqs, callbacks, names, &desc); |
a62a8ef9 SH |
837 | if (ret < 0) |
838 | goto out; | |
839 | ||
840 | for (i = 0; i < fs->nvqs; i++) | |
841 | fs->vqs[i].vq = vqs[i]; | |
842 | ||
843 | virtio_fs_start_all_queues(fs); | |
844 | out: | |
845 | kfree(names); | |
846 | kfree(callbacks); | |
847 | kfree(vqs); | |
529395d2 | 848 | if (ret) { |
a62a8ef9 | 849 | kfree(fs->vqs); |
529395d2 PJG |
850 | kfree(fs->mq_map); |
851 | } | |
a62a8ef9 SH |
852 | return ret; |
853 | } | |
854 | ||
855 | /* Free virtqueues (device must already be reset) */ | |
1e5b9e04 | 856 | static void virtio_fs_cleanup_vqs(struct virtio_device *vdev) |
a62a8ef9 SH |
857 | { |
858 | vdev->config->del_vqs(vdev); | |
859 | } | |
860 | ||
22f3787e SH |
861 | /* Map a window offset to a page frame number. The window offset will have |
862 | * been produced by .iomap_begin(), which maps a file offset to a window | |
863 | * offset. | |
864 | */ | |
865 | static long virtio_fs_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, | |
e511c4a3 JC |
866 | long nr_pages, enum dax_access_mode mode, |
867 | void **kaddr, pfn_t *pfn) | |
22f3787e SH |
868 | { |
869 | struct virtio_fs *fs = dax_get_private(dax_dev); | |
870 | phys_addr_t offset = PFN_PHYS(pgoff); | |
73fb2c8b | 871 | size_t max_nr_pages = fs->window_len / PAGE_SIZE - pgoff; |
22f3787e SH |
872 | |
873 | if (kaddr) | |
874 | *kaddr = fs->window_kaddr + offset; | |
875 | if (pfn) | |
876 | *pfn = phys_to_pfn_t(fs->window_phys_addr + offset, | |
877 | PFN_DEV | PFN_MAP); | |
878 | return nr_pages > max_nr_pages ? max_nr_pages : nr_pages; | |
879 | } | |
880 | ||
22f3787e SH |
881 | static int virtio_fs_zero_page_range(struct dax_device *dax_dev, |
882 | pgoff_t pgoff, size_t nr_pages) | |
883 | { | |
884 | long rc; | |
885 | void *kaddr; | |
886 | ||
e511c4a3 JC |
887 | rc = dax_direct_access(dax_dev, pgoff, nr_pages, DAX_ACCESS, &kaddr, |
888 | NULL); | |
22f3787e | 889 | if (rc < 0) |
1ea7ca1b JC |
890 | return dax_mem2blk_err(rc); |
891 | ||
22f3787e SH |
892 | memset(kaddr, 0, nr_pages << PAGE_SHIFT); |
893 | dax_flush(dax_dev, kaddr, nr_pages << PAGE_SHIFT); | |
894 | return 0; | |
895 | } | |
896 | ||
897 | static const struct dax_operations virtio_fs_dax_ops = { | |
898 | .direct_access = virtio_fs_direct_access, | |
22f3787e SH |
899 | .zero_page_range = virtio_fs_zero_page_range, |
900 | }; | |
901 | ||
902 | static void virtio_fs_cleanup_dax(void *data) | |
903 | { | |
904 | struct dax_device *dax_dev = data; | |
905 | ||
906 | kill_dax(dax_dev); | |
907 | put_dax(dax_dev); | |
908 | } | |
909 | ||
562ce828 MD |
910 | DEFINE_FREE(cleanup_dax, struct dax_dev *, if (!IS_ERR_OR_NULL(_T)) virtio_fs_cleanup_dax(_T)) |
911 | ||
22f3787e SH |
912 | static int virtio_fs_setup_dax(struct virtio_device *vdev, struct virtio_fs *fs) |
913 | { | |
562ce828 | 914 | struct dax_device *dax_dev __free(cleanup_dax) = NULL; |
22f3787e SH |
915 | struct virtio_shm_region cache_reg; |
916 | struct dev_pagemap *pgmap; | |
917 | bool have_cache; | |
918 | ||
919 | if (!IS_ENABLED(CONFIG_FUSE_DAX)) | |
920 | return 0; | |
921 | ||
562ce828 MD |
922 | dax_dev = alloc_dax(fs, &virtio_fs_dax_ops); |
923 | if (IS_ERR(dax_dev)) { | |
924 | int rc = PTR_ERR(dax_dev); | |
925 | return rc == -EOPNOTSUPP ? 0 : rc; | |
926 | } | |
927 | ||
22f3787e SH |
928 | /* Get cache region */ |
929 | have_cache = virtio_get_shm_region(vdev, &cache_reg, | |
930 | (u8)VIRTIO_FS_SHMCAP_ID_CACHE); | |
931 | if (!have_cache) { | |
932 | dev_notice(&vdev->dev, "%s: No cache capability\n", __func__); | |
933 | return 0; | |
934 | } | |
935 | ||
936 | if (!devm_request_mem_region(&vdev->dev, cache_reg.addr, cache_reg.len, | |
937 | dev_name(&vdev->dev))) { | |
938 | dev_warn(&vdev->dev, "could not reserve region addr=0x%llx len=0x%llx\n", | |
939 | cache_reg.addr, cache_reg.len); | |
940 | return -EBUSY; | |
941 | } | |
942 | ||
943 | dev_notice(&vdev->dev, "Cache len: 0x%llx @ 0x%llx\n", cache_reg.len, | |
944 | cache_reg.addr); | |
945 | ||
946 | pgmap = devm_kzalloc(&vdev->dev, sizeof(*pgmap), GFP_KERNEL); | |
947 | if (!pgmap) | |
948 | return -ENOMEM; | |
949 | ||
950 | pgmap->type = MEMORY_DEVICE_FS_DAX; | |
951 | ||
952 | /* Ideally we would directly use the PCI BAR resource but | |
953 | * devm_memremap_pages() wants its own copy in pgmap. So | |
954 | * initialize a struct resource from scratch (only the start | |
955 | * and end fields will be used). | |
956 | */ | |
69456535 | 957 | pgmap->range = (struct range) { |
22f3787e SH |
958 | .start = (phys_addr_t) cache_reg.addr, |
959 | .end = (phys_addr_t) cache_reg.addr + cache_reg.len - 1, | |
960 | }; | |
69456535 | 961 | pgmap->nr_range = 1; |
22f3787e SH |
962 | |
963 | fs->window_kaddr = devm_memremap_pages(&vdev->dev, pgmap); | |
964 | if (IS_ERR(fs->window_kaddr)) | |
965 | return PTR_ERR(fs->window_kaddr); | |
966 | ||
967 | fs->window_phys_addr = (phys_addr_t) cache_reg.addr; | |
968 | fs->window_len = (phys_addr_t) cache_reg.len; | |
969 | ||
970 | dev_dbg(&vdev->dev, "%s: window kaddr 0x%px phys_addr 0x%llx len 0x%llx\n", | |
971 | __func__, fs->window_kaddr, cache_reg.addr, cache_reg.len); | |
972 | ||
562ce828 | 973 | fs->dax_dev = no_free_ptr(dax_dev); |
22f3787e SH |
974 | return devm_add_action_or_reset(&vdev->dev, virtio_fs_cleanup_dax, |
975 | fs->dax_dev); | |
976 | } | |
977 | ||
a62a8ef9 SH |
978 | static int virtio_fs_probe(struct virtio_device *vdev) |
979 | { | |
980 | struct virtio_fs *fs; | |
981 | int ret; | |
982 | ||
983 | fs = kzalloc(sizeof(*fs), GFP_KERNEL); | |
984 | if (!fs) | |
985 | return -ENOMEM; | |
a8f62f50 | 986 | kobject_init(&fs->kobj, &virtio_fs_ktype); |
a62a8ef9 SH |
987 | vdev->priv = fs; |
988 | ||
989 | ret = virtio_fs_read_tag(vdev, fs); | |
990 | if (ret < 0) | |
991 | goto out; | |
992 | ||
993 | ret = virtio_fs_setup_vqs(vdev, fs); | |
994 | if (ret < 0) | |
995 | goto out; | |
996 | ||
529395d2 | 997 | virtio_fs_map_queues(vdev, fs); |
a62a8ef9 | 998 | |
22f3787e SH |
999 | ret = virtio_fs_setup_dax(vdev, fs); |
1000 | if (ret < 0) | |
1001 | goto out_vqs; | |
1002 | ||
a62a8ef9 SH |
1003 | /* Bring the device online in case the filesystem is mounted and |
1004 | * requests need to be sent before we return. | |
1005 | */ | |
1006 | virtio_device_ready(vdev); | |
1007 | ||
a8f62f50 | 1008 | ret = virtio_fs_add_instance(vdev, fs); |
a62a8ef9 SH |
1009 | if (ret < 0) |
1010 | goto out_vqs; | |
1011 | ||
1012 | return 0; | |
1013 | ||
1014 | out_vqs: | |
d9679d00 | 1015 | virtio_reset_device(vdev); |
1e5b9e04 | 1016 | virtio_fs_cleanup_vqs(vdev); |
a62a8ef9 SH |
1017 | |
1018 | out: | |
1019 | vdev->priv = NULL; | |
a8f62f50 | 1020 | kobject_put(&fs->kobj); |
a62a8ef9 SH |
1021 | return ret; |
1022 | } | |
1023 | ||
1024 | static void virtio_fs_stop_all_queues(struct virtio_fs *fs) | |
1025 | { | |
1026 | struct virtio_fs_vq *fsvq; | |
1027 | int i; | |
1028 | ||
1029 | for (i = 0; i < fs->nvqs; i++) { | |
1030 | fsvq = &fs->vqs[i]; | |
1031 | spin_lock(&fsvq->lock); | |
1032 | fsvq->connected = false; | |
1033 | spin_unlock(&fsvq->lock); | |
1034 | } | |
1035 | } | |
1036 | ||
1037 | static void virtio_fs_remove(struct virtio_device *vdev) | |
1038 | { | |
1039 | struct virtio_fs *fs = vdev->priv; | |
1040 | ||
1041 | mutex_lock(&virtio_fs_mutex); | |
1042 | /* This device is going away. No one should get new reference */ | |
1043 | list_del_init(&fs->list); | |
a8f62f50 SH |
1044 | sysfs_remove_link(&fs->kobj, "device"); |
1045 | kobject_del(&fs->kobj); | |
a62a8ef9 | 1046 | virtio_fs_stop_all_queues(fs); |
724c15a4 | 1047 | virtio_fs_drain_all_queues_locked(fs); |
d9679d00 | 1048 | virtio_reset_device(vdev); |
1e5b9e04 | 1049 | virtio_fs_cleanup_vqs(vdev); |
a62a8ef9 SH |
1050 | |
1051 | vdev->priv = NULL; | |
1052 | /* Put device reference on virtio_fs object */ | |
1053 | virtio_fs_put(fs); | |
1054 | mutex_unlock(&virtio_fs_mutex); | |
1055 | } | |
1056 | ||
1057 | #ifdef CONFIG_PM_SLEEP | |
1058 | static int virtio_fs_freeze(struct virtio_device *vdev) | |
1059 | { | |
1060 | /* TODO need to save state here */ | |
1061 | pr_warn("virtio-fs: suspend/resume not yet supported\n"); | |
1062 | return -EOPNOTSUPP; | |
1063 | } | |
1064 | ||
1065 | static int virtio_fs_restore(struct virtio_device *vdev) | |
1066 | { | |
1067 | /* TODO need to restore state here */ | |
1068 | return 0; | |
1069 | } | |
1070 | #endif /* CONFIG_PM_SLEEP */ | |
1071 | ||
00929447 | 1072 | static const struct virtio_device_id id_table[] = { |
a62a8ef9 SH |
1073 | { VIRTIO_ID_FS, VIRTIO_DEV_ANY_ID }, |
1074 | {}, | |
1075 | }; | |
1076 | ||
00929447 | 1077 | static const unsigned int feature_table[] = {}; |
a62a8ef9 SH |
1078 | |
1079 | static struct virtio_driver virtio_fs_driver = { | |
1080 | .driver.name = KBUILD_MODNAME, | |
1081 | .driver.owner = THIS_MODULE, | |
1082 | .id_table = id_table, | |
1083 | .feature_table = feature_table, | |
1084 | .feature_table_size = ARRAY_SIZE(feature_table), | |
1085 | .probe = virtio_fs_probe, | |
1086 | .remove = virtio_fs_remove, | |
1087 | #ifdef CONFIG_PM_SLEEP | |
1088 | .freeze = virtio_fs_freeze, | |
1089 | .restore = virtio_fs_restore, | |
1090 | #endif | |
1091 | }; | |
1092 | ||
1093 | static void virtio_fs_wake_forget_and_unlock(struct fuse_iqueue *fiq) | |
1094 | __releases(fiq->lock) | |
1095 | { | |
1096 | struct fuse_forget_link *link; | |
1097 | struct virtio_fs_forget *forget; | |
1efcf39e | 1098 | struct virtio_fs_forget_req *req; |
a62a8ef9 | 1099 | struct virtio_fs *fs; |
a62a8ef9 | 1100 | struct virtio_fs_vq *fsvq; |
a62a8ef9 | 1101 | u64 unique; |
a62a8ef9 SH |
1102 | |
1103 | link = fuse_dequeue_forget(fiq, 1, NULL); | |
1104 | unique = fuse_get_unique(fiq); | |
1105 | ||
1106 | fs = fiq->priv; | |
1107 | fsvq = &fs->vqs[VQ_HIPRIO]; | |
1108 | spin_unlock(&fiq->lock); | |
1109 | ||
1110 | /* Allocate a buffer for the request */ | |
1111 | forget = kmalloc(sizeof(*forget), GFP_NOFS | __GFP_NOFAIL); | |
1efcf39e | 1112 | req = &forget->req; |
a62a8ef9 | 1113 | |
1efcf39e | 1114 | req->ih = (struct fuse_in_header){ |
a62a8ef9 SH |
1115 | .opcode = FUSE_FORGET, |
1116 | .nodeid = link->forget_one.nodeid, | |
1117 | .unique = unique, | |
1efcf39e | 1118 | .len = sizeof(*req), |
a62a8ef9 | 1119 | }; |
1efcf39e | 1120 | req->arg = (struct fuse_forget_in){ |
a62a8ef9 SH |
1121 | .nlookup = link->forget_one.nlookup, |
1122 | }; | |
1123 | ||
58ada94f | 1124 | send_forget_request(fsvq, forget, false); |
a62a8ef9 SH |
1125 | kfree(link); |
1126 | } | |
1127 | ||
1128 | static void virtio_fs_wake_interrupt_and_unlock(struct fuse_iqueue *fiq) | |
1129 | __releases(fiq->lock) | |
1130 | { | |
1131 | /* | |
1132 | * TODO interrupts. | |
1133 | * | |
1134 | * Normal fs operations on a local filesystems aren't interruptible. | |
1135 | * Exceptions are blocking lock operations; for example fcntl(F_SETLKW) | |
1136 | * with shared lock between host and guest. | |
1137 | */ | |
1138 | spin_unlock(&fiq->lock); | |
1139 | } | |
1140 | ||
42d3e2d0 VG |
1141 | /* Count number of scatter-gather elements required */ |
1142 | static unsigned int sg_count_fuse_pages(struct fuse_page_desc *page_descs, | |
1143 | unsigned int num_pages, | |
1144 | unsigned int total_len) | |
1145 | { | |
1146 | unsigned int i; | |
1147 | unsigned int this_len; | |
1148 | ||
1149 | for (i = 0; i < num_pages && total_len; i++) { | |
1150 | this_len = min(page_descs[i].length, total_len); | |
1151 | total_len -= this_len; | |
1152 | } | |
1153 | ||
1154 | return i; | |
1155 | } | |
1156 | ||
a62a8ef9 SH |
1157 | /* Return the number of scatter-gather list elements required */ |
1158 | static unsigned int sg_count_fuse_req(struct fuse_req *req) | |
1159 | { | |
1160 | struct fuse_args *args = req->args; | |
1161 | struct fuse_args_pages *ap = container_of(args, typeof(*ap), args); | |
42d3e2d0 | 1162 | unsigned int size, total_sgs = 1 /* fuse_in_header */; |
a62a8ef9 SH |
1163 | |
1164 | if (args->in_numargs - args->in_pages) | |
1165 | total_sgs += 1; | |
1166 | ||
42d3e2d0 VG |
1167 | if (args->in_pages) { |
1168 | size = args->in_args[args->in_numargs - 1].size; | |
1169 | total_sgs += sg_count_fuse_pages(ap->descs, ap->num_pages, | |
1170 | size); | |
1171 | } | |
a62a8ef9 SH |
1172 | |
1173 | if (!test_bit(FR_ISREPLY, &req->flags)) | |
1174 | return total_sgs; | |
1175 | ||
1176 | total_sgs += 1 /* fuse_out_header */; | |
1177 | ||
1178 | if (args->out_numargs - args->out_pages) | |
1179 | total_sgs += 1; | |
1180 | ||
42d3e2d0 VG |
1181 | if (args->out_pages) { |
1182 | size = args->out_args[args->out_numargs - 1].size; | |
1183 | total_sgs += sg_count_fuse_pages(ap->descs, ap->num_pages, | |
1184 | size); | |
1185 | } | |
a62a8ef9 SH |
1186 | |
1187 | return total_sgs; | |
1188 | } | |
1189 | ||
1190 | /* Add pages to scatter-gather list and return number of elements used */ | |
1191 | static unsigned int sg_init_fuse_pages(struct scatterlist *sg, | |
1192 | struct page **pages, | |
1193 | struct fuse_page_desc *page_descs, | |
1194 | unsigned int num_pages, | |
1195 | unsigned int total_len) | |
1196 | { | |
1197 | unsigned int i; | |
1198 | unsigned int this_len; | |
1199 | ||
1200 | for (i = 0; i < num_pages && total_len; i++) { | |
1201 | sg_init_table(&sg[i], 1); | |
1202 | this_len = min(page_descs[i].length, total_len); | |
1203 | sg_set_page(&sg[i], pages[i], this_len, page_descs[i].offset); | |
1204 | total_len -= this_len; | |
1205 | } | |
1206 | ||
1207 | return i; | |
1208 | } | |
1209 | ||
1210 | /* Add args to scatter-gather list and return number of elements used */ | |
1211 | static unsigned int sg_init_fuse_args(struct scatterlist *sg, | |
1212 | struct fuse_req *req, | |
1213 | struct fuse_arg *args, | |
1214 | unsigned int numargs, | |
1215 | bool argpages, | |
1216 | void *argbuf, | |
1217 | unsigned int *len_used) | |
1218 | { | |
1219 | struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args); | |
1220 | unsigned int total_sgs = 0; | |
1221 | unsigned int len; | |
1222 | ||
1223 | len = fuse_len_args(numargs - argpages, args); | |
1224 | if (len) | |
1225 | sg_init_one(&sg[total_sgs++], argbuf, len); | |
1226 | ||
1227 | if (argpages) | |
1228 | total_sgs += sg_init_fuse_pages(&sg[total_sgs], | |
1229 | ap->pages, ap->descs, | |
1230 | ap->num_pages, | |
1231 | args[numargs - 1].size); | |
1232 | ||
1233 | if (len_used) | |
1234 | *len_used = len; | |
1235 | ||
1236 | return total_sgs; | |
1237 | } | |
1238 | ||
1239 | /* Add a request to a virtqueue and kick the device */ | |
1240 | static int virtio_fs_enqueue_req(struct virtio_fs_vq *fsvq, | |
a9bfd9dd | 1241 | struct fuse_req *req, bool in_flight) |
a62a8ef9 SH |
1242 | { |
1243 | /* requests need at least 4 elements */ | |
1244 | struct scatterlist *stack_sgs[6]; | |
1245 | struct scatterlist stack_sg[ARRAY_SIZE(stack_sgs)]; | |
1246 | struct scatterlist **sgs = stack_sgs; | |
1247 | struct scatterlist *sg = stack_sg; | |
1248 | struct virtqueue *vq; | |
1249 | struct fuse_args *args = req->args; | |
1250 | unsigned int argbuf_used = 0; | |
1251 | unsigned int out_sgs = 0; | |
1252 | unsigned int in_sgs = 0; | |
1253 | unsigned int total_sgs; | |
1254 | unsigned int i; | |
1255 | int ret; | |
1256 | bool notify; | |
5dbe190f | 1257 | struct fuse_pqueue *fpq; |
a62a8ef9 SH |
1258 | |
1259 | /* Does the sglist fit on the stack? */ | |
1260 | total_sgs = sg_count_fuse_req(req); | |
1261 | if (total_sgs > ARRAY_SIZE(stack_sgs)) { | |
1262 | sgs = kmalloc_array(total_sgs, sizeof(sgs[0]), GFP_ATOMIC); | |
1263 | sg = kmalloc_array(total_sgs, sizeof(sg[0]), GFP_ATOMIC); | |
1264 | if (!sgs || !sg) { | |
1265 | ret = -ENOMEM; | |
1266 | goto out; | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | /* Use a bounce buffer since stack args cannot be mapped */ | |
1271 | ret = copy_args_to_argbuf(req); | |
1272 | if (ret < 0) | |
1273 | goto out; | |
1274 | ||
1275 | /* Request elements */ | |
1276 | sg_init_one(&sg[out_sgs++], &req->in.h, sizeof(req->in.h)); | |
1277 | out_sgs += sg_init_fuse_args(&sg[out_sgs], req, | |
1278 | (struct fuse_arg *)args->in_args, | |
1279 | args->in_numargs, args->in_pages, | |
1280 | req->argbuf, &argbuf_used); | |
1281 | ||
1282 | /* Reply elements */ | |
1283 | if (test_bit(FR_ISREPLY, &req->flags)) { | |
1284 | sg_init_one(&sg[out_sgs + in_sgs++], | |
1285 | &req->out.h, sizeof(req->out.h)); | |
1286 | in_sgs += sg_init_fuse_args(&sg[out_sgs + in_sgs], req, | |
1287 | args->out_args, args->out_numargs, | |
1288 | args->out_pages, | |
1289 | req->argbuf + argbuf_used, NULL); | |
1290 | } | |
1291 | ||
1292 | WARN_ON(out_sgs + in_sgs != total_sgs); | |
1293 | ||
1294 | for (i = 0; i < total_sgs; i++) | |
1295 | sgs[i] = &sg[i]; | |
1296 | ||
1297 | spin_lock(&fsvq->lock); | |
1298 | ||
1299 | if (!fsvq->connected) { | |
1300 | spin_unlock(&fsvq->lock); | |
1301 | ret = -ENOTCONN; | |
1302 | goto out; | |
1303 | } | |
1304 | ||
1305 | vq = fsvq->vq; | |
1306 | ret = virtqueue_add_sgs(vq, sgs, out_sgs, in_sgs, req, GFP_ATOMIC); | |
1307 | if (ret < 0) { | |
1308 | spin_unlock(&fsvq->lock); | |
1309 | goto out; | |
1310 | } | |
1311 | ||
5dbe190f VG |
1312 | /* Request successfully sent. */ |
1313 | fpq = &fsvq->fud->pq; | |
1314 | spin_lock(&fpq->lock); | |
1315 | list_add_tail(&req->list, fpq->processing); | |
1316 | spin_unlock(&fpq->lock); | |
1317 | set_bit(FR_SENT, &req->flags); | |
1318 | /* matches barrier in request_wait_answer() */ | |
1319 | smp_mb__after_atomic(); | |
1320 | ||
a9bfd9dd VG |
1321 | if (!in_flight) |
1322 | inc_in_flight_req(fsvq); | |
a62a8ef9 SH |
1323 | notify = virtqueue_kick_prepare(vq); |
1324 | ||
1325 | spin_unlock(&fsvq->lock); | |
1326 | ||
1327 | if (notify) | |
1328 | virtqueue_notify(vq); | |
1329 | ||
1330 | out: | |
1331 | if (ret < 0 && req->argbuf) { | |
1332 | kfree(req->argbuf); | |
1333 | req->argbuf = NULL; | |
1334 | } | |
1335 | if (sgs != stack_sgs) { | |
1336 | kfree(sgs); | |
1337 | kfree(sg); | |
1338 | } | |
1339 | ||
1340 | return ret; | |
1341 | } | |
1342 | ||
1343 | static void virtio_fs_wake_pending_and_unlock(struct fuse_iqueue *fiq) | |
1344 | __releases(fiq->lock) | |
1345 | { | |
529395d2 | 1346 | unsigned int queue_id; |
a62a8ef9 | 1347 | struct virtio_fs *fs; |
a62a8ef9 | 1348 | struct fuse_req *req; |
51fecdd2 | 1349 | struct virtio_fs_vq *fsvq; |
a62a8ef9 SH |
1350 | int ret; |
1351 | ||
1352 | WARN_ON(list_empty(&fiq->pending)); | |
1353 | req = list_last_entry(&fiq->pending, struct fuse_req, list); | |
1354 | clear_bit(FR_PENDING, &req->flags); | |
1355 | list_del_init(&req->list); | |
1356 | WARN_ON(!list_empty(&fiq->pending)); | |
1357 | spin_unlock(&fiq->lock); | |
1358 | ||
1359 | fs = fiq->priv; | |
529395d2 | 1360 | queue_id = VQ_REQUEST + fs->mq_map[raw_smp_processor_id()]; |
a62a8ef9 | 1361 | |
529395d2 PJG |
1362 | pr_debug("%s: opcode %u unique %#llx nodeid %#llx in.len %u out.len %u queue_id %u\n", |
1363 | __func__, req->in.h.opcode, req->in.h.unique, | |
a62a8ef9 | 1364 | req->in.h.nodeid, req->in.h.len, |
529395d2 PJG |
1365 | fuse_len_args(req->args->out_numargs, req->args->out_args), |
1366 | queue_id); | |
a62a8ef9 | 1367 | |
51fecdd2 | 1368 | fsvq = &fs->vqs[queue_id]; |
a9bfd9dd | 1369 | ret = virtio_fs_enqueue_req(fsvq, req, false); |
a62a8ef9 SH |
1370 | if (ret < 0) { |
1371 | if (ret == -ENOMEM || ret == -ENOSPC) { | |
a9bfd9dd VG |
1372 | /* |
1373 | * Virtqueue full. Retry submission from worker | |
1374 | * context as we might be holding fc->bg_lock. | |
1375 | */ | |
1376 | spin_lock(&fsvq->lock); | |
1377 | list_add_tail(&req->list, &fsvq->queued_reqs); | |
1378 | inc_in_flight_req(fsvq); | |
1379 | schedule_delayed_work(&fsvq->dispatch_work, | |
1380 | msecs_to_jiffies(1)); | |
1381 | spin_unlock(&fsvq->lock); | |
1382 | return; | |
a62a8ef9 SH |
1383 | } |
1384 | req->out.h.error = ret; | |
1385 | pr_err("virtio-fs: virtio_fs_enqueue_req() failed %d\n", ret); | |
51fecdd2 VG |
1386 | |
1387 | /* Can't end request in submission context. Use a worker */ | |
1388 | spin_lock(&fsvq->lock); | |
1389 | list_add_tail(&req->list, &fsvq->end_reqs); | |
1390 | schedule_delayed_work(&fsvq->dispatch_work, 0); | |
1391 | spin_unlock(&fsvq->lock); | |
a62a8ef9 SH |
1392 | return; |
1393 | } | |
1394 | } | |
1395 | ||
00929447 | 1396 | static const struct fuse_iqueue_ops virtio_fs_fiq_ops = { |
a62a8ef9 SH |
1397 | .wake_forget_and_unlock = virtio_fs_wake_forget_and_unlock, |
1398 | .wake_interrupt_and_unlock = virtio_fs_wake_interrupt_and_unlock, | |
1399 | .wake_pending_and_unlock = virtio_fs_wake_pending_and_unlock, | |
1400 | .release = virtio_fs_fiq_release, | |
1401 | }; | |
1402 | ||
1dd53957 | 1403 | static inline void virtio_fs_ctx_set_defaults(struct fuse_fs_context *ctx) |
a62a8ef9 | 1404 | { |
1dd53957 VG |
1405 | ctx->rootmode = S_IFDIR; |
1406 | ctx->default_permissions = 1; | |
1407 | ctx->allow_other = 1; | |
1408 | ctx->max_read = UINT_MAX; | |
1409 | ctx->blksize = 512; | |
1410 | ctx->destroy = true; | |
1411 | ctx->no_control = true; | |
1412 | ctx->no_force_umount = true; | |
1413 | } | |
1414 | ||
1415 | static int virtio_fs_fill_super(struct super_block *sb, struct fs_context *fsc) | |
a62a8ef9 | 1416 | { |
fcee216b MR |
1417 | struct fuse_mount *fm = get_fuse_mount_super(sb); |
1418 | struct fuse_conn *fc = fm->fc; | |
a62a8ef9 | 1419 | struct virtio_fs *fs = fc->iq.priv; |
1dd53957 | 1420 | struct fuse_fs_context *ctx = fsc->fs_private; |
a62a8ef9 SH |
1421 | unsigned int i; |
1422 | int err; | |
a62a8ef9 | 1423 | |
1dd53957 | 1424 | virtio_fs_ctx_set_defaults(ctx); |
a62a8ef9 SH |
1425 | mutex_lock(&virtio_fs_mutex); |
1426 | ||
1427 | /* After holding mutex, make sure virtiofs device is still there. | |
1428 | * Though we are holding a reference to it, drive ->remove might | |
1429 | * still have cleaned up virtual queues. In that case bail out. | |
1430 | */ | |
1431 | err = -EINVAL; | |
1432 | if (list_empty(&fs->list)) { | |
1433 | pr_info("virtio-fs: tag <%s> not found\n", fs->tag); | |
1434 | goto err; | |
1435 | } | |
1436 | ||
1437 | err = -ENOMEM; | |
1438 | /* Allocate fuse_dev for hiprio and notification queues */ | |
7fd3abfa | 1439 | for (i = 0; i < fs->nvqs; i++) { |
a62a8ef9 SH |
1440 | struct virtio_fs_vq *fsvq = &fs->vqs[i]; |
1441 | ||
1442 | fsvq->fud = fuse_dev_alloc(); | |
1443 | if (!fsvq->fud) | |
1444 | goto err_free_fuse_devs; | |
1445 | } | |
1446 | ||
7fd3abfa | 1447 | /* virtiofs allocates and installs its own fuse devices */ |
1dd53957 | 1448 | ctx->fudptr = NULL; |
780b1b95 JX |
1449 | if (ctx->dax_mode != FUSE_DAX_NEVER) { |
1450 | if (ctx->dax_mode == FUSE_DAX_ALWAYS && !fs->dax_dev) { | |
3f9b9efd VG |
1451 | err = -EINVAL; |
1452 | pr_err("virtio-fs: dax can't be enabled as filesystem" | |
1453 | " device does not support it.\n"); | |
1454 | goto err_free_fuse_devs; | |
1455 | } | |
1dd53957 | 1456 | ctx->dax_dev = fs->dax_dev; |
3f9b9efd | 1457 | } |
1dd53957 | 1458 | err = fuse_fill_super_common(sb, ctx); |
a62a8ef9 SH |
1459 | if (err < 0) |
1460 | goto err_free_fuse_devs; | |
1461 | ||
a62a8ef9 SH |
1462 | for (i = 0; i < fs->nvqs; i++) { |
1463 | struct virtio_fs_vq *fsvq = &fs->vqs[i]; | |
1464 | ||
a62a8ef9 SH |
1465 | fuse_dev_install(fsvq->fud, fc); |
1466 | } | |
1467 | ||
1468 | /* Previous unmount will stop all queues. Start these again */ | |
1469 | virtio_fs_start_all_queues(fs); | |
fcee216b | 1470 | fuse_send_init(fm); |
a62a8ef9 SH |
1471 | mutex_unlock(&virtio_fs_mutex); |
1472 | return 0; | |
1473 | ||
1474 | err_free_fuse_devs: | |
1475 | virtio_fs_free_devs(fs); | |
1476 | err: | |
1477 | mutex_unlock(&virtio_fs_mutex); | |
1478 | return err; | |
1479 | } | |
1480 | ||
fcee216b | 1481 | static void virtio_fs_conn_destroy(struct fuse_mount *fm) |
a62a8ef9 | 1482 | { |
fcee216b MR |
1483 | struct fuse_conn *fc = fm->fc; |
1484 | struct virtio_fs *vfs = fc->iq.priv; | |
1485 | struct virtio_fs_vq *fsvq = &vfs->vqs[VQ_HIPRIO]; | |
a62a8ef9 | 1486 | |
fcee216b MR |
1487 | /* Stop dax worker. Soon evict_inodes() will be called which |
1488 | * will free all memory ranges belonging to all inodes. | |
9a752d18 VG |
1489 | */ |
1490 | if (IS_ENABLED(CONFIG_FUSE_DAX)) | |
1491 | fuse_dax_cancel_work(fc); | |
a62a8ef9 SH |
1492 | |
1493 | /* Stop forget queue. Soon destroy will be sent */ | |
1494 | spin_lock(&fsvq->lock); | |
1495 | fsvq->connected = false; | |
1496 | spin_unlock(&fsvq->lock); | |
1497 | virtio_fs_drain_all_queues(vfs); | |
1498 | ||
fcee216b | 1499 | fuse_conn_destroy(fm); |
a62a8ef9 | 1500 | |
fcee216b | 1501 | /* fuse_conn_destroy() must have sent destroy. Stop all queues |
a62a8ef9 SH |
1502 | * and drain one more time and free fuse devices. Freeing fuse |
1503 | * devices will drop their reference on fuse_conn and that in | |
1504 | * turn will drop its reference on virtio_fs object. | |
1505 | */ | |
1506 | virtio_fs_stop_all_queues(vfs); | |
1507 | virtio_fs_drain_all_queues(vfs); | |
1508 | virtio_fs_free_devs(vfs); | |
1509 | } | |
1510 | ||
fcee216b MR |
1511 | static void virtio_kill_sb(struct super_block *sb) |
1512 | { | |
1513 | struct fuse_mount *fm = get_fuse_mount_super(sb); | |
1514 | bool last; | |
1515 | ||
1516 | /* If mount failed, we can still be called without any fc */ | |
d534d31d | 1517 | if (sb->s_root) { |
fcee216b MR |
1518 | last = fuse_mount_remove(fm); |
1519 | if (last) | |
1520 | virtio_fs_conn_destroy(fm); | |
1521 | } | |
1522 | kill_anon_super(sb); | |
a27c061a | 1523 | fuse_mount_destroy(fm); |
fcee216b MR |
1524 | } |
1525 | ||
a62a8ef9 SH |
1526 | static int virtio_fs_test_super(struct super_block *sb, |
1527 | struct fs_context *fsc) | |
1528 | { | |
fcee216b MR |
1529 | struct fuse_mount *fsc_fm = fsc->s_fs_info; |
1530 | struct fuse_mount *sb_fm = get_fuse_mount_super(sb); | |
a62a8ef9 | 1531 | |
fcee216b | 1532 | return fsc_fm->fc->iq.priv == sb_fm->fc->iq.priv; |
a62a8ef9 SH |
1533 | } |
1534 | ||
a62a8ef9 SH |
1535 | static int virtio_fs_get_tree(struct fs_context *fsc) |
1536 | { | |
1537 | struct virtio_fs *fs; | |
1538 | struct super_block *sb; | |
a7f0d7aa | 1539 | struct fuse_conn *fc = NULL; |
fcee216b | 1540 | struct fuse_mount *fm; |
a7f0d7aa CK |
1541 | unsigned int virtqueue_size; |
1542 | int err = -EIO; | |
a62a8ef9 SH |
1543 | |
1544 | /* This gets a reference on virtio_fs object. This ptr gets installed | |
1545 | * in fc->iq->priv. Once fuse_conn is going away, it calls ->put() | |
1546 | * to drop the reference to this object. | |
1547 | */ | |
1548 | fs = virtio_fs_find_instance(fsc->source); | |
1549 | if (!fs) { | |
1550 | pr_info("virtio-fs: tag <%s> not found\n", fsc->source); | |
1551 | return -EINVAL; | |
1552 | } | |
1553 | ||
a7f0d7aa CK |
1554 | virtqueue_size = virtqueue_get_vring_size(fs->vqs[VQ_REQUEST].vq); |
1555 | if (WARN_ON(virtqueue_size <= FUSE_HEADER_OVERHEAD)) | |
1556 | goto out_err; | |
1557 | ||
833c5a42 | 1558 | err = -ENOMEM; |
a62a8ef9 | 1559 | fc = kzalloc(sizeof(struct fuse_conn), GFP_KERNEL); |
833c5a42 MS |
1560 | if (!fc) |
1561 | goto out_err; | |
a62a8ef9 | 1562 | |
fcee216b | 1563 | fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL); |
833c5a42 MS |
1564 | if (!fm) |
1565 | goto out_err; | |
fcee216b | 1566 | |
0a7419c6 | 1567 | fuse_conn_init(fc, fm, fsc->user_ns, &virtio_fs_fiq_ops, fs); |
a62a8ef9 SH |
1568 | fc->release = fuse_free_conn; |
1569 | fc->delete_stale = true; | |
bf109c64 | 1570 | fc->auto_submounts = true; |
2d82ab25 | 1571 | fc->sync_fs = true; |
a62a8ef9 | 1572 | |
a7f0d7aa CK |
1573 | /* Tell FUSE to split requests that exceed the virtqueue's size */ |
1574 | fc->max_pages_limit = min_t(unsigned int, fc->max_pages_limit, | |
1575 | virtqueue_size - FUSE_HEADER_OVERHEAD); | |
1576 | ||
fcee216b | 1577 | fsc->s_fs_info = fm; |
b19d3d00 | 1578 | sb = sget_fc(fsc, virtio_fs_test_super, set_anon_super_fc); |
c191cd07 MS |
1579 | if (fsc->s_fs_info) |
1580 | fuse_mount_destroy(fm); | |
a62a8ef9 SH |
1581 | if (IS_ERR(sb)) |
1582 | return PTR_ERR(sb); | |
1583 | ||
1584 | if (!sb->s_root) { | |
1dd53957 | 1585 | err = virtio_fs_fill_super(sb, fsc); |
a62a8ef9 SH |
1586 | if (err) { |
1587 | deactivate_locked_super(sb); | |
1588 | return err; | |
1589 | } | |
1590 | ||
1591 | sb->s_flags |= SB_ACTIVE; | |
1592 | } | |
1593 | ||
1594 | WARN_ON(fsc->root); | |
1595 | fsc->root = dget(sb->s_root); | |
1596 | return 0; | |
833c5a42 MS |
1597 | |
1598 | out_err: | |
1599 | kfree(fc); | |
1600 | mutex_lock(&virtio_fs_mutex); | |
1601 | virtio_fs_put(fs); | |
1602 | mutex_unlock(&virtio_fs_mutex); | |
1603 | return err; | |
a62a8ef9 SH |
1604 | } |
1605 | ||
1606 | static const struct fs_context_operations virtio_fs_context_ops = { | |
84c21507 | 1607 | .free = virtio_fs_free_fsc, |
1dd53957 | 1608 | .parse_param = virtio_fs_parse_param, |
a62a8ef9 SH |
1609 | .get_tree = virtio_fs_get_tree, |
1610 | }; | |
1611 | ||
1612 | static int virtio_fs_init_fs_context(struct fs_context *fsc) | |
1613 | { | |
1dd53957 VG |
1614 | struct fuse_fs_context *ctx; |
1615 | ||
fe0a7bd8 GK |
1616 | if (fsc->purpose == FS_CONTEXT_FOR_SUBMOUNT) |
1617 | return fuse_init_fs_context_submount(fsc); | |
1618 | ||
1dd53957 VG |
1619 | ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL); |
1620 | if (!ctx) | |
1621 | return -ENOMEM; | |
1622 | fsc->fs_private = ctx; | |
a62a8ef9 SH |
1623 | fsc->ops = &virtio_fs_context_ops; |
1624 | return 0; | |
1625 | } | |
1626 | ||
1627 | static struct file_system_type virtio_fs_type = { | |
1628 | .owner = THIS_MODULE, | |
1629 | .name = "virtiofs", | |
1630 | .init_fs_context = virtio_fs_init_fs_context, | |
1631 | .kill_sb = virtio_kill_sb, | |
1632 | }; | |
1633 | ||
9086b2d9 SH |
1634 | static int virtio_fs_uevent(const struct kobject *kobj, struct kobj_uevent_env *env) |
1635 | { | |
1636 | const struct virtio_fs *fs = container_of(kobj, struct virtio_fs, kobj); | |
1637 | ||
1638 | add_uevent_var(env, "TAG=%s", fs->tag); | |
1639 | return 0; | |
1640 | } | |
1641 | ||
1642 | static const struct kset_uevent_ops virtio_fs_uevent_ops = { | |
1643 | .uevent = virtio_fs_uevent, | |
1644 | }; | |
1645 | ||
a8f62f50 SH |
1646 | static int __init virtio_fs_sysfs_init(void) |
1647 | { | |
9086b2d9 SH |
1648 | virtio_fs_kset = kset_create_and_add("virtiofs", &virtio_fs_uevent_ops, |
1649 | fs_kobj); | |
a8f62f50 SH |
1650 | if (!virtio_fs_kset) |
1651 | return -ENOMEM; | |
1652 | return 0; | |
1653 | } | |
1654 | ||
d30ff898 | 1655 | static void virtio_fs_sysfs_exit(void) |
a8f62f50 SH |
1656 | { |
1657 | kset_unregister(virtio_fs_kset); | |
1658 | virtio_fs_kset = NULL; | |
1659 | } | |
1660 | ||
a62a8ef9 SH |
1661 | static int __init virtio_fs_init(void) |
1662 | { | |
1663 | int ret; | |
1664 | ||
a8f62f50 | 1665 | ret = virtio_fs_sysfs_init(); |
a62a8ef9 SH |
1666 | if (ret < 0) |
1667 | return ret; | |
1668 | ||
a8f62f50 SH |
1669 | ret = register_virtio_driver(&virtio_fs_driver); |
1670 | if (ret < 0) | |
1671 | goto sysfs_exit; | |
1672 | ||
a62a8ef9 | 1673 | ret = register_filesystem(&virtio_fs_type); |
a8f62f50 SH |
1674 | if (ret < 0) |
1675 | goto unregister_virtio_driver; | |
a62a8ef9 SH |
1676 | |
1677 | return 0; | |
a8f62f50 SH |
1678 | |
1679 | unregister_virtio_driver: | |
1680 | unregister_virtio_driver(&virtio_fs_driver); | |
1681 | sysfs_exit: | |
1682 | virtio_fs_sysfs_exit(); | |
1683 | return ret; | |
a62a8ef9 SH |
1684 | } |
1685 | module_init(virtio_fs_init); | |
1686 | ||
1687 | static void __exit virtio_fs_exit(void) | |
1688 | { | |
1689 | unregister_filesystem(&virtio_fs_type); | |
1690 | unregister_virtio_driver(&virtio_fs_driver); | |
a8f62f50 | 1691 | virtio_fs_sysfs_exit(); |
a62a8ef9 SH |
1692 | } |
1693 | module_exit(virtio_fs_exit); | |
1694 | ||
1695 | MODULE_AUTHOR("Stefan Hajnoczi <stefanha@redhat.com>"); | |
1696 | MODULE_DESCRIPTION("Virtio Filesystem"); | |
1697 | MODULE_LICENSE("GPL"); | |
1698 | MODULE_ALIAS_FS(KBUILD_MODNAME); | |
1699 | MODULE_DEVICE_TABLE(virtio, id_table); |