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> | |
10 | #include <linux/pfn_t.h> | |
dc90f084 | 11 | #include <linux/memremap.h> |
a62a8ef9 SH |
12 | #include <linux/module.h> |
13 | #include <linux/virtio.h> | |
14 | #include <linux/virtio_fs.h> | |
15 | #include <linux/delay.h> | |
16 | #include <linux/fs_context.h> | |
1dd53957 | 17 | #include <linux/fs_parser.h> |
a62a8ef9 | 18 | #include <linux/highmem.h> |
562ce828 | 19 | #include <linux/cleanup.h> |
22f3787e | 20 | #include <linux/uio.h> |
a62a8ef9 SH |
21 | #include "fuse_i.h" |
22 | ||
a7f0d7aa CK |
23 | /* Used to help calculate the FUSE connection's max_pages limit for a request's |
24 | * size. Parts of the struct fuse_req are sliced into scattergather lists in | |
25 | * addition to the pages used, so this can help account for that overhead. | |
26 | */ | |
27 | #define FUSE_HEADER_OVERHEAD 4 | |
28 | ||
a62a8ef9 SH |
29 | /* List of virtio-fs device instances and a lock for the list. Also provides |
30 | * mutual exclusion in device removal and mounting path | |
31 | */ | |
32 | static DEFINE_MUTEX(virtio_fs_mutex); | |
33 | static LIST_HEAD(virtio_fs_instances); | |
34 | ||
35 | enum { | |
36 | VQ_HIPRIO, | |
37 | VQ_REQUEST | |
38 | }; | |
39 | ||
b43b7e81 VG |
40 | #define VQ_NAME_LEN 24 |
41 | ||
a62a8ef9 SH |
42 | /* Per-virtqueue state */ |
43 | struct virtio_fs_vq { | |
44 | spinlock_t lock; | |
45 | struct virtqueue *vq; /* protected by ->lock */ | |
46 | struct work_struct done_work; | |
47 | struct list_head queued_reqs; | |
51fecdd2 | 48 | struct list_head end_reqs; /* End these requests */ |
a62a8ef9 SH |
49 | struct delayed_work dispatch_work; |
50 | struct fuse_dev *fud; | |
51 | bool connected; | |
52 | long in_flight; | |
724c15a4 | 53 | struct completion in_flight_zero; /* No inflight requests */ |
b43b7e81 | 54 | char name[VQ_NAME_LEN]; |
a62a8ef9 SH |
55 | } ____cacheline_aligned_in_smp; |
56 | ||
57 | /* A virtio-fs device instance */ | |
58 | struct virtio_fs { | |
59 | struct kref refcount; | |
60 | struct list_head list; /* on virtio_fs_instances */ | |
61 | char *tag; | |
62 | struct virtio_fs_vq *vqs; | |
63 | unsigned int nvqs; /* number of virtqueues */ | |
64 | unsigned int num_request_queues; /* number of request queues */ | |
22f3787e SH |
65 | struct dax_device *dax_dev; |
66 | ||
67 | /* DAX memory window where file contents are mapped */ | |
68 | void *window_kaddr; | |
69 | phys_addr_t window_phys_addr; | |
70 | size_t window_len; | |
a62a8ef9 SH |
71 | }; |
72 | ||
1efcf39e | 73 | struct virtio_fs_forget_req { |
a62a8ef9 SH |
74 | struct fuse_in_header ih; |
75 | struct fuse_forget_in arg; | |
1efcf39e VG |
76 | }; |
77 | ||
78 | struct virtio_fs_forget { | |
a62a8ef9 SH |
79 | /* This request can be temporarily queued on virt queue */ |
80 | struct list_head list; | |
1efcf39e | 81 | struct virtio_fs_forget_req req; |
a62a8ef9 SH |
82 | }; |
83 | ||
bb737bbe VG |
84 | struct virtio_fs_req_work { |
85 | struct fuse_req *req; | |
86 | struct virtio_fs_vq *fsvq; | |
87 | struct work_struct done_work; | |
88 | }; | |
89 | ||
a9bfd9dd VG |
90 | static int virtio_fs_enqueue_req(struct virtio_fs_vq *fsvq, |
91 | struct fuse_req *req, bool in_flight); | |
92 | ||
780b1b95 JX |
93 | static const struct constant_table dax_param_enums[] = { |
94 | {"always", FUSE_DAX_ALWAYS }, | |
95 | {"never", FUSE_DAX_NEVER }, | |
96 | {"inode", FUSE_DAX_INODE_USER }, | |
97 | {} | |
98 | }; | |
99 | ||
1dd53957 VG |
100 | enum { |
101 | OPT_DAX, | |
780b1b95 | 102 | OPT_DAX_ENUM, |
1dd53957 VG |
103 | }; |
104 | ||
105 | static const struct fs_parameter_spec virtio_fs_parameters[] = { | |
106 | fsparam_flag("dax", OPT_DAX), | |
780b1b95 | 107 | fsparam_enum("dax", OPT_DAX_ENUM, dax_param_enums), |
1dd53957 VG |
108 | {} |
109 | }; | |
110 | ||
84c21507 | 111 | static int virtio_fs_parse_param(struct fs_context *fsc, |
1dd53957 VG |
112 | struct fs_parameter *param) |
113 | { | |
114 | struct fs_parse_result result; | |
84c21507 | 115 | struct fuse_fs_context *ctx = fsc->fs_private; |
1dd53957 VG |
116 | int opt; |
117 | ||
84c21507 | 118 | opt = fs_parse(fsc, virtio_fs_parameters, param, &result); |
1dd53957 VG |
119 | if (opt < 0) |
120 | return opt; | |
121 | ||
122 | switch (opt) { | |
123 | case OPT_DAX: | |
780b1b95 JX |
124 | ctx->dax_mode = FUSE_DAX_ALWAYS; |
125 | break; | |
126 | case OPT_DAX_ENUM: | |
127 | ctx->dax_mode = result.uint_32; | |
1dd53957 VG |
128 | break; |
129 | default: | |
130 | return -EINVAL; | |
131 | } | |
132 | ||
133 | return 0; | |
134 | } | |
135 | ||
84c21507 | 136 | static void virtio_fs_free_fsc(struct fs_context *fsc) |
1dd53957 | 137 | { |
84c21507 | 138 | struct fuse_fs_context *ctx = fsc->fs_private; |
1dd53957 VG |
139 | |
140 | kfree(ctx); | |
141 | } | |
142 | ||
a62a8ef9 SH |
143 | static inline struct virtio_fs_vq *vq_to_fsvq(struct virtqueue *vq) |
144 | { | |
145 | struct virtio_fs *fs = vq->vdev->priv; | |
146 | ||
147 | return &fs->vqs[vq->index]; | |
148 | } | |
149 | ||
c17ea009 VG |
150 | /* Should be called with fsvq->lock held. */ |
151 | static inline void inc_in_flight_req(struct virtio_fs_vq *fsvq) | |
152 | { | |
153 | fsvq->in_flight++; | |
154 | } | |
155 | ||
156 | /* Should be called with fsvq->lock held. */ | |
157 | static inline void dec_in_flight_req(struct virtio_fs_vq *fsvq) | |
158 | { | |
159 | WARN_ON(fsvq->in_flight <= 0); | |
160 | fsvq->in_flight--; | |
724c15a4 VG |
161 | if (!fsvq->in_flight) |
162 | complete(&fsvq->in_flight_zero); | |
c17ea009 VG |
163 | } |
164 | ||
a62a8ef9 SH |
165 | static void release_virtio_fs_obj(struct kref *ref) |
166 | { | |
167 | struct virtio_fs *vfs = container_of(ref, struct virtio_fs, refcount); | |
168 | ||
169 | kfree(vfs->vqs); | |
170 | kfree(vfs); | |
171 | } | |
172 | ||
173 | /* Make sure virtiofs_mutex is held */ | |
174 | static void virtio_fs_put(struct virtio_fs *fs) | |
175 | { | |
176 | kref_put(&fs->refcount, release_virtio_fs_obj); | |
177 | } | |
178 | ||
179 | static void virtio_fs_fiq_release(struct fuse_iqueue *fiq) | |
180 | { | |
181 | struct virtio_fs *vfs = fiq->priv; | |
182 | ||
183 | mutex_lock(&virtio_fs_mutex); | |
184 | virtio_fs_put(vfs); | |
185 | mutex_unlock(&virtio_fs_mutex); | |
186 | } | |
187 | ||
188 | static void virtio_fs_drain_queue(struct virtio_fs_vq *fsvq) | |
189 | { | |
190 | WARN_ON(fsvq->in_flight < 0); | |
191 | ||
192 | /* Wait for in flight requests to finish.*/ | |
724c15a4 VG |
193 | spin_lock(&fsvq->lock); |
194 | if (fsvq->in_flight) { | |
195 | /* We are holding virtio_fs_mutex. There should not be any | |
196 | * waiters waiting for completion. | |
197 | */ | |
198 | reinit_completion(&fsvq->in_flight_zero); | |
199 | spin_unlock(&fsvq->lock); | |
200 | wait_for_completion(&fsvq->in_flight_zero); | |
201 | } else { | |
a62a8ef9 | 202 | spin_unlock(&fsvq->lock); |
a62a8ef9 SH |
203 | } |
204 | ||
205 | flush_work(&fsvq->done_work); | |
206 | flush_delayed_work(&fsvq->dispatch_work); | |
207 | } | |
208 | ||
724c15a4 | 209 | static void virtio_fs_drain_all_queues_locked(struct virtio_fs *fs) |
a62a8ef9 SH |
210 | { |
211 | struct virtio_fs_vq *fsvq; | |
212 | int i; | |
213 | ||
214 | for (i = 0; i < fs->nvqs; i++) { | |
215 | fsvq = &fs->vqs[i]; | |
a62a8ef9 SH |
216 | virtio_fs_drain_queue(fsvq); |
217 | } | |
218 | } | |
219 | ||
724c15a4 VG |
220 | static void virtio_fs_drain_all_queues(struct virtio_fs *fs) |
221 | { | |
222 | /* Provides mutual exclusion between ->remove and ->kill_sb | |
223 | * paths. We don't want both of these draining queue at the | |
224 | * same time. Current completion logic reinits completion | |
225 | * and that means there should not be any other thread | |
226 | * doing reinit or waiting for completion already. | |
227 | */ | |
228 | mutex_lock(&virtio_fs_mutex); | |
229 | virtio_fs_drain_all_queues_locked(fs); | |
230 | mutex_unlock(&virtio_fs_mutex); | |
231 | } | |
232 | ||
a62a8ef9 SH |
233 | static void virtio_fs_start_all_queues(struct virtio_fs *fs) |
234 | { | |
235 | struct virtio_fs_vq *fsvq; | |
236 | int i; | |
237 | ||
238 | for (i = 0; i < fs->nvqs; i++) { | |
239 | fsvq = &fs->vqs[i]; | |
240 | spin_lock(&fsvq->lock); | |
241 | fsvq->connected = true; | |
242 | spin_unlock(&fsvq->lock); | |
243 | } | |
244 | } | |
245 | ||
246 | /* Add a new instance to the list or return -EEXIST if tag name exists*/ | |
247 | static int virtio_fs_add_instance(struct virtio_fs *fs) | |
248 | { | |
249 | struct virtio_fs *fs2; | |
250 | bool duplicate = false; | |
251 | ||
252 | mutex_lock(&virtio_fs_mutex); | |
253 | ||
254 | list_for_each_entry(fs2, &virtio_fs_instances, list) { | |
255 | if (strcmp(fs->tag, fs2->tag) == 0) | |
256 | duplicate = true; | |
257 | } | |
258 | ||
259 | if (!duplicate) | |
260 | list_add_tail(&fs->list, &virtio_fs_instances); | |
261 | ||
262 | mutex_unlock(&virtio_fs_mutex); | |
263 | ||
264 | if (duplicate) | |
265 | return -EEXIST; | |
266 | return 0; | |
267 | } | |
268 | ||
269 | /* Return the virtio_fs with a given tag, or NULL */ | |
270 | static struct virtio_fs *virtio_fs_find_instance(const char *tag) | |
271 | { | |
272 | struct virtio_fs *fs; | |
273 | ||
274 | mutex_lock(&virtio_fs_mutex); | |
275 | ||
276 | list_for_each_entry(fs, &virtio_fs_instances, list) { | |
277 | if (strcmp(fs->tag, tag) == 0) { | |
278 | kref_get(&fs->refcount); | |
279 | goto found; | |
280 | } | |
281 | } | |
282 | ||
283 | fs = NULL; /* not found */ | |
284 | ||
285 | found: | |
286 | mutex_unlock(&virtio_fs_mutex); | |
287 | ||
288 | return fs; | |
289 | } | |
290 | ||
291 | static void virtio_fs_free_devs(struct virtio_fs *fs) | |
292 | { | |
293 | unsigned int i; | |
294 | ||
295 | for (i = 0; i < fs->nvqs; i++) { | |
296 | struct virtio_fs_vq *fsvq = &fs->vqs[i]; | |
297 | ||
298 | if (!fsvq->fud) | |
299 | continue; | |
300 | ||
301 | fuse_dev_free(fsvq->fud); | |
302 | fsvq->fud = NULL; | |
303 | } | |
304 | } | |
305 | ||
306 | /* Read filesystem name from virtio config into fs->tag (must kfree()). */ | |
307 | static int virtio_fs_read_tag(struct virtio_device *vdev, struct virtio_fs *fs) | |
308 | { | |
309 | char tag_buf[sizeof_field(struct virtio_fs_config, tag)]; | |
310 | char *end; | |
311 | size_t len; | |
312 | ||
313 | virtio_cread_bytes(vdev, offsetof(struct virtio_fs_config, tag), | |
314 | &tag_buf, sizeof(tag_buf)); | |
315 | end = memchr(tag_buf, '\0', sizeof(tag_buf)); | |
316 | if (end == tag_buf) | |
317 | return -EINVAL; /* empty tag */ | |
318 | if (!end) | |
319 | end = &tag_buf[sizeof(tag_buf)]; | |
320 | ||
321 | len = end - tag_buf; | |
322 | fs->tag = devm_kmalloc(&vdev->dev, len + 1, GFP_KERNEL); | |
323 | if (!fs->tag) | |
324 | return -ENOMEM; | |
325 | memcpy(fs->tag, tag_buf, len); | |
326 | fs->tag[len] = '\0'; | |
327 | return 0; | |
328 | } | |
329 | ||
330 | /* Work function for hiprio completion */ | |
331 | static void virtio_fs_hiprio_done_work(struct work_struct *work) | |
332 | { | |
333 | struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq, | |
334 | done_work); | |
335 | struct virtqueue *vq = fsvq->vq; | |
336 | ||
337 | /* Free completed FUSE_FORGET requests */ | |
338 | spin_lock(&fsvq->lock); | |
339 | do { | |
340 | unsigned int len; | |
341 | void *req; | |
342 | ||
343 | virtqueue_disable_cb(vq); | |
344 | ||
345 | while ((req = virtqueue_get_buf(vq, &len)) != NULL) { | |
346 | kfree(req); | |
c17ea009 | 347 | dec_in_flight_req(fsvq); |
a62a8ef9 SH |
348 | } |
349 | } while (!virtqueue_enable_cb(vq) && likely(!virtqueue_is_broken(vq))); | |
350 | spin_unlock(&fsvq->lock); | |
351 | } | |
352 | ||
51fecdd2 | 353 | static void virtio_fs_request_dispatch_work(struct work_struct *work) |
a62a8ef9 | 354 | { |
51fecdd2 VG |
355 | struct fuse_req *req; |
356 | struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq, | |
357 | dispatch_work.work); | |
a9bfd9dd | 358 | int ret; |
51fecdd2 VG |
359 | |
360 | pr_debug("virtio-fs: worker %s called.\n", __func__); | |
361 | while (1) { | |
362 | spin_lock(&fsvq->lock); | |
363 | req = list_first_entry_or_null(&fsvq->end_reqs, struct fuse_req, | |
364 | list); | |
365 | if (!req) { | |
366 | spin_unlock(&fsvq->lock); | |
a9bfd9dd | 367 | break; |
51fecdd2 VG |
368 | } |
369 | ||
370 | list_del_init(&req->list); | |
371 | spin_unlock(&fsvq->lock); | |
8f622e94 | 372 | fuse_request_end(req); |
51fecdd2 | 373 | } |
a9bfd9dd VG |
374 | |
375 | /* Dispatch pending requests */ | |
376 | while (1) { | |
377 | spin_lock(&fsvq->lock); | |
378 | req = list_first_entry_or_null(&fsvq->queued_reqs, | |
379 | struct fuse_req, list); | |
380 | if (!req) { | |
381 | spin_unlock(&fsvq->lock); | |
382 | return; | |
383 | } | |
384 | list_del_init(&req->list); | |
385 | spin_unlock(&fsvq->lock); | |
386 | ||
387 | ret = virtio_fs_enqueue_req(fsvq, req, true); | |
388 | if (ret < 0) { | |
389 | if (ret == -ENOMEM || ret == -ENOSPC) { | |
390 | spin_lock(&fsvq->lock); | |
391 | list_add_tail(&req->list, &fsvq->queued_reqs); | |
392 | schedule_delayed_work(&fsvq->dispatch_work, | |
393 | msecs_to_jiffies(1)); | |
394 | spin_unlock(&fsvq->lock); | |
395 | return; | |
396 | } | |
397 | req->out.h.error = ret; | |
398 | spin_lock(&fsvq->lock); | |
399 | dec_in_flight_req(fsvq); | |
400 | spin_unlock(&fsvq->lock); | |
401 | pr_err("virtio-fs: virtio_fs_enqueue_req() failed %d\n", | |
402 | ret); | |
8f622e94 | 403 | fuse_request_end(req); |
a9bfd9dd VG |
404 | } |
405 | } | |
a62a8ef9 SH |
406 | } |
407 | ||
58ada94f VG |
408 | /* |
409 | * Returns 1 if queue is full and sender should wait a bit before sending | |
410 | * next request, 0 otherwise. | |
411 | */ | |
412 | static int send_forget_request(struct virtio_fs_vq *fsvq, | |
413 | struct virtio_fs_forget *forget, | |
414 | bool in_flight) | |
415 | { | |
416 | struct scatterlist sg; | |
417 | struct virtqueue *vq; | |
418 | int ret = 0; | |
419 | bool notify; | |
1efcf39e | 420 | struct virtio_fs_forget_req *req = &forget->req; |
58ada94f VG |
421 | |
422 | spin_lock(&fsvq->lock); | |
423 | if (!fsvq->connected) { | |
424 | if (in_flight) | |
425 | dec_in_flight_req(fsvq); | |
426 | kfree(forget); | |
427 | goto out; | |
428 | } | |
429 | ||
1efcf39e | 430 | sg_init_one(&sg, req, sizeof(*req)); |
58ada94f VG |
431 | vq = fsvq->vq; |
432 | dev_dbg(&vq->vdev->dev, "%s\n", __func__); | |
433 | ||
434 | ret = virtqueue_add_outbuf(vq, &sg, 1, forget, GFP_ATOMIC); | |
435 | if (ret < 0) { | |
436 | if (ret == -ENOMEM || ret == -ENOSPC) { | |
437 | pr_debug("virtio-fs: Could not queue FORGET: err=%d. Will try later\n", | |
438 | ret); | |
439 | list_add_tail(&forget->list, &fsvq->queued_reqs); | |
440 | schedule_delayed_work(&fsvq->dispatch_work, | |
441 | msecs_to_jiffies(1)); | |
442 | if (!in_flight) | |
443 | inc_in_flight_req(fsvq); | |
444 | /* Queue is full */ | |
445 | ret = 1; | |
446 | } else { | |
447 | pr_debug("virtio-fs: Could not queue FORGET: err=%d. Dropping it.\n", | |
448 | ret); | |
449 | kfree(forget); | |
450 | if (in_flight) | |
451 | dec_in_flight_req(fsvq); | |
452 | } | |
453 | goto out; | |
454 | } | |
455 | ||
456 | if (!in_flight) | |
457 | inc_in_flight_req(fsvq); | |
458 | notify = virtqueue_kick_prepare(vq); | |
459 | spin_unlock(&fsvq->lock); | |
460 | ||
461 | if (notify) | |
462 | virtqueue_notify(vq); | |
463 | return ret; | |
464 | out: | |
465 | spin_unlock(&fsvq->lock); | |
466 | return ret; | |
467 | } | |
468 | ||
a62a8ef9 SH |
469 | static void virtio_fs_hiprio_dispatch_work(struct work_struct *work) |
470 | { | |
471 | struct virtio_fs_forget *forget; | |
472 | struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq, | |
473 | dispatch_work.work); | |
a62a8ef9 SH |
474 | pr_debug("virtio-fs: worker %s called.\n", __func__); |
475 | while (1) { | |
476 | spin_lock(&fsvq->lock); | |
477 | forget = list_first_entry_or_null(&fsvq->queued_reqs, | |
478 | struct virtio_fs_forget, list); | |
479 | if (!forget) { | |
480 | spin_unlock(&fsvq->lock); | |
481 | return; | |
482 | } | |
483 | ||
484 | list_del(&forget->list); | |
a62a8ef9 | 485 | spin_unlock(&fsvq->lock); |
58ada94f VG |
486 | if (send_forget_request(fsvq, forget, true)) |
487 | return; | |
a62a8ef9 SH |
488 | } |
489 | } | |
490 | ||
491 | /* Allocate and copy args into req->argbuf */ | |
492 | static int copy_args_to_argbuf(struct fuse_req *req) | |
493 | { | |
494 | struct fuse_args *args = req->args; | |
495 | unsigned int offset = 0; | |
496 | unsigned int num_in; | |
497 | unsigned int num_out; | |
498 | unsigned int len; | |
499 | unsigned int i; | |
500 | ||
501 | num_in = args->in_numargs - args->in_pages; | |
502 | num_out = args->out_numargs - args->out_pages; | |
503 | len = fuse_len_args(num_in, (struct fuse_arg *) args->in_args) + | |
504 | fuse_len_args(num_out, args->out_args); | |
505 | ||
506 | req->argbuf = kmalloc(len, GFP_ATOMIC); | |
507 | if (!req->argbuf) | |
508 | return -ENOMEM; | |
509 | ||
510 | for (i = 0; i < num_in; i++) { | |
511 | memcpy(req->argbuf + offset, | |
512 | args->in_args[i].value, | |
513 | args->in_args[i].size); | |
514 | offset += args->in_args[i].size; | |
515 | } | |
516 | ||
517 | return 0; | |
518 | } | |
519 | ||
520 | /* Copy args out of and free req->argbuf */ | |
521 | static void copy_args_from_argbuf(struct fuse_args *args, struct fuse_req *req) | |
522 | { | |
523 | unsigned int remaining; | |
524 | unsigned int offset; | |
525 | unsigned int num_in; | |
526 | unsigned int num_out; | |
527 | unsigned int i; | |
528 | ||
529 | remaining = req->out.h.len - sizeof(req->out.h); | |
530 | num_in = args->in_numargs - args->in_pages; | |
531 | num_out = args->out_numargs - args->out_pages; | |
532 | offset = fuse_len_args(num_in, (struct fuse_arg *)args->in_args); | |
533 | ||
534 | for (i = 0; i < num_out; i++) { | |
535 | unsigned int argsize = args->out_args[i].size; | |
536 | ||
537 | if (args->out_argvar && | |
538 | i == args->out_numargs - 1 && | |
539 | argsize > remaining) { | |
540 | argsize = remaining; | |
541 | } | |
542 | ||
543 | memcpy(args->out_args[i].value, req->argbuf + offset, argsize); | |
544 | offset += argsize; | |
545 | ||
546 | if (i != args->out_numargs - 1) | |
547 | remaining -= argsize; | |
548 | } | |
549 | ||
550 | /* Store the actual size of the variable-length arg */ | |
551 | if (args->out_argvar) | |
552 | args->out_args[args->out_numargs - 1].size = remaining; | |
553 | ||
554 | kfree(req->argbuf); | |
555 | req->argbuf = NULL; | |
556 | } | |
557 | ||
558 | /* Work function for request completion */ | |
bb737bbe VG |
559 | static void virtio_fs_request_complete(struct fuse_req *req, |
560 | struct virtio_fs_vq *fsvq) | |
561 | { | |
562 | struct fuse_pqueue *fpq = &fsvq->fud->pq; | |
bb737bbe VG |
563 | struct fuse_args *args; |
564 | struct fuse_args_pages *ap; | |
565 | unsigned int len, i, thislen; | |
566 | struct page *page; | |
567 | ||
568 | /* | |
569 | * TODO verify that server properly follows FUSE protocol | |
570 | * (oh.uniq, oh.len) | |
571 | */ | |
572 | args = req->args; | |
573 | copy_args_from_argbuf(args, req); | |
574 | ||
575 | if (args->out_pages && args->page_zeroing) { | |
576 | len = args->out_args[args->out_numargs - 1].size; | |
577 | ap = container_of(args, typeof(*ap), args); | |
578 | for (i = 0; i < ap->num_pages; i++) { | |
579 | thislen = ap->descs[i].length; | |
580 | if (len < thislen) { | |
581 | WARN_ON(ap->descs[i].offset); | |
582 | page = ap->pages[i]; | |
583 | zero_user_segment(page, len, thislen); | |
584 | len = 0; | |
585 | } else { | |
586 | len -= thislen; | |
587 | } | |
588 | } | |
589 | } | |
590 | ||
591 | spin_lock(&fpq->lock); | |
592 | clear_bit(FR_SENT, &req->flags); | |
593 | spin_unlock(&fpq->lock); | |
594 | ||
8f622e94 | 595 | fuse_request_end(req); |
bb737bbe VG |
596 | spin_lock(&fsvq->lock); |
597 | dec_in_flight_req(fsvq); | |
598 | spin_unlock(&fsvq->lock); | |
599 | } | |
600 | ||
601 | static void virtio_fs_complete_req_work(struct work_struct *work) | |
602 | { | |
603 | struct virtio_fs_req_work *w = | |
604 | container_of(work, typeof(*w), done_work); | |
605 | ||
606 | virtio_fs_request_complete(w->req, w->fsvq); | |
607 | kfree(w); | |
608 | } | |
609 | ||
a62a8ef9 SH |
610 | static void virtio_fs_requests_done_work(struct work_struct *work) |
611 | { | |
612 | struct virtio_fs_vq *fsvq = container_of(work, struct virtio_fs_vq, | |
613 | done_work); | |
614 | struct fuse_pqueue *fpq = &fsvq->fud->pq; | |
a62a8ef9 SH |
615 | struct virtqueue *vq = fsvq->vq; |
616 | struct fuse_req *req; | |
a62a8ef9 | 617 | struct fuse_req *next; |
bb737bbe | 618 | unsigned int len; |
a62a8ef9 SH |
619 | LIST_HEAD(reqs); |
620 | ||
621 | /* Collect completed requests off the virtqueue */ | |
622 | spin_lock(&fsvq->lock); | |
623 | do { | |
624 | virtqueue_disable_cb(vq); | |
625 | ||
626 | while ((req = virtqueue_get_buf(vq, &len)) != NULL) { | |
627 | spin_lock(&fpq->lock); | |
628 | list_move_tail(&req->list, &reqs); | |
629 | spin_unlock(&fpq->lock); | |
630 | } | |
631 | } while (!virtqueue_enable_cb(vq) && likely(!virtqueue_is_broken(vq))); | |
632 | spin_unlock(&fsvq->lock); | |
633 | ||
634 | /* End requests */ | |
635 | list_for_each_entry_safe(req, next, &reqs, list) { | |
a62a8ef9 | 636 | list_del_init(&req->list); |
a62a8ef9 | 637 | |
bb737bbe VG |
638 | /* blocking async request completes in a worker context */ |
639 | if (req->args->may_block) { | |
640 | struct virtio_fs_req_work *w; | |
641 | ||
642 | w = kzalloc(sizeof(*w), GFP_NOFS | __GFP_NOFAIL); | |
643 | INIT_WORK(&w->done_work, virtio_fs_complete_req_work); | |
644 | w->fsvq = fsvq; | |
645 | w->req = req; | |
646 | schedule_work(&w->done_work); | |
647 | } else { | |
648 | virtio_fs_request_complete(req, fsvq); | |
649 | } | |
a62a8ef9 SH |
650 | } |
651 | } | |
652 | ||
653 | /* Virtqueue interrupt handler */ | |
654 | static void virtio_fs_vq_done(struct virtqueue *vq) | |
655 | { | |
656 | struct virtio_fs_vq *fsvq = vq_to_fsvq(vq); | |
657 | ||
658 | dev_dbg(&vq->vdev->dev, "%s %s\n", __func__, fsvq->name); | |
659 | ||
660 | schedule_work(&fsvq->done_work); | |
661 | } | |
662 | ||
b43b7e81 VG |
663 | static void virtio_fs_init_vq(struct virtio_fs_vq *fsvq, char *name, |
664 | int vq_type) | |
665 | { | |
7c594bbd | 666 | strscpy(fsvq->name, name, VQ_NAME_LEN); |
b43b7e81 VG |
667 | spin_lock_init(&fsvq->lock); |
668 | INIT_LIST_HEAD(&fsvq->queued_reqs); | |
669 | INIT_LIST_HEAD(&fsvq->end_reqs); | |
670 | init_completion(&fsvq->in_flight_zero); | |
671 | ||
672 | if (vq_type == VQ_REQUEST) { | |
673 | INIT_WORK(&fsvq->done_work, virtio_fs_requests_done_work); | |
674 | INIT_DELAYED_WORK(&fsvq->dispatch_work, | |
675 | virtio_fs_request_dispatch_work); | |
676 | } else { | |
677 | INIT_WORK(&fsvq->done_work, virtio_fs_hiprio_done_work); | |
678 | INIT_DELAYED_WORK(&fsvq->dispatch_work, | |
679 | virtio_fs_hiprio_dispatch_work); | |
680 | } | |
681 | } | |
682 | ||
a62a8ef9 SH |
683 | /* Initialize virtqueues */ |
684 | static int virtio_fs_setup_vqs(struct virtio_device *vdev, | |
685 | struct virtio_fs *fs) | |
686 | { | |
687 | struct virtqueue **vqs; | |
688 | vq_callback_t **callbacks; | |
689 | const char **names; | |
690 | unsigned int i; | |
691 | int ret = 0; | |
692 | ||
2c0349ec MT |
693 | virtio_cread_le(vdev, struct virtio_fs_config, num_request_queues, |
694 | &fs->num_request_queues); | |
a62a8ef9 SH |
695 | if (fs->num_request_queues == 0) |
696 | return -EINVAL; | |
697 | ||
b43b7e81 | 698 | fs->nvqs = VQ_REQUEST + fs->num_request_queues; |
a62a8ef9 SH |
699 | fs->vqs = kcalloc(fs->nvqs, sizeof(fs->vqs[VQ_HIPRIO]), GFP_KERNEL); |
700 | if (!fs->vqs) | |
701 | return -ENOMEM; | |
702 | ||
703 | vqs = kmalloc_array(fs->nvqs, sizeof(vqs[VQ_HIPRIO]), GFP_KERNEL); | |
704 | callbacks = kmalloc_array(fs->nvqs, sizeof(callbacks[VQ_HIPRIO]), | |
705 | GFP_KERNEL); | |
706 | names = kmalloc_array(fs->nvqs, sizeof(names[VQ_HIPRIO]), GFP_KERNEL); | |
707 | if (!vqs || !callbacks || !names) { | |
708 | ret = -ENOMEM; | |
709 | goto out; | |
710 | } | |
711 | ||
b43b7e81 | 712 | /* Initialize the hiprio/forget request virtqueue */ |
a62a8ef9 | 713 | callbacks[VQ_HIPRIO] = virtio_fs_vq_done; |
b43b7e81 | 714 | virtio_fs_init_vq(&fs->vqs[VQ_HIPRIO], "hiprio", VQ_HIPRIO); |
a62a8ef9 | 715 | names[VQ_HIPRIO] = fs->vqs[VQ_HIPRIO].name; |
a62a8ef9 SH |
716 | |
717 | /* Initialize the requests virtqueues */ | |
718 | for (i = VQ_REQUEST; i < fs->nvqs; i++) { | |
b43b7e81 VG |
719 | char vq_name[VQ_NAME_LEN]; |
720 | ||
721 | snprintf(vq_name, VQ_NAME_LEN, "requests.%u", i - VQ_REQUEST); | |
722 | virtio_fs_init_vq(&fs->vqs[i], vq_name, VQ_REQUEST); | |
a62a8ef9 SH |
723 | callbacks[i] = virtio_fs_vq_done; |
724 | names[i] = fs->vqs[i].name; | |
725 | } | |
726 | ||
727 | ret = virtio_find_vqs(vdev, fs->nvqs, vqs, callbacks, names, NULL); | |
728 | if (ret < 0) | |
729 | goto out; | |
730 | ||
731 | for (i = 0; i < fs->nvqs; i++) | |
732 | fs->vqs[i].vq = vqs[i]; | |
733 | ||
734 | virtio_fs_start_all_queues(fs); | |
735 | out: | |
736 | kfree(names); | |
737 | kfree(callbacks); | |
738 | kfree(vqs); | |
739 | if (ret) | |
740 | kfree(fs->vqs); | |
741 | return ret; | |
742 | } | |
743 | ||
744 | /* Free virtqueues (device must already be reset) */ | |
1e5b9e04 | 745 | static void virtio_fs_cleanup_vqs(struct virtio_device *vdev) |
a62a8ef9 SH |
746 | { |
747 | vdev->config->del_vqs(vdev); | |
748 | } | |
749 | ||
22f3787e SH |
750 | /* Map a window offset to a page frame number. The window offset will have |
751 | * been produced by .iomap_begin(), which maps a file offset to a window | |
752 | * offset. | |
753 | */ | |
754 | static long virtio_fs_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, | |
e511c4a3 JC |
755 | long nr_pages, enum dax_access_mode mode, |
756 | void **kaddr, pfn_t *pfn) | |
22f3787e SH |
757 | { |
758 | struct virtio_fs *fs = dax_get_private(dax_dev); | |
759 | phys_addr_t offset = PFN_PHYS(pgoff); | |
73fb2c8b | 760 | size_t max_nr_pages = fs->window_len / PAGE_SIZE - pgoff; |
22f3787e SH |
761 | |
762 | if (kaddr) | |
763 | *kaddr = fs->window_kaddr + offset; | |
764 | if (pfn) | |
765 | *pfn = phys_to_pfn_t(fs->window_phys_addr + offset, | |
766 | PFN_DEV | PFN_MAP); | |
767 | return nr_pages > max_nr_pages ? max_nr_pages : nr_pages; | |
768 | } | |
769 | ||
22f3787e SH |
770 | static int virtio_fs_zero_page_range(struct dax_device *dax_dev, |
771 | pgoff_t pgoff, size_t nr_pages) | |
772 | { | |
773 | long rc; | |
774 | void *kaddr; | |
775 | ||
e511c4a3 JC |
776 | rc = dax_direct_access(dax_dev, pgoff, nr_pages, DAX_ACCESS, &kaddr, |
777 | NULL); | |
22f3787e | 778 | if (rc < 0) |
1ea7ca1b JC |
779 | return dax_mem2blk_err(rc); |
780 | ||
22f3787e SH |
781 | memset(kaddr, 0, nr_pages << PAGE_SHIFT); |
782 | dax_flush(dax_dev, kaddr, nr_pages << PAGE_SHIFT); | |
783 | return 0; | |
784 | } | |
785 | ||
786 | static const struct dax_operations virtio_fs_dax_ops = { | |
787 | .direct_access = virtio_fs_direct_access, | |
22f3787e SH |
788 | .zero_page_range = virtio_fs_zero_page_range, |
789 | }; | |
790 | ||
791 | static void virtio_fs_cleanup_dax(void *data) | |
792 | { | |
793 | struct dax_device *dax_dev = data; | |
794 | ||
795 | kill_dax(dax_dev); | |
796 | put_dax(dax_dev); | |
797 | } | |
798 | ||
562ce828 MD |
799 | DEFINE_FREE(cleanup_dax, struct dax_dev *, if (!IS_ERR_OR_NULL(_T)) virtio_fs_cleanup_dax(_T)) |
800 | ||
22f3787e SH |
801 | static int virtio_fs_setup_dax(struct virtio_device *vdev, struct virtio_fs *fs) |
802 | { | |
562ce828 | 803 | struct dax_device *dax_dev __free(cleanup_dax) = NULL; |
22f3787e SH |
804 | struct virtio_shm_region cache_reg; |
805 | struct dev_pagemap *pgmap; | |
806 | bool have_cache; | |
807 | ||
808 | if (!IS_ENABLED(CONFIG_FUSE_DAX)) | |
809 | return 0; | |
810 | ||
562ce828 MD |
811 | dax_dev = alloc_dax(fs, &virtio_fs_dax_ops); |
812 | if (IS_ERR(dax_dev)) { | |
813 | int rc = PTR_ERR(dax_dev); | |
814 | return rc == -EOPNOTSUPP ? 0 : rc; | |
815 | } | |
816 | ||
22f3787e SH |
817 | /* Get cache region */ |
818 | have_cache = virtio_get_shm_region(vdev, &cache_reg, | |
819 | (u8)VIRTIO_FS_SHMCAP_ID_CACHE); | |
820 | if (!have_cache) { | |
821 | dev_notice(&vdev->dev, "%s: No cache capability\n", __func__); | |
822 | return 0; | |
823 | } | |
824 | ||
825 | if (!devm_request_mem_region(&vdev->dev, cache_reg.addr, cache_reg.len, | |
826 | dev_name(&vdev->dev))) { | |
827 | dev_warn(&vdev->dev, "could not reserve region addr=0x%llx len=0x%llx\n", | |
828 | cache_reg.addr, cache_reg.len); | |
829 | return -EBUSY; | |
830 | } | |
831 | ||
832 | dev_notice(&vdev->dev, "Cache len: 0x%llx @ 0x%llx\n", cache_reg.len, | |
833 | cache_reg.addr); | |
834 | ||
835 | pgmap = devm_kzalloc(&vdev->dev, sizeof(*pgmap), GFP_KERNEL); | |
836 | if (!pgmap) | |
837 | return -ENOMEM; | |
838 | ||
839 | pgmap->type = MEMORY_DEVICE_FS_DAX; | |
840 | ||
841 | /* Ideally we would directly use the PCI BAR resource but | |
842 | * devm_memremap_pages() wants its own copy in pgmap. So | |
843 | * initialize a struct resource from scratch (only the start | |
844 | * and end fields will be used). | |
845 | */ | |
69456535 | 846 | pgmap->range = (struct range) { |
22f3787e SH |
847 | .start = (phys_addr_t) cache_reg.addr, |
848 | .end = (phys_addr_t) cache_reg.addr + cache_reg.len - 1, | |
849 | }; | |
69456535 | 850 | pgmap->nr_range = 1; |
22f3787e SH |
851 | |
852 | fs->window_kaddr = devm_memremap_pages(&vdev->dev, pgmap); | |
853 | if (IS_ERR(fs->window_kaddr)) | |
854 | return PTR_ERR(fs->window_kaddr); | |
855 | ||
856 | fs->window_phys_addr = (phys_addr_t) cache_reg.addr; | |
857 | fs->window_len = (phys_addr_t) cache_reg.len; | |
858 | ||
859 | dev_dbg(&vdev->dev, "%s: window kaddr 0x%px phys_addr 0x%llx len 0x%llx\n", | |
860 | __func__, fs->window_kaddr, cache_reg.addr, cache_reg.len); | |
861 | ||
562ce828 | 862 | fs->dax_dev = no_free_ptr(dax_dev); |
22f3787e SH |
863 | return devm_add_action_or_reset(&vdev->dev, virtio_fs_cleanup_dax, |
864 | fs->dax_dev); | |
865 | } | |
866 | ||
a62a8ef9 SH |
867 | static int virtio_fs_probe(struct virtio_device *vdev) |
868 | { | |
869 | struct virtio_fs *fs; | |
870 | int ret; | |
871 | ||
872 | fs = kzalloc(sizeof(*fs), GFP_KERNEL); | |
873 | if (!fs) | |
874 | return -ENOMEM; | |
875 | kref_init(&fs->refcount); | |
876 | vdev->priv = fs; | |
877 | ||
878 | ret = virtio_fs_read_tag(vdev, fs); | |
879 | if (ret < 0) | |
880 | goto out; | |
881 | ||
882 | ret = virtio_fs_setup_vqs(vdev, fs); | |
883 | if (ret < 0) | |
884 | goto out; | |
885 | ||
886 | /* TODO vq affinity */ | |
887 | ||
22f3787e SH |
888 | ret = virtio_fs_setup_dax(vdev, fs); |
889 | if (ret < 0) | |
890 | goto out_vqs; | |
891 | ||
a62a8ef9 SH |
892 | /* Bring the device online in case the filesystem is mounted and |
893 | * requests need to be sent before we return. | |
894 | */ | |
895 | virtio_device_ready(vdev); | |
896 | ||
897 | ret = virtio_fs_add_instance(fs); | |
898 | if (ret < 0) | |
899 | goto out_vqs; | |
900 | ||
901 | return 0; | |
902 | ||
903 | out_vqs: | |
d9679d00 | 904 | virtio_reset_device(vdev); |
1e5b9e04 | 905 | virtio_fs_cleanup_vqs(vdev); |
c79c5e01 | 906 | kfree(fs->vqs); |
a62a8ef9 SH |
907 | |
908 | out: | |
909 | vdev->priv = NULL; | |
910 | kfree(fs); | |
911 | return ret; | |
912 | } | |
913 | ||
914 | static void virtio_fs_stop_all_queues(struct virtio_fs *fs) | |
915 | { | |
916 | struct virtio_fs_vq *fsvq; | |
917 | int i; | |
918 | ||
919 | for (i = 0; i < fs->nvqs; i++) { | |
920 | fsvq = &fs->vqs[i]; | |
921 | spin_lock(&fsvq->lock); | |
922 | fsvq->connected = false; | |
923 | spin_unlock(&fsvq->lock); | |
924 | } | |
925 | } | |
926 | ||
927 | static void virtio_fs_remove(struct virtio_device *vdev) | |
928 | { | |
929 | struct virtio_fs *fs = vdev->priv; | |
930 | ||
931 | mutex_lock(&virtio_fs_mutex); | |
932 | /* This device is going away. No one should get new reference */ | |
933 | list_del_init(&fs->list); | |
934 | virtio_fs_stop_all_queues(fs); | |
724c15a4 | 935 | virtio_fs_drain_all_queues_locked(fs); |
d9679d00 | 936 | virtio_reset_device(vdev); |
1e5b9e04 | 937 | virtio_fs_cleanup_vqs(vdev); |
a62a8ef9 SH |
938 | |
939 | vdev->priv = NULL; | |
940 | /* Put device reference on virtio_fs object */ | |
941 | virtio_fs_put(fs); | |
942 | mutex_unlock(&virtio_fs_mutex); | |
943 | } | |
944 | ||
945 | #ifdef CONFIG_PM_SLEEP | |
946 | static int virtio_fs_freeze(struct virtio_device *vdev) | |
947 | { | |
948 | /* TODO need to save state here */ | |
949 | pr_warn("virtio-fs: suspend/resume not yet supported\n"); | |
950 | return -EOPNOTSUPP; | |
951 | } | |
952 | ||
953 | static int virtio_fs_restore(struct virtio_device *vdev) | |
954 | { | |
955 | /* TODO need to restore state here */ | |
956 | return 0; | |
957 | } | |
958 | #endif /* CONFIG_PM_SLEEP */ | |
959 | ||
00929447 | 960 | static const struct virtio_device_id id_table[] = { |
a62a8ef9 SH |
961 | { VIRTIO_ID_FS, VIRTIO_DEV_ANY_ID }, |
962 | {}, | |
963 | }; | |
964 | ||
00929447 | 965 | static const unsigned int feature_table[] = {}; |
a62a8ef9 SH |
966 | |
967 | static struct virtio_driver virtio_fs_driver = { | |
968 | .driver.name = KBUILD_MODNAME, | |
969 | .driver.owner = THIS_MODULE, | |
970 | .id_table = id_table, | |
971 | .feature_table = feature_table, | |
972 | .feature_table_size = ARRAY_SIZE(feature_table), | |
973 | .probe = virtio_fs_probe, | |
974 | .remove = virtio_fs_remove, | |
975 | #ifdef CONFIG_PM_SLEEP | |
976 | .freeze = virtio_fs_freeze, | |
977 | .restore = virtio_fs_restore, | |
978 | #endif | |
979 | }; | |
980 | ||
981 | static void virtio_fs_wake_forget_and_unlock(struct fuse_iqueue *fiq) | |
982 | __releases(fiq->lock) | |
983 | { | |
984 | struct fuse_forget_link *link; | |
985 | struct virtio_fs_forget *forget; | |
1efcf39e | 986 | struct virtio_fs_forget_req *req; |
a62a8ef9 | 987 | struct virtio_fs *fs; |
a62a8ef9 | 988 | struct virtio_fs_vq *fsvq; |
a62a8ef9 | 989 | u64 unique; |
a62a8ef9 SH |
990 | |
991 | link = fuse_dequeue_forget(fiq, 1, NULL); | |
992 | unique = fuse_get_unique(fiq); | |
993 | ||
994 | fs = fiq->priv; | |
995 | fsvq = &fs->vqs[VQ_HIPRIO]; | |
996 | spin_unlock(&fiq->lock); | |
997 | ||
998 | /* Allocate a buffer for the request */ | |
999 | forget = kmalloc(sizeof(*forget), GFP_NOFS | __GFP_NOFAIL); | |
1efcf39e | 1000 | req = &forget->req; |
a62a8ef9 | 1001 | |
1efcf39e | 1002 | req->ih = (struct fuse_in_header){ |
a62a8ef9 SH |
1003 | .opcode = FUSE_FORGET, |
1004 | .nodeid = link->forget_one.nodeid, | |
1005 | .unique = unique, | |
1efcf39e | 1006 | .len = sizeof(*req), |
a62a8ef9 | 1007 | }; |
1efcf39e | 1008 | req->arg = (struct fuse_forget_in){ |
a62a8ef9 SH |
1009 | .nlookup = link->forget_one.nlookup, |
1010 | }; | |
1011 | ||
58ada94f | 1012 | send_forget_request(fsvq, forget, false); |
a62a8ef9 SH |
1013 | kfree(link); |
1014 | } | |
1015 | ||
1016 | static void virtio_fs_wake_interrupt_and_unlock(struct fuse_iqueue *fiq) | |
1017 | __releases(fiq->lock) | |
1018 | { | |
1019 | /* | |
1020 | * TODO interrupts. | |
1021 | * | |
1022 | * Normal fs operations on a local filesystems aren't interruptible. | |
1023 | * Exceptions are blocking lock operations; for example fcntl(F_SETLKW) | |
1024 | * with shared lock between host and guest. | |
1025 | */ | |
1026 | spin_unlock(&fiq->lock); | |
1027 | } | |
1028 | ||
42d3e2d0 VG |
1029 | /* Count number of scatter-gather elements required */ |
1030 | static unsigned int sg_count_fuse_pages(struct fuse_page_desc *page_descs, | |
1031 | unsigned int num_pages, | |
1032 | unsigned int total_len) | |
1033 | { | |
1034 | unsigned int i; | |
1035 | unsigned int this_len; | |
1036 | ||
1037 | for (i = 0; i < num_pages && total_len; i++) { | |
1038 | this_len = min(page_descs[i].length, total_len); | |
1039 | total_len -= this_len; | |
1040 | } | |
1041 | ||
1042 | return i; | |
1043 | } | |
1044 | ||
a62a8ef9 SH |
1045 | /* Return the number of scatter-gather list elements required */ |
1046 | static unsigned int sg_count_fuse_req(struct fuse_req *req) | |
1047 | { | |
1048 | struct fuse_args *args = req->args; | |
1049 | struct fuse_args_pages *ap = container_of(args, typeof(*ap), args); | |
42d3e2d0 | 1050 | unsigned int size, total_sgs = 1 /* fuse_in_header */; |
a62a8ef9 SH |
1051 | |
1052 | if (args->in_numargs - args->in_pages) | |
1053 | total_sgs += 1; | |
1054 | ||
42d3e2d0 VG |
1055 | if (args->in_pages) { |
1056 | size = args->in_args[args->in_numargs - 1].size; | |
1057 | total_sgs += sg_count_fuse_pages(ap->descs, ap->num_pages, | |
1058 | size); | |
1059 | } | |
a62a8ef9 SH |
1060 | |
1061 | if (!test_bit(FR_ISREPLY, &req->flags)) | |
1062 | return total_sgs; | |
1063 | ||
1064 | total_sgs += 1 /* fuse_out_header */; | |
1065 | ||
1066 | if (args->out_numargs - args->out_pages) | |
1067 | total_sgs += 1; | |
1068 | ||
42d3e2d0 VG |
1069 | if (args->out_pages) { |
1070 | size = args->out_args[args->out_numargs - 1].size; | |
1071 | total_sgs += sg_count_fuse_pages(ap->descs, ap->num_pages, | |
1072 | size); | |
1073 | } | |
a62a8ef9 SH |
1074 | |
1075 | return total_sgs; | |
1076 | } | |
1077 | ||
1078 | /* Add pages to scatter-gather list and return number of elements used */ | |
1079 | static unsigned int sg_init_fuse_pages(struct scatterlist *sg, | |
1080 | struct page **pages, | |
1081 | struct fuse_page_desc *page_descs, | |
1082 | unsigned int num_pages, | |
1083 | unsigned int total_len) | |
1084 | { | |
1085 | unsigned int i; | |
1086 | unsigned int this_len; | |
1087 | ||
1088 | for (i = 0; i < num_pages && total_len; i++) { | |
1089 | sg_init_table(&sg[i], 1); | |
1090 | this_len = min(page_descs[i].length, total_len); | |
1091 | sg_set_page(&sg[i], pages[i], this_len, page_descs[i].offset); | |
1092 | total_len -= this_len; | |
1093 | } | |
1094 | ||
1095 | return i; | |
1096 | } | |
1097 | ||
1098 | /* Add args to scatter-gather list and return number of elements used */ | |
1099 | static unsigned int sg_init_fuse_args(struct scatterlist *sg, | |
1100 | struct fuse_req *req, | |
1101 | struct fuse_arg *args, | |
1102 | unsigned int numargs, | |
1103 | bool argpages, | |
1104 | void *argbuf, | |
1105 | unsigned int *len_used) | |
1106 | { | |
1107 | struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args); | |
1108 | unsigned int total_sgs = 0; | |
1109 | unsigned int len; | |
1110 | ||
1111 | len = fuse_len_args(numargs - argpages, args); | |
1112 | if (len) | |
1113 | sg_init_one(&sg[total_sgs++], argbuf, len); | |
1114 | ||
1115 | if (argpages) | |
1116 | total_sgs += sg_init_fuse_pages(&sg[total_sgs], | |
1117 | ap->pages, ap->descs, | |
1118 | ap->num_pages, | |
1119 | args[numargs - 1].size); | |
1120 | ||
1121 | if (len_used) | |
1122 | *len_used = len; | |
1123 | ||
1124 | return total_sgs; | |
1125 | } | |
1126 | ||
1127 | /* Add a request to a virtqueue and kick the device */ | |
1128 | static int virtio_fs_enqueue_req(struct virtio_fs_vq *fsvq, | |
a9bfd9dd | 1129 | struct fuse_req *req, bool in_flight) |
a62a8ef9 SH |
1130 | { |
1131 | /* requests need at least 4 elements */ | |
1132 | struct scatterlist *stack_sgs[6]; | |
1133 | struct scatterlist stack_sg[ARRAY_SIZE(stack_sgs)]; | |
1134 | struct scatterlist **sgs = stack_sgs; | |
1135 | struct scatterlist *sg = stack_sg; | |
1136 | struct virtqueue *vq; | |
1137 | struct fuse_args *args = req->args; | |
1138 | unsigned int argbuf_used = 0; | |
1139 | unsigned int out_sgs = 0; | |
1140 | unsigned int in_sgs = 0; | |
1141 | unsigned int total_sgs; | |
1142 | unsigned int i; | |
1143 | int ret; | |
1144 | bool notify; | |
5dbe190f | 1145 | struct fuse_pqueue *fpq; |
a62a8ef9 SH |
1146 | |
1147 | /* Does the sglist fit on the stack? */ | |
1148 | total_sgs = sg_count_fuse_req(req); | |
1149 | if (total_sgs > ARRAY_SIZE(stack_sgs)) { | |
1150 | sgs = kmalloc_array(total_sgs, sizeof(sgs[0]), GFP_ATOMIC); | |
1151 | sg = kmalloc_array(total_sgs, sizeof(sg[0]), GFP_ATOMIC); | |
1152 | if (!sgs || !sg) { | |
1153 | ret = -ENOMEM; | |
1154 | goto out; | |
1155 | } | |
1156 | } | |
1157 | ||
1158 | /* Use a bounce buffer since stack args cannot be mapped */ | |
1159 | ret = copy_args_to_argbuf(req); | |
1160 | if (ret < 0) | |
1161 | goto out; | |
1162 | ||
1163 | /* Request elements */ | |
1164 | sg_init_one(&sg[out_sgs++], &req->in.h, sizeof(req->in.h)); | |
1165 | out_sgs += sg_init_fuse_args(&sg[out_sgs], req, | |
1166 | (struct fuse_arg *)args->in_args, | |
1167 | args->in_numargs, args->in_pages, | |
1168 | req->argbuf, &argbuf_used); | |
1169 | ||
1170 | /* Reply elements */ | |
1171 | if (test_bit(FR_ISREPLY, &req->flags)) { | |
1172 | sg_init_one(&sg[out_sgs + in_sgs++], | |
1173 | &req->out.h, sizeof(req->out.h)); | |
1174 | in_sgs += sg_init_fuse_args(&sg[out_sgs + in_sgs], req, | |
1175 | args->out_args, args->out_numargs, | |
1176 | args->out_pages, | |
1177 | req->argbuf + argbuf_used, NULL); | |
1178 | } | |
1179 | ||
1180 | WARN_ON(out_sgs + in_sgs != total_sgs); | |
1181 | ||
1182 | for (i = 0; i < total_sgs; i++) | |
1183 | sgs[i] = &sg[i]; | |
1184 | ||
1185 | spin_lock(&fsvq->lock); | |
1186 | ||
1187 | if (!fsvq->connected) { | |
1188 | spin_unlock(&fsvq->lock); | |
1189 | ret = -ENOTCONN; | |
1190 | goto out; | |
1191 | } | |
1192 | ||
1193 | vq = fsvq->vq; | |
1194 | ret = virtqueue_add_sgs(vq, sgs, out_sgs, in_sgs, req, GFP_ATOMIC); | |
1195 | if (ret < 0) { | |
1196 | spin_unlock(&fsvq->lock); | |
1197 | goto out; | |
1198 | } | |
1199 | ||
5dbe190f VG |
1200 | /* Request successfully sent. */ |
1201 | fpq = &fsvq->fud->pq; | |
1202 | spin_lock(&fpq->lock); | |
1203 | list_add_tail(&req->list, fpq->processing); | |
1204 | spin_unlock(&fpq->lock); | |
1205 | set_bit(FR_SENT, &req->flags); | |
1206 | /* matches barrier in request_wait_answer() */ | |
1207 | smp_mb__after_atomic(); | |
1208 | ||
a9bfd9dd VG |
1209 | if (!in_flight) |
1210 | inc_in_flight_req(fsvq); | |
a62a8ef9 SH |
1211 | notify = virtqueue_kick_prepare(vq); |
1212 | ||
1213 | spin_unlock(&fsvq->lock); | |
1214 | ||
1215 | if (notify) | |
1216 | virtqueue_notify(vq); | |
1217 | ||
1218 | out: | |
1219 | if (ret < 0 && req->argbuf) { | |
1220 | kfree(req->argbuf); | |
1221 | req->argbuf = NULL; | |
1222 | } | |
1223 | if (sgs != stack_sgs) { | |
1224 | kfree(sgs); | |
1225 | kfree(sg); | |
1226 | } | |
1227 | ||
1228 | return ret; | |
1229 | } | |
1230 | ||
1231 | static void virtio_fs_wake_pending_and_unlock(struct fuse_iqueue *fiq) | |
1232 | __releases(fiq->lock) | |
1233 | { | |
1234 | unsigned int queue_id = VQ_REQUEST; /* TODO multiqueue */ | |
1235 | struct virtio_fs *fs; | |
a62a8ef9 | 1236 | struct fuse_req *req; |
51fecdd2 | 1237 | struct virtio_fs_vq *fsvq; |
a62a8ef9 SH |
1238 | int ret; |
1239 | ||
1240 | WARN_ON(list_empty(&fiq->pending)); | |
1241 | req = list_last_entry(&fiq->pending, struct fuse_req, list); | |
1242 | clear_bit(FR_PENDING, &req->flags); | |
1243 | list_del_init(&req->list); | |
1244 | WARN_ON(!list_empty(&fiq->pending)); | |
1245 | spin_unlock(&fiq->lock); | |
1246 | ||
1247 | fs = fiq->priv; | |
a62a8ef9 SH |
1248 | |
1249 | pr_debug("%s: opcode %u unique %#llx nodeid %#llx in.len %u out.len %u\n", | |
1250 | __func__, req->in.h.opcode, req->in.h.unique, | |
1251 | req->in.h.nodeid, req->in.h.len, | |
1252 | fuse_len_args(req->args->out_numargs, req->args->out_args)); | |
1253 | ||
51fecdd2 | 1254 | fsvq = &fs->vqs[queue_id]; |
a9bfd9dd | 1255 | ret = virtio_fs_enqueue_req(fsvq, req, false); |
a62a8ef9 SH |
1256 | if (ret < 0) { |
1257 | if (ret == -ENOMEM || ret == -ENOSPC) { | |
a9bfd9dd VG |
1258 | /* |
1259 | * Virtqueue full. Retry submission from worker | |
1260 | * context as we might be holding fc->bg_lock. | |
1261 | */ | |
1262 | spin_lock(&fsvq->lock); | |
1263 | list_add_tail(&req->list, &fsvq->queued_reqs); | |
1264 | inc_in_flight_req(fsvq); | |
1265 | schedule_delayed_work(&fsvq->dispatch_work, | |
1266 | msecs_to_jiffies(1)); | |
1267 | spin_unlock(&fsvq->lock); | |
1268 | return; | |
a62a8ef9 SH |
1269 | } |
1270 | req->out.h.error = ret; | |
1271 | pr_err("virtio-fs: virtio_fs_enqueue_req() failed %d\n", ret); | |
51fecdd2 VG |
1272 | |
1273 | /* Can't end request in submission context. Use a worker */ | |
1274 | spin_lock(&fsvq->lock); | |
1275 | list_add_tail(&req->list, &fsvq->end_reqs); | |
1276 | schedule_delayed_work(&fsvq->dispatch_work, 0); | |
1277 | spin_unlock(&fsvq->lock); | |
a62a8ef9 SH |
1278 | return; |
1279 | } | |
1280 | } | |
1281 | ||
00929447 | 1282 | static const struct fuse_iqueue_ops virtio_fs_fiq_ops = { |
a62a8ef9 SH |
1283 | .wake_forget_and_unlock = virtio_fs_wake_forget_and_unlock, |
1284 | .wake_interrupt_and_unlock = virtio_fs_wake_interrupt_and_unlock, | |
1285 | .wake_pending_and_unlock = virtio_fs_wake_pending_and_unlock, | |
1286 | .release = virtio_fs_fiq_release, | |
1287 | }; | |
1288 | ||
1dd53957 | 1289 | static inline void virtio_fs_ctx_set_defaults(struct fuse_fs_context *ctx) |
a62a8ef9 | 1290 | { |
1dd53957 VG |
1291 | ctx->rootmode = S_IFDIR; |
1292 | ctx->default_permissions = 1; | |
1293 | ctx->allow_other = 1; | |
1294 | ctx->max_read = UINT_MAX; | |
1295 | ctx->blksize = 512; | |
1296 | ctx->destroy = true; | |
1297 | ctx->no_control = true; | |
1298 | ctx->no_force_umount = true; | |
1299 | } | |
1300 | ||
1301 | static int virtio_fs_fill_super(struct super_block *sb, struct fs_context *fsc) | |
a62a8ef9 | 1302 | { |
fcee216b MR |
1303 | struct fuse_mount *fm = get_fuse_mount_super(sb); |
1304 | struct fuse_conn *fc = fm->fc; | |
a62a8ef9 | 1305 | struct virtio_fs *fs = fc->iq.priv; |
1dd53957 | 1306 | struct fuse_fs_context *ctx = fsc->fs_private; |
a62a8ef9 SH |
1307 | unsigned int i; |
1308 | int err; | |
a62a8ef9 | 1309 | |
1dd53957 | 1310 | virtio_fs_ctx_set_defaults(ctx); |
a62a8ef9 SH |
1311 | mutex_lock(&virtio_fs_mutex); |
1312 | ||
1313 | /* After holding mutex, make sure virtiofs device is still there. | |
1314 | * Though we are holding a reference to it, drive ->remove might | |
1315 | * still have cleaned up virtual queues. In that case bail out. | |
1316 | */ | |
1317 | err = -EINVAL; | |
1318 | if (list_empty(&fs->list)) { | |
1319 | pr_info("virtio-fs: tag <%s> not found\n", fs->tag); | |
1320 | goto err; | |
1321 | } | |
1322 | ||
1323 | err = -ENOMEM; | |
1324 | /* Allocate fuse_dev for hiprio and notification queues */ | |
7fd3abfa | 1325 | for (i = 0; i < fs->nvqs; i++) { |
a62a8ef9 SH |
1326 | struct virtio_fs_vq *fsvq = &fs->vqs[i]; |
1327 | ||
1328 | fsvq->fud = fuse_dev_alloc(); | |
1329 | if (!fsvq->fud) | |
1330 | goto err_free_fuse_devs; | |
1331 | } | |
1332 | ||
7fd3abfa | 1333 | /* virtiofs allocates and installs its own fuse devices */ |
1dd53957 | 1334 | ctx->fudptr = NULL; |
780b1b95 JX |
1335 | if (ctx->dax_mode != FUSE_DAX_NEVER) { |
1336 | if (ctx->dax_mode == FUSE_DAX_ALWAYS && !fs->dax_dev) { | |
3f9b9efd VG |
1337 | err = -EINVAL; |
1338 | pr_err("virtio-fs: dax can't be enabled as filesystem" | |
1339 | " device does not support it.\n"); | |
1340 | goto err_free_fuse_devs; | |
1341 | } | |
1dd53957 | 1342 | ctx->dax_dev = fs->dax_dev; |
3f9b9efd | 1343 | } |
1dd53957 | 1344 | err = fuse_fill_super_common(sb, ctx); |
a62a8ef9 SH |
1345 | if (err < 0) |
1346 | goto err_free_fuse_devs; | |
1347 | ||
a62a8ef9 SH |
1348 | for (i = 0; i < fs->nvqs; i++) { |
1349 | struct virtio_fs_vq *fsvq = &fs->vqs[i]; | |
1350 | ||
a62a8ef9 SH |
1351 | fuse_dev_install(fsvq->fud, fc); |
1352 | } | |
1353 | ||
1354 | /* Previous unmount will stop all queues. Start these again */ | |
1355 | virtio_fs_start_all_queues(fs); | |
fcee216b | 1356 | fuse_send_init(fm); |
a62a8ef9 SH |
1357 | mutex_unlock(&virtio_fs_mutex); |
1358 | return 0; | |
1359 | ||
1360 | err_free_fuse_devs: | |
1361 | virtio_fs_free_devs(fs); | |
1362 | err: | |
1363 | mutex_unlock(&virtio_fs_mutex); | |
1364 | return err; | |
1365 | } | |
1366 | ||
fcee216b | 1367 | static void virtio_fs_conn_destroy(struct fuse_mount *fm) |
a62a8ef9 | 1368 | { |
fcee216b MR |
1369 | struct fuse_conn *fc = fm->fc; |
1370 | struct virtio_fs *vfs = fc->iq.priv; | |
1371 | struct virtio_fs_vq *fsvq = &vfs->vqs[VQ_HIPRIO]; | |
a62a8ef9 | 1372 | |
fcee216b MR |
1373 | /* Stop dax worker. Soon evict_inodes() will be called which |
1374 | * will free all memory ranges belonging to all inodes. | |
9a752d18 VG |
1375 | */ |
1376 | if (IS_ENABLED(CONFIG_FUSE_DAX)) | |
1377 | fuse_dax_cancel_work(fc); | |
a62a8ef9 SH |
1378 | |
1379 | /* Stop forget queue. Soon destroy will be sent */ | |
1380 | spin_lock(&fsvq->lock); | |
1381 | fsvq->connected = false; | |
1382 | spin_unlock(&fsvq->lock); | |
1383 | virtio_fs_drain_all_queues(vfs); | |
1384 | ||
fcee216b | 1385 | fuse_conn_destroy(fm); |
a62a8ef9 | 1386 | |
fcee216b | 1387 | /* fuse_conn_destroy() must have sent destroy. Stop all queues |
a62a8ef9 SH |
1388 | * and drain one more time and free fuse devices. Freeing fuse |
1389 | * devices will drop their reference on fuse_conn and that in | |
1390 | * turn will drop its reference on virtio_fs object. | |
1391 | */ | |
1392 | virtio_fs_stop_all_queues(vfs); | |
1393 | virtio_fs_drain_all_queues(vfs); | |
1394 | virtio_fs_free_devs(vfs); | |
1395 | } | |
1396 | ||
fcee216b MR |
1397 | static void virtio_kill_sb(struct super_block *sb) |
1398 | { | |
1399 | struct fuse_mount *fm = get_fuse_mount_super(sb); | |
1400 | bool last; | |
1401 | ||
1402 | /* If mount failed, we can still be called without any fc */ | |
d534d31d | 1403 | if (sb->s_root) { |
fcee216b MR |
1404 | last = fuse_mount_remove(fm); |
1405 | if (last) | |
1406 | virtio_fs_conn_destroy(fm); | |
1407 | } | |
1408 | kill_anon_super(sb); | |
a27c061a | 1409 | fuse_mount_destroy(fm); |
fcee216b MR |
1410 | } |
1411 | ||
a62a8ef9 SH |
1412 | static int virtio_fs_test_super(struct super_block *sb, |
1413 | struct fs_context *fsc) | |
1414 | { | |
fcee216b MR |
1415 | struct fuse_mount *fsc_fm = fsc->s_fs_info; |
1416 | struct fuse_mount *sb_fm = get_fuse_mount_super(sb); | |
a62a8ef9 | 1417 | |
fcee216b | 1418 | return fsc_fm->fc->iq.priv == sb_fm->fc->iq.priv; |
a62a8ef9 SH |
1419 | } |
1420 | ||
a62a8ef9 SH |
1421 | static int virtio_fs_get_tree(struct fs_context *fsc) |
1422 | { | |
1423 | struct virtio_fs *fs; | |
1424 | struct super_block *sb; | |
a7f0d7aa | 1425 | struct fuse_conn *fc = NULL; |
fcee216b | 1426 | struct fuse_mount *fm; |
a7f0d7aa CK |
1427 | unsigned int virtqueue_size; |
1428 | int err = -EIO; | |
a62a8ef9 SH |
1429 | |
1430 | /* This gets a reference on virtio_fs object. This ptr gets installed | |
1431 | * in fc->iq->priv. Once fuse_conn is going away, it calls ->put() | |
1432 | * to drop the reference to this object. | |
1433 | */ | |
1434 | fs = virtio_fs_find_instance(fsc->source); | |
1435 | if (!fs) { | |
1436 | pr_info("virtio-fs: tag <%s> not found\n", fsc->source); | |
1437 | return -EINVAL; | |
1438 | } | |
1439 | ||
a7f0d7aa CK |
1440 | virtqueue_size = virtqueue_get_vring_size(fs->vqs[VQ_REQUEST].vq); |
1441 | if (WARN_ON(virtqueue_size <= FUSE_HEADER_OVERHEAD)) | |
1442 | goto out_err; | |
1443 | ||
833c5a42 | 1444 | err = -ENOMEM; |
a62a8ef9 | 1445 | fc = kzalloc(sizeof(struct fuse_conn), GFP_KERNEL); |
833c5a42 MS |
1446 | if (!fc) |
1447 | goto out_err; | |
a62a8ef9 | 1448 | |
fcee216b | 1449 | fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL); |
833c5a42 MS |
1450 | if (!fm) |
1451 | goto out_err; | |
fcee216b | 1452 | |
0a7419c6 | 1453 | fuse_conn_init(fc, fm, fsc->user_ns, &virtio_fs_fiq_ops, fs); |
a62a8ef9 SH |
1454 | fc->release = fuse_free_conn; |
1455 | fc->delete_stale = true; | |
bf109c64 | 1456 | fc->auto_submounts = true; |
2d82ab25 | 1457 | fc->sync_fs = true; |
a62a8ef9 | 1458 | |
a7f0d7aa CK |
1459 | /* Tell FUSE to split requests that exceed the virtqueue's size */ |
1460 | fc->max_pages_limit = min_t(unsigned int, fc->max_pages_limit, | |
1461 | virtqueue_size - FUSE_HEADER_OVERHEAD); | |
1462 | ||
fcee216b | 1463 | fsc->s_fs_info = fm; |
b19d3d00 | 1464 | sb = sget_fc(fsc, virtio_fs_test_super, set_anon_super_fc); |
c191cd07 MS |
1465 | if (fsc->s_fs_info) |
1466 | fuse_mount_destroy(fm); | |
a62a8ef9 SH |
1467 | if (IS_ERR(sb)) |
1468 | return PTR_ERR(sb); | |
1469 | ||
1470 | if (!sb->s_root) { | |
1dd53957 | 1471 | err = virtio_fs_fill_super(sb, fsc); |
a62a8ef9 SH |
1472 | if (err) { |
1473 | deactivate_locked_super(sb); | |
1474 | return err; | |
1475 | } | |
1476 | ||
1477 | sb->s_flags |= SB_ACTIVE; | |
1478 | } | |
1479 | ||
1480 | WARN_ON(fsc->root); | |
1481 | fsc->root = dget(sb->s_root); | |
1482 | return 0; | |
833c5a42 MS |
1483 | |
1484 | out_err: | |
1485 | kfree(fc); | |
1486 | mutex_lock(&virtio_fs_mutex); | |
1487 | virtio_fs_put(fs); | |
1488 | mutex_unlock(&virtio_fs_mutex); | |
1489 | return err; | |
a62a8ef9 SH |
1490 | } |
1491 | ||
1492 | static const struct fs_context_operations virtio_fs_context_ops = { | |
84c21507 | 1493 | .free = virtio_fs_free_fsc, |
1dd53957 | 1494 | .parse_param = virtio_fs_parse_param, |
a62a8ef9 SH |
1495 | .get_tree = virtio_fs_get_tree, |
1496 | }; | |
1497 | ||
1498 | static int virtio_fs_init_fs_context(struct fs_context *fsc) | |
1499 | { | |
1dd53957 VG |
1500 | struct fuse_fs_context *ctx; |
1501 | ||
fe0a7bd8 GK |
1502 | if (fsc->purpose == FS_CONTEXT_FOR_SUBMOUNT) |
1503 | return fuse_init_fs_context_submount(fsc); | |
1504 | ||
1dd53957 VG |
1505 | ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL); |
1506 | if (!ctx) | |
1507 | return -ENOMEM; | |
1508 | fsc->fs_private = ctx; | |
a62a8ef9 SH |
1509 | fsc->ops = &virtio_fs_context_ops; |
1510 | return 0; | |
1511 | } | |
1512 | ||
1513 | static struct file_system_type virtio_fs_type = { | |
1514 | .owner = THIS_MODULE, | |
1515 | .name = "virtiofs", | |
1516 | .init_fs_context = virtio_fs_init_fs_context, | |
1517 | .kill_sb = virtio_kill_sb, | |
1518 | }; | |
1519 | ||
1520 | static int __init virtio_fs_init(void) | |
1521 | { | |
1522 | int ret; | |
1523 | ||
1524 | ret = register_virtio_driver(&virtio_fs_driver); | |
1525 | if (ret < 0) | |
1526 | return ret; | |
1527 | ||
1528 | ret = register_filesystem(&virtio_fs_type); | |
1529 | if (ret < 0) { | |
1530 | unregister_virtio_driver(&virtio_fs_driver); | |
1531 | return ret; | |
1532 | } | |
1533 | ||
1534 | return 0; | |
1535 | } | |
1536 | module_init(virtio_fs_init); | |
1537 | ||
1538 | static void __exit virtio_fs_exit(void) | |
1539 | { | |
1540 | unregister_filesystem(&virtio_fs_type); | |
1541 | unregister_virtio_driver(&virtio_fs_driver); | |
1542 | } | |
1543 | module_exit(virtio_fs_exit); | |
1544 | ||
1545 | MODULE_AUTHOR("Stefan Hajnoczi <stefanha@redhat.com>"); | |
1546 | MODULE_DESCRIPTION("Virtio Filesystem"); | |
1547 | MODULE_LICENSE("GPL"); | |
1548 | MODULE_ALIAS_FS(KBUILD_MODNAME); | |
1549 | MODULE_DEVICE_TABLE(virtio, id_table); |