Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* Provide a way to create a superblock configuration context within the kernel
3 * that allows a superblock to be set up prior to mounting.
4 *
5 * Copyright (C) 2017 Red Hat, Inc. All Rights Reserved.
6 * Written by David Howells (dhowells@redhat.com)
7 */
8
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10#include <linux/module.h>
11#include <linux/fs_context.h>
12#include <linux/fs_parser.h>
13#include <linux/fs.h>
14#include <linux/mount.h>
15#include <linux/nsproxy.h>
16#include <linux/slab.h>
17#include <linux/magic.h>
18#include <linux/security.h>
19#include <linux/mnt_namespace.h>
20#include <linux/pid_namespace.h>
21#include <linux/user_namespace.h>
22#include <net/net_namespace.h>
23#include <asm/sections.h>
24#include "mount.h"
25#include "internal.h"
26
27enum legacy_fs_param {
28 LEGACY_FS_UNSET_PARAMS,
29 LEGACY_FS_MONOLITHIC_PARAMS,
30 LEGACY_FS_INDIVIDUAL_PARAMS,
31};
32
33struct legacy_fs_context {
34 char *legacy_data; /* Data page for legacy filesystems */
35 size_t data_size;
36 enum legacy_fs_param param_type;
37};
38
39static int legacy_init_fs_context(struct fs_context *fc);
40
41static const struct constant_table common_set_sb_flag[] = {
42 { "dirsync", SB_DIRSYNC },
43 { "lazytime", SB_LAZYTIME },
44 { "mand", SB_MANDLOCK },
45 { "ro", SB_RDONLY },
46 { "sync", SB_SYNCHRONOUS },
47 { },
48};
49
50static const struct constant_table common_clear_sb_flag[] = {
51 { "async", SB_SYNCHRONOUS },
52 { "nolazytime", SB_LAZYTIME },
53 { "nomand", SB_MANDLOCK },
54 { "rw", SB_RDONLY },
55 { },
56};
57
58/*
59 * Check for a common mount option that manipulates s_flags.
60 */
61static int vfs_parse_sb_flag(struct fs_context *fc, const char *key)
62{
63 unsigned int token;
64
65 token = lookup_constant(common_set_sb_flag, key, 0);
66 if (token) {
67 fc->sb_flags |= token;
68 fc->sb_flags_mask |= token;
69 return 0;
70 }
71
72 token = lookup_constant(common_clear_sb_flag, key, 0);
73 if (token) {
74 fc->sb_flags &= ~token;
75 fc->sb_flags_mask |= token;
76 return 0;
77 }
78
79 return -ENOPARAM;
80}
81
82/**
83 * vfs_parse_fs_param_source - Handle setting "source" via parameter
84 * @fc: The filesystem context to modify
85 * @param: The parameter
86 *
87 * This is a simple helper for filesystems to verify that the "source" they
88 * accept is sane.
89 *
90 * Returns 0 on success, -ENOPARAM if this is not "source" parameter, and
91 * -EINVAL otherwise. In the event of failure, supplementary error information
92 * is logged.
93 */
94int vfs_parse_fs_param_source(struct fs_context *fc, struct fs_parameter *param)
95{
96 if (strcmp(param->key, "source") != 0)
97 return -ENOPARAM;
98
99 if (param->type != fs_value_is_string)
100 return invalf(fc, "Non-string source");
101
102 if (fc->source)
103 return invalf(fc, "Multiple sources");
104
105 fc->source = param->string;
106 param->string = NULL;
107 return 0;
108}
109EXPORT_SYMBOL(vfs_parse_fs_param_source);
110
111/**
112 * vfs_parse_fs_param - Add a single parameter to a superblock config
113 * @fc: The filesystem context to modify
114 * @param: The parameter
115 *
116 * A single mount option in string form is applied to the filesystem context
117 * being set up. Certain standard options (for example "ro") are translated
118 * into flag bits without going to the filesystem. The active security module
119 * is allowed to observe and poach options. Any other options are passed over
120 * to the filesystem to parse.
121 *
122 * This may be called multiple times for a context.
123 *
124 * Returns 0 on success and a negative error code on failure. In the event of
125 * failure, supplementary error information may have been set.
126 */
127int vfs_parse_fs_param(struct fs_context *fc, struct fs_parameter *param)
128{
129 int ret;
130
131 if (!param->key)
132 return invalf(fc, "Unnamed parameter\n");
133
134 ret = vfs_parse_sb_flag(fc, param->key);
135 if (ret != -ENOPARAM)
136 return ret;
137
138 ret = security_fs_context_parse_param(fc, param);
139 if (ret != -ENOPARAM)
140 /* Param belongs to the LSM or is disallowed by the LSM; so
141 * don't pass to the FS.
142 */
143 return ret;
144
145 if (fc->ops->parse_param) {
146 ret = fc->ops->parse_param(fc, param);
147 if (ret != -ENOPARAM)
148 return ret;
149 }
150
151 /* If the filesystem doesn't take any arguments, give it the
152 * default handling of source.
153 */
154 ret = vfs_parse_fs_param_source(fc, param);
155 if (ret != -ENOPARAM)
156 return ret;
157
158 return invalf(fc, "%s: Unknown parameter '%s'",
159 fc->fs_type->name, param->key);
160}
161EXPORT_SYMBOL(vfs_parse_fs_param);
162
163/**
164 * vfs_parse_fs_string - Convenience function to just parse a string.
165 * @fc: Filesystem context.
166 * @key: Parameter name.
167 * @value: Default value.
168 * @v_size: Maximum number of bytes in the value.
169 */
170int vfs_parse_fs_string(struct fs_context *fc, const char *key,
171 const char *value, size_t v_size)
172{
173 int ret;
174
175 struct fs_parameter param = {
176 .key = key,
177 .type = fs_value_is_flag,
178 .size = v_size,
179 };
180
181 if (value) {
182 param.string = kmemdup_nul(value, v_size, GFP_KERNEL);
183 if (!param.string)
184 return -ENOMEM;
185 param.type = fs_value_is_string;
186 }
187
188 ret = vfs_parse_fs_param(fc, ¶m);
189 kfree(param.string);
190 return ret;
191}
192EXPORT_SYMBOL(vfs_parse_fs_string);
193
194/**
195 * vfs_parse_monolithic_sep - Parse key[=val][,key[=val]]* mount data
196 * @fc: The superblock configuration to fill in.
197 * @data: The data to parse
198 * @sep: callback for separating next option
199 *
200 * Parse a blob of data that's in key[=val][,key[=val]]* form with a custom
201 * option separator callback.
202 *
203 * Returns 0 on success or the error returned by the ->parse_option() fs_context
204 * operation on failure.
205 */
206int vfs_parse_monolithic_sep(struct fs_context *fc, void *data,
207 char *(*sep)(char **))
208{
209 char *options = data, *key;
210 int ret = 0;
211
212 if (!options)
213 return 0;
214
215 ret = security_sb_eat_lsm_opts(options, &fc->security);
216 if (ret)
217 return ret;
218
219 while ((key = sep(&options)) != NULL) {
220 if (*key) {
221 size_t v_len = 0;
222 char *value = strchr(key, '=');
223
224 if (value) {
225 if (value == key)
226 continue;
227 *value++ = 0;
228 v_len = strlen(value);
229 }
230 ret = vfs_parse_fs_string(fc, key, value, v_len);
231 if (ret < 0)
232 break;
233 }
234 }
235
236 return ret;
237}
238EXPORT_SYMBOL(vfs_parse_monolithic_sep);
239
240static char *vfs_parse_comma_sep(char **s)
241{
242 return strsep(s, ",");
243}
244
245/**
246 * generic_parse_monolithic - Parse key[=val][,key[=val]]* mount data
247 * @fc: The superblock configuration to fill in.
248 * @data: The data to parse
249 *
250 * Parse a blob of data that's in key[=val][,key[=val]]* form. This can be
251 * called from the ->monolithic_mount_data() fs_context operation.
252 *
253 * Returns 0 on success or the error returned by the ->parse_option() fs_context
254 * operation on failure.
255 */
256int generic_parse_monolithic(struct fs_context *fc, void *data)
257{
258 return vfs_parse_monolithic_sep(fc, data, vfs_parse_comma_sep);
259}
260EXPORT_SYMBOL(generic_parse_monolithic);
261
262/**
263 * alloc_fs_context - Create a filesystem context.
264 * @fs_type: The filesystem type.
265 * @reference: The dentry from which this one derives (or NULL)
266 * @sb_flags: Filesystem/superblock flags (SB_*)
267 * @sb_flags_mask: Applicable members of @sb_flags
268 * @purpose: The purpose that this configuration shall be used for.
269 *
270 * Open a filesystem and create a mount context. The mount context is
271 * initialised with the supplied flags and, if a submount/automount from
272 * another superblock (referred to by @reference) is supplied, may have
273 * parameters such as namespaces copied across from that superblock.
274 */
275static struct fs_context *alloc_fs_context(struct file_system_type *fs_type,
276 struct dentry *reference,
277 unsigned int sb_flags,
278 unsigned int sb_flags_mask,
279 enum fs_context_purpose purpose)
280{
281 int (*init_fs_context)(struct fs_context *);
282 struct fs_context *fc;
283 int ret = -ENOMEM;
284
285 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL_ACCOUNT);
286 if (!fc)
287 return ERR_PTR(-ENOMEM);
288
289 fc->purpose = purpose;
290 fc->sb_flags = sb_flags;
291 fc->sb_flags_mask = sb_flags_mask;
292 fc->fs_type = get_filesystem(fs_type);
293 fc->cred = get_current_cred();
294 fc->net_ns = get_net(current->nsproxy->net_ns);
295 fc->log.prefix = fs_type->name;
296
297 mutex_init(&fc->uapi_mutex);
298
299 switch (purpose) {
300 case FS_CONTEXT_FOR_MOUNT:
301 fc->user_ns = get_user_ns(fc->cred->user_ns);
302 break;
303 case FS_CONTEXT_FOR_SUBMOUNT:
304 fc->user_ns = get_user_ns(reference->d_sb->s_user_ns);
305 break;
306 case FS_CONTEXT_FOR_RECONFIGURE:
307 atomic_inc(&reference->d_sb->s_active);
308 fc->user_ns = get_user_ns(reference->d_sb->s_user_ns);
309 fc->root = dget(reference);
310 break;
311 }
312
313 /* TODO: Make all filesystems support this unconditionally */
314 init_fs_context = fc->fs_type->init_fs_context;
315 if (!init_fs_context)
316 init_fs_context = legacy_init_fs_context;
317
318 ret = init_fs_context(fc);
319 if (ret < 0)
320 goto err_fc;
321 fc->need_free = true;
322 return fc;
323
324err_fc:
325 put_fs_context(fc);
326 return ERR_PTR(ret);
327}
328
329struct fs_context *fs_context_for_mount(struct file_system_type *fs_type,
330 unsigned int sb_flags)
331{
332 return alloc_fs_context(fs_type, NULL, sb_flags, 0,
333 FS_CONTEXT_FOR_MOUNT);
334}
335EXPORT_SYMBOL(fs_context_for_mount);
336
337struct fs_context *fs_context_for_reconfigure(struct dentry *dentry,
338 unsigned int sb_flags,
339 unsigned int sb_flags_mask)
340{
341 return alloc_fs_context(dentry->d_sb->s_type, dentry, sb_flags,
342 sb_flags_mask, FS_CONTEXT_FOR_RECONFIGURE);
343}
344EXPORT_SYMBOL(fs_context_for_reconfigure);
345
346/**
347 * fs_context_for_submount: allocate a new fs_context for a submount
348 * @type: file_system_type of the new context
349 * @reference: reference dentry from which to copy relevant info
350 *
351 * Allocate a new fs_context suitable for a submount. This also ensures that
352 * the fc->security object is inherited from @reference (if needed).
353 */
354struct fs_context *fs_context_for_submount(struct file_system_type *type,
355 struct dentry *reference)
356{
357 struct fs_context *fc;
358 int ret;
359
360 fc = alloc_fs_context(type, reference, 0, 0, FS_CONTEXT_FOR_SUBMOUNT);
361 if (IS_ERR(fc))
362 return fc;
363
364 ret = security_fs_context_submount(fc, reference->d_sb);
365 if (ret) {
366 put_fs_context(fc);
367 return ERR_PTR(ret);
368 }
369
370 return fc;
371}
372EXPORT_SYMBOL(fs_context_for_submount);
373
374void fc_drop_locked(struct fs_context *fc)
375{
376 struct super_block *sb = fc->root->d_sb;
377 dput(fc->root);
378 fc->root = NULL;
379 deactivate_locked_super(sb);
380}
381
382static void legacy_fs_context_free(struct fs_context *fc);
383
384/**
385 * vfs_dup_fs_context - Duplicate a filesystem context.
386 * @src_fc: The context to copy.
387 */
388struct fs_context *vfs_dup_fs_context(struct fs_context *src_fc)
389{
390 struct fs_context *fc;
391 int ret;
392
393 if (!src_fc->ops->dup)
394 return ERR_PTR(-EOPNOTSUPP);
395
396 fc = kmemdup(src_fc, sizeof(struct fs_context), GFP_KERNEL);
397 if (!fc)
398 return ERR_PTR(-ENOMEM);
399
400 mutex_init(&fc->uapi_mutex);
401
402 fc->fs_private = NULL;
403 fc->s_fs_info = NULL;
404 fc->source = NULL;
405 fc->security = NULL;
406 get_filesystem(fc->fs_type);
407 get_net(fc->net_ns);
408 get_user_ns(fc->user_ns);
409 get_cred(fc->cred);
410 if (fc->log.log)
411 refcount_inc(&fc->log.log->usage);
412
413 /* Can't call put until we've called ->dup */
414 ret = fc->ops->dup(fc, src_fc);
415 if (ret < 0)
416 goto err_fc;
417
418 ret = security_fs_context_dup(fc, src_fc);
419 if (ret < 0)
420 goto err_fc;
421 return fc;
422
423err_fc:
424 put_fs_context(fc);
425 return ERR_PTR(ret);
426}
427EXPORT_SYMBOL(vfs_dup_fs_context);
428
429/**
430 * logfc - Log a message to a filesystem context
431 * @log: The filesystem context to log to, or NULL to use printk.
432 * @prefix: A string to prefix the output with, or NULL.
433 * @level: 'w' for a warning, 'e' for an error. Anything else is a notice.
434 * @fmt: The format of the buffer.
435 */
436void logfc(struct fc_log *log, const char *prefix, char level, const char *fmt, ...)
437{
438 va_list va;
439 struct va_format vaf = {.fmt = fmt, .va = &va};
440
441 va_start(va, fmt);
442 if (!log) {
443 switch (level) {
444 case 'w':
445 printk(KERN_WARNING "%s%s%pV\n", prefix ? prefix : "",
446 prefix ? ": " : "", &vaf);
447 break;
448 case 'e':
449 printk(KERN_ERR "%s%s%pV\n", prefix ? prefix : "",
450 prefix ? ": " : "", &vaf);
451 break;
452 default:
453 printk(KERN_NOTICE "%s%s%pV\n", prefix ? prefix : "",
454 prefix ? ": " : "", &vaf);
455 break;
456 }
457 } else {
458 unsigned int logsize = ARRAY_SIZE(log->buffer);
459 u8 index;
460 char *q = kasprintf(GFP_KERNEL, "%c %s%s%pV\n", level,
461 prefix ? prefix : "",
462 prefix ? ": " : "", &vaf);
463
464 index = log->head & (logsize - 1);
465 BUILD_BUG_ON(sizeof(log->head) != sizeof(u8) ||
466 sizeof(log->tail) != sizeof(u8));
467 if ((u8)(log->head - log->tail) == logsize) {
468 /* The buffer is full, discard the oldest message */
469 if (log->need_free & (1 << index))
470 kfree(log->buffer[index]);
471 log->tail++;
472 }
473
474 log->buffer[index] = q ? q : "OOM: Can't store error string";
475 if (q)
476 log->need_free |= 1 << index;
477 else
478 log->need_free &= ~(1 << index);
479 log->head++;
480 }
481 va_end(va);
482}
483EXPORT_SYMBOL(logfc);
484
485/*
486 * Free a logging structure.
487 */
488static void put_fc_log(struct fs_context *fc)
489{
490 struct fc_log *log = fc->log.log;
491 int i;
492
493 if (log) {
494 if (refcount_dec_and_test(&log->usage)) {
495 fc->log.log = NULL;
496 for (i = 0; i <= 7; i++)
497 if (log->need_free & (1 << i))
498 kfree(log->buffer[i]);
499 kfree(log);
500 }
501 }
502}
503
504/**
505 * put_fs_context - Dispose of a superblock configuration context.
506 * @fc: The context to dispose of.
507 */
508void put_fs_context(struct fs_context *fc)
509{
510 struct super_block *sb;
511
512 if (fc->root) {
513 sb = fc->root->d_sb;
514 dput(fc->root);
515 fc->root = NULL;
516 deactivate_super(sb);
517 }
518
519 if (fc->need_free && fc->ops && fc->ops->free)
520 fc->ops->free(fc);
521
522 security_free_mnt_opts(&fc->security);
523 put_net(fc->net_ns);
524 put_user_ns(fc->user_ns);
525 put_cred(fc->cred);
526 put_fc_log(fc);
527 put_filesystem(fc->fs_type);
528 kfree(fc->source);
529 kfree(fc);
530}
531EXPORT_SYMBOL(put_fs_context);
532
533/*
534 * Free the config for a filesystem that doesn't support fs_context.
535 */
536static void legacy_fs_context_free(struct fs_context *fc)
537{
538 struct legacy_fs_context *ctx = fc->fs_private;
539
540 if (ctx) {
541 if (ctx->param_type == LEGACY_FS_INDIVIDUAL_PARAMS)
542 kfree(ctx->legacy_data);
543 kfree(ctx);
544 }
545}
546
547/*
548 * Duplicate a legacy config.
549 */
550static int legacy_fs_context_dup(struct fs_context *fc, struct fs_context *src_fc)
551{
552 struct legacy_fs_context *ctx;
553 struct legacy_fs_context *src_ctx = src_fc->fs_private;
554
555 ctx = kmemdup(src_ctx, sizeof(*src_ctx), GFP_KERNEL);
556 if (!ctx)
557 return -ENOMEM;
558
559 if (ctx->param_type == LEGACY_FS_INDIVIDUAL_PARAMS) {
560 ctx->legacy_data = kmemdup(src_ctx->legacy_data,
561 src_ctx->data_size, GFP_KERNEL);
562 if (!ctx->legacy_data) {
563 kfree(ctx);
564 return -ENOMEM;
565 }
566 }
567
568 fc->fs_private = ctx;
569 return 0;
570}
571
572/*
573 * Add a parameter to a legacy config. We build up a comma-separated list of
574 * options.
575 */
576static int legacy_parse_param(struct fs_context *fc, struct fs_parameter *param)
577{
578 struct legacy_fs_context *ctx = fc->fs_private;
579 unsigned int size = ctx->data_size;
580 size_t len = 0;
581 int ret;
582
583 ret = vfs_parse_fs_param_source(fc, param);
584 if (ret != -ENOPARAM)
585 return ret;
586
587 if (ctx->param_type == LEGACY_FS_MONOLITHIC_PARAMS)
588 return invalf(fc, "VFS: Legacy: Can't mix monolithic and individual options");
589
590 switch (param->type) {
591 case fs_value_is_string:
592 len = 1 + param->size;
593 fallthrough;
594 case fs_value_is_flag:
595 len += strlen(param->key);
596 break;
597 default:
598 return invalf(fc, "VFS: Legacy: Parameter type for '%s' not supported",
599 param->key);
600 }
601
602 if (size + len + 2 > PAGE_SIZE)
603 return invalf(fc, "VFS: Legacy: Cumulative options too large");
604 if (strchr(param->key, ',') ||
605 (param->type == fs_value_is_string &&
606 memchr(param->string, ',', param->size)))
607 return invalf(fc, "VFS: Legacy: Option '%s' contained comma",
608 param->key);
609 if (!ctx->legacy_data) {
610 ctx->legacy_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
611 if (!ctx->legacy_data)
612 return -ENOMEM;
613 }
614
615 if (size)
616 ctx->legacy_data[size++] = ',';
617 len = strlen(param->key);
618 memcpy(ctx->legacy_data + size, param->key, len);
619 size += len;
620 if (param->type == fs_value_is_string) {
621 ctx->legacy_data[size++] = '=';
622 memcpy(ctx->legacy_data + size, param->string, param->size);
623 size += param->size;
624 }
625 ctx->legacy_data[size] = '\0';
626 ctx->data_size = size;
627 ctx->param_type = LEGACY_FS_INDIVIDUAL_PARAMS;
628 return 0;
629}
630
631/*
632 * Add monolithic mount data.
633 */
634static int legacy_parse_monolithic(struct fs_context *fc, void *data)
635{
636 struct legacy_fs_context *ctx = fc->fs_private;
637
638 if (ctx->param_type != LEGACY_FS_UNSET_PARAMS) {
639 pr_warn("VFS: Can't mix monolithic and individual options\n");
640 return -EINVAL;
641 }
642
643 ctx->legacy_data = data;
644 ctx->param_type = LEGACY_FS_MONOLITHIC_PARAMS;
645 if (!ctx->legacy_data)
646 return 0;
647
648 if (fc->fs_type->fs_flags & FS_BINARY_MOUNTDATA)
649 return 0;
650 return security_sb_eat_lsm_opts(ctx->legacy_data, &fc->security);
651}
652
653/*
654 * Get a mountable root with the legacy mount command.
655 */
656static int legacy_get_tree(struct fs_context *fc)
657{
658 struct legacy_fs_context *ctx = fc->fs_private;
659 struct super_block *sb;
660 struct dentry *root;
661
662 root = fc->fs_type->mount(fc->fs_type, fc->sb_flags,
663 fc->source, ctx->legacy_data);
664 if (IS_ERR(root))
665 return PTR_ERR(root);
666
667 sb = root->d_sb;
668 BUG_ON(!sb);
669
670 fc->root = root;
671 return 0;
672}
673
674/*
675 * Handle remount.
676 */
677static int legacy_reconfigure(struct fs_context *fc)
678{
679 struct legacy_fs_context *ctx = fc->fs_private;
680 struct super_block *sb = fc->root->d_sb;
681
682 if (!sb->s_op->remount_fs)
683 return 0;
684
685 return sb->s_op->remount_fs(sb, &fc->sb_flags,
686 ctx ? ctx->legacy_data : NULL);
687}
688
689const struct fs_context_operations legacy_fs_context_ops = {
690 .free = legacy_fs_context_free,
691 .dup = legacy_fs_context_dup,
692 .parse_param = legacy_parse_param,
693 .parse_monolithic = legacy_parse_monolithic,
694 .get_tree = legacy_get_tree,
695 .reconfigure = legacy_reconfigure,
696};
697
698/*
699 * Initialise a legacy context for a filesystem that doesn't support
700 * fs_context.
701 */
702static int legacy_init_fs_context(struct fs_context *fc)
703{
704 fc->fs_private = kzalloc(sizeof(struct legacy_fs_context), GFP_KERNEL_ACCOUNT);
705 if (!fc->fs_private)
706 return -ENOMEM;
707 fc->ops = &legacy_fs_context_ops;
708 return 0;
709}
710
711int parse_monolithic_mount_data(struct fs_context *fc, void *data)
712{
713 int (*monolithic_mount_data)(struct fs_context *, void *);
714
715 monolithic_mount_data = fc->ops->parse_monolithic;
716 if (!monolithic_mount_data)
717 monolithic_mount_data = generic_parse_monolithic;
718
719 return monolithic_mount_data(fc, data);
720}
721
722/*
723 * Clean up a context after performing an action on it and put it into a state
724 * from where it can be used to reconfigure a superblock.
725 *
726 * Note that here we do only the parts that can't fail; the rest is in
727 * finish_clean_context() below and in between those fs_context is marked
728 * FS_CONTEXT_AWAITING_RECONF. The reason for splitup is that after
729 * successful mount or remount we need to report success to userland.
730 * Trying to do full reinit (for the sake of possible subsequent remount)
731 * and failing to allocate memory would've put us into a nasty situation.
732 * So here we only discard the old state and reinitialization is left
733 * until we actually try to reconfigure.
734 */
735void vfs_clean_context(struct fs_context *fc)
736{
737 if (fc->need_free && fc->ops && fc->ops->free)
738 fc->ops->free(fc);
739 fc->need_free = false;
740 fc->fs_private = NULL;
741 fc->s_fs_info = NULL;
742 fc->sb_flags = 0;
743 security_free_mnt_opts(&fc->security);
744 kfree(fc->source);
745 fc->source = NULL;
746 fc->exclusive = false;
747
748 fc->purpose = FS_CONTEXT_FOR_RECONFIGURE;
749 fc->phase = FS_CONTEXT_AWAITING_RECONF;
750}
751
752int finish_clean_context(struct fs_context *fc)
753{
754 int error;
755
756 if (fc->phase != FS_CONTEXT_AWAITING_RECONF)
757 return 0;
758
759 if (fc->fs_type->init_fs_context)
760 error = fc->fs_type->init_fs_context(fc);
761 else
762 error = legacy_init_fs_context(fc);
763 if (unlikely(error)) {
764 fc->phase = FS_CONTEXT_FAILED;
765 return error;
766 }
767 fc->need_free = true;
768 fc->phase = FS_CONTEXT_RECONF_PARAMS;
769 return 0;
770}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* Provide a way to create a superblock configuration context within the kernel
3 * that allows a superblock to be set up prior to mounting.
4 *
5 * Copyright (C) 2017 Red Hat, Inc. All Rights Reserved.
6 * Written by David Howells (dhowells@redhat.com)
7 */
8
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10#include <linux/module.h>
11#include <linux/fs_context.h>
12#include <linux/fs_parser.h>
13#include <linux/fs.h>
14#include <linux/mount.h>
15#include <linux/nsproxy.h>
16#include <linux/slab.h>
17#include <linux/magic.h>
18#include <linux/security.h>
19#include <linux/mnt_namespace.h>
20#include <linux/pid_namespace.h>
21#include <linux/user_namespace.h>
22#include <net/net_namespace.h>
23#include <asm/sections.h>
24#include "mount.h"
25#include "internal.h"
26
27enum legacy_fs_param {
28 LEGACY_FS_UNSET_PARAMS,
29 LEGACY_FS_MONOLITHIC_PARAMS,
30 LEGACY_FS_INDIVIDUAL_PARAMS,
31};
32
33struct legacy_fs_context {
34 char *legacy_data; /* Data page for legacy filesystems */
35 size_t data_size;
36 enum legacy_fs_param param_type;
37};
38
39static int legacy_init_fs_context(struct fs_context *fc);
40
41static const struct constant_table common_set_sb_flag[] = {
42 { "dirsync", SB_DIRSYNC },
43 { "lazytime", SB_LAZYTIME },
44 { "mand", SB_MANDLOCK },
45 { "ro", SB_RDONLY },
46 { "sync", SB_SYNCHRONOUS },
47 { },
48};
49
50static const struct constant_table common_clear_sb_flag[] = {
51 { "async", SB_SYNCHRONOUS },
52 { "nolazytime", SB_LAZYTIME },
53 { "nomand", SB_MANDLOCK },
54 { "rw", SB_RDONLY },
55 { },
56};
57
58/*
59 * Check for a common mount option that manipulates s_flags.
60 */
61static int vfs_parse_sb_flag(struct fs_context *fc, const char *key)
62{
63 unsigned int token;
64
65 token = lookup_constant(common_set_sb_flag, key, 0);
66 if (token) {
67 fc->sb_flags |= token;
68 fc->sb_flags_mask |= token;
69 return 0;
70 }
71
72 token = lookup_constant(common_clear_sb_flag, key, 0);
73 if (token) {
74 fc->sb_flags &= ~token;
75 fc->sb_flags_mask |= token;
76 return 0;
77 }
78
79 return -ENOPARAM;
80}
81
82/**
83 * vfs_parse_fs_param_source - Handle setting "source" via parameter
84 * @fc: The filesystem context to modify
85 * @param: The parameter
86 *
87 * This is a simple helper for filesystems to verify that the "source" they
88 * accept is sane.
89 *
90 * Returns 0 on success, -ENOPARAM if this is not "source" parameter, and
91 * -EINVAL otherwise. In the event of failure, supplementary error information
92 * is logged.
93 */
94int vfs_parse_fs_param_source(struct fs_context *fc, struct fs_parameter *param)
95{
96 if (strcmp(param->key, "source") != 0)
97 return -ENOPARAM;
98
99 if (param->type != fs_value_is_string)
100 return invalf(fc, "Non-string source");
101
102 if (fc->source)
103 return invalf(fc, "Multiple sources");
104
105 fc->source = param->string;
106 param->string = NULL;
107 return 0;
108}
109EXPORT_SYMBOL(vfs_parse_fs_param_source);
110
111/**
112 * vfs_parse_fs_param - Add a single parameter to a superblock config
113 * @fc: The filesystem context to modify
114 * @param: The parameter
115 *
116 * A single mount option in string form is applied to the filesystem context
117 * being set up. Certain standard options (for example "ro") are translated
118 * into flag bits without going to the filesystem. The active security module
119 * is allowed to observe and poach options. Any other options are passed over
120 * to the filesystem to parse.
121 *
122 * This may be called multiple times for a context.
123 *
124 * Returns 0 on success and a negative error code on failure. In the event of
125 * failure, supplementary error information may have been set.
126 */
127int vfs_parse_fs_param(struct fs_context *fc, struct fs_parameter *param)
128{
129 int ret;
130
131 if (!param->key)
132 return invalf(fc, "Unnamed parameter\n");
133
134 ret = vfs_parse_sb_flag(fc, param->key);
135 if (ret != -ENOPARAM)
136 return ret;
137
138 ret = security_fs_context_parse_param(fc, param);
139 if (ret != -ENOPARAM)
140 /* Param belongs to the LSM or is disallowed by the LSM; so
141 * don't pass to the FS.
142 */
143 return ret;
144
145 if (fc->ops->parse_param) {
146 ret = fc->ops->parse_param(fc, param);
147 if (ret != -ENOPARAM)
148 return ret;
149 }
150
151 /* If the filesystem doesn't take any arguments, give it the
152 * default handling of source.
153 */
154 ret = vfs_parse_fs_param_source(fc, param);
155 if (ret != -ENOPARAM)
156 return ret;
157
158 return invalf(fc, "%s: Unknown parameter '%s'",
159 fc->fs_type->name, param->key);
160}
161EXPORT_SYMBOL(vfs_parse_fs_param);
162
163/**
164 * vfs_parse_fs_string - Convenience function to just parse a string.
165 */
166int vfs_parse_fs_string(struct fs_context *fc, const char *key,
167 const char *value, size_t v_size)
168{
169 int ret;
170
171 struct fs_parameter param = {
172 .key = key,
173 .type = fs_value_is_flag,
174 .size = v_size,
175 };
176
177 if (value) {
178 param.string = kmemdup_nul(value, v_size, GFP_KERNEL);
179 if (!param.string)
180 return -ENOMEM;
181 param.type = fs_value_is_string;
182 }
183
184 ret = vfs_parse_fs_param(fc, ¶m);
185 kfree(param.string);
186 return ret;
187}
188EXPORT_SYMBOL(vfs_parse_fs_string);
189
190/**
191 * generic_parse_monolithic - Parse key[=val][,key[=val]]* mount data
192 * @ctx: The superblock configuration to fill in.
193 * @data: The data to parse
194 *
195 * Parse a blob of data that's in key[=val][,key[=val]]* form. This can be
196 * called from the ->monolithic_mount_data() fs_context operation.
197 *
198 * Returns 0 on success or the error returned by the ->parse_option() fs_context
199 * operation on failure.
200 */
201int generic_parse_monolithic(struct fs_context *fc, void *data)
202{
203 char *options = data, *key;
204 int ret = 0;
205
206 if (!options)
207 return 0;
208
209 ret = security_sb_eat_lsm_opts(options, &fc->security);
210 if (ret)
211 return ret;
212
213 while ((key = strsep(&options, ",")) != NULL) {
214 if (*key) {
215 size_t v_len = 0;
216 char *value = strchr(key, '=');
217
218 if (value) {
219 if (value == key)
220 continue;
221 *value++ = 0;
222 v_len = strlen(value);
223 }
224 ret = vfs_parse_fs_string(fc, key, value, v_len);
225 if (ret < 0)
226 break;
227 }
228 }
229
230 return ret;
231}
232EXPORT_SYMBOL(generic_parse_monolithic);
233
234/**
235 * alloc_fs_context - Create a filesystem context.
236 * @fs_type: The filesystem type.
237 * @reference: The dentry from which this one derives (or NULL)
238 * @sb_flags: Filesystem/superblock flags (SB_*)
239 * @sb_flags_mask: Applicable members of @sb_flags
240 * @purpose: The purpose that this configuration shall be used for.
241 *
242 * Open a filesystem and create a mount context. The mount context is
243 * initialised with the supplied flags and, if a submount/automount from
244 * another superblock (referred to by @reference) is supplied, may have
245 * parameters such as namespaces copied across from that superblock.
246 */
247static struct fs_context *alloc_fs_context(struct file_system_type *fs_type,
248 struct dentry *reference,
249 unsigned int sb_flags,
250 unsigned int sb_flags_mask,
251 enum fs_context_purpose purpose)
252{
253 int (*init_fs_context)(struct fs_context *);
254 struct fs_context *fc;
255 int ret = -ENOMEM;
256
257 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL);
258 if (!fc)
259 return ERR_PTR(-ENOMEM);
260
261 fc->purpose = purpose;
262 fc->sb_flags = sb_flags;
263 fc->sb_flags_mask = sb_flags_mask;
264 fc->fs_type = get_filesystem(fs_type);
265 fc->cred = get_current_cred();
266 fc->net_ns = get_net(current->nsproxy->net_ns);
267 fc->log.prefix = fs_type->name;
268
269 mutex_init(&fc->uapi_mutex);
270
271 switch (purpose) {
272 case FS_CONTEXT_FOR_MOUNT:
273 fc->user_ns = get_user_ns(fc->cred->user_ns);
274 break;
275 case FS_CONTEXT_FOR_SUBMOUNT:
276 fc->user_ns = get_user_ns(reference->d_sb->s_user_ns);
277 break;
278 case FS_CONTEXT_FOR_RECONFIGURE:
279 atomic_inc(&reference->d_sb->s_active);
280 fc->user_ns = get_user_ns(reference->d_sb->s_user_ns);
281 fc->root = dget(reference);
282 break;
283 }
284
285 /* TODO: Make all filesystems support this unconditionally */
286 init_fs_context = fc->fs_type->init_fs_context;
287 if (!init_fs_context)
288 init_fs_context = legacy_init_fs_context;
289
290 ret = init_fs_context(fc);
291 if (ret < 0)
292 goto err_fc;
293 fc->need_free = true;
294 return fc;
295
296err_fc:
297 put_fs_context(fc);
298 return ERR_PTR(ret);
299}
300
301struct fs_context *fs_context_for_mount(struct file_system_type *fs_type,
302 unsigned int sb_flags)
303{
304 return alloc_fs_context(fs_type, NULL, sb_flags, 0,
305 FS_CONTEXT_FOR_MOUNT);
306}
307EXPORT_SYMBOL(fs_context_for_mount);
308
309struct fs_context *fs_context_for_reconfigure(struct dentry *dentry,
310 unsigned int sb_flags,
311 unsigned int sb_flags_mask)
312{
313 return alloc_fs_context(dentry->d_sb->s_type, dentry, sb_flags,
314 sb_flags_mask, FS_CONTEXT_FOR_RECONFIGURE);
315}
316EXPORT_SYMBOL(fs_context_for_reconfigure);
317
318struct fs_context *fs_context_for_submount(struct file_system_type *type,
319 struct dentry *reference)
320{
321 return alloc_fs_context(type, reference, 0, 0, FS_CONTEXT_FOR_SUBMOUNT);
322}
323EXPORT_SYMBOL(fs_context_for_submount);
324
325void fc_drop_locked(struct fs_context *fc)
326{
327 struct super_block *sb = fc->root->d_sb;
328 dput(fc->root);
329 fc->root = NULL;
330 deactivate_locked_super(sb);
331}
332
333static void legacy_fs_context_free(struct fs_context *fc);
334
335/**
336 * vfs_dup_fc_config: Duplicate a filesystem context.
337 * @src_fc: The context to copy.
338 */
339struct fs_context *vfs_dup_fs_context(struct fs_context *src_fc)
340{
341 struct fs_context *fc;
342 int ret;
343
344 if (!src_fc->ops->dup)
345 return ERR_PTR(-EOPNOTSUPP);
346
347 fc = kmemdup(src_fc, sizeof(struct fs_context), GFP_KERNEL);
348 if (!fc)
349 return ERR_PTR(-ENOMEM);
350
351 mutex_init(&fc->uapi_mutex);
352
353 fc->fs_private = NULL;
354 fc->s_fs_info = NULL;
355 fc->source = NULL;
356 fc->security = NULL;
357 get_filesystem(fc->fs_type);
358 get_net(fc->net_ns);
359 get_user_ns(fc->user_ns);
360 get_cred(fc->cred);
361 if (fc->log.log)
362 refcount_inc(&fc->log.log->usage);
363
364 /* Can't call put until we've called ->dup */
365 ret = fc->ops->dup(fc, src_fc);
366 if (ret < 0)
367 goto err_fc;
368
369 ret = security_fs_context_dup(fc, src_fc);
370 if (ret < 0)
371 goto err_fc;
372 return fc;
373
374err_fc:
375 put_fs_context(fc);
376 return ERR_PTR(ret);
377}
378EXPORT_SYMBOL(vfs_dup_fs_context);
379
380/**
381 * logfc - Log a message to a filesystem context
382 * @fc: The filesystem context to log to.
383 * @fmt: The format of the buffer.
384 */
385void logfc(struct fc_log *log, const char *prefix, char level, const char *fmt, ...)
386{
387 va_list va;
388 struct va_format vaf = {.fmt = fmt, .va = &va};
389
390 va_start(va, fmt);
391 if (!log) {
392 switch (level) {
393 case 'w':
394 printk(KERN_WARNING "%s%s%pV\n", prefix ? prefix : "",
395 prefix ? ": " : "", &vaf);
396 break;
397 case 'e':
398 printk(KERN_ERR "%s%s%pV\n", prefix ? prefix : "",
399 prefix ? ": " : "", &vaf);
400 break;
401 default:
402 printk(KERN_NOTICE "%s%s%pV\n", prefix ? prefix : "",
403 prefix ? ": " : "", &vaf);
404 break;
405 }
406 } else {
407 unsigned int logsize = ARRAY_SIZE(log->buffer);
408 u8 index;
409 char *q = kasprintf(GFP_KERNEL, "%c %s%s%pV\n", level,
410 prefix ? prefix : "",
411 prefix ? ": " : "", &vaf);
412
413 index = log->head & (logsize - 1);
414 BUILD_BUG_ON(sizeof(log->head) != sizeof(u8) ||
415 sizeof(log->tail) != sizeof(u8));
416 if ((u8)(log->head - log->tail) == logsize) {
417 /* The buffer is full, discard the oldest message */
418 if (log->need_free & (1 << index))
419 kfree(log->buffer[index]);
420 log->tail++;
421 }
422
423 log->buffer[index] = q ? q : "OOM: Can't store error string";
424 if (q)
425 log->need_free |= 1 << index;
426 else
427 log->need_free &= ~(1 << index);
428 log->head++;
429 }
430 va_end(va);
431}
432EXPORT_SYMBOL(logfc);
433
434/*
435 * Free a logging structure.
436 */
437static void put_fc_log(struct fs_context *fc)
438{
439 struct fc_log *log = fc->log.log;
440 int i;
441
442 if (log) {
443 if (refcount_dec_and_test(&log->usage)) {
444 fc->log.log = NULL;
445 for (i = 0; i <= 7; i++)
446 if (log->need_free & (1 << i))
447 kfree(log->buffer[i]);
448 kfree(log);
449 }
450 }
451}
452
453/**
454 * put_fs_context - Dispose of a superblock configuration context.
455 * @fc: The context to dispose of.
456 */
457void put_fs_context(struct fs_context *fc)
458{
459 struct super_block *sb;
460
461 if (fc->root) {
462 sb = fc->root->d_sb;
463 dput(fc->root);
464 fc->root = NULL;
465 deactivate_super(sb);
466 }
467
468 if (fc->need_free && fc->ops && fc->ops->free)
469 fc->ops->free(fc);
470
471 security_free_mnt_opts(&fc->security);
472 put_net(fc->net_ns);
473 put_user_ns(fc->user_ns);
474 put_cred(fc->cred);
475 put_fc_log(fc);
476 put_filesystem(fc->fs_type);
477 kfree(fc->source);
478 kfree(fc);
479}
480EXPORT_SYMBOL(put_fs_context);
481
482/*
483 * Free the config for a filesystem that doesn't support fs_context.
484 */
485static void legacy_fs_context_free(struct fs_context *fc)
486{
487 struct legacy_fs_context *ctx = fc->fs_private;
488
489 if (ctx) {
490 if (ctx->param_type == LEGACY_FS_INDIVIDUAL_PARAMS)
491 kfree(ctx->legacy_data);
492 kfree(ctx);
493 }
494}
495
496/*
497 * Duplicate a legacy config.
498 */
499static int legacy_fs_context_dup(struct fs_context *fc, struct fs_context *src_fc)
500{
501 struct legacy_fs_context *ctx;
502 struct legacy_fs_context *src_ctx = src_fc->fs_private;
503
504 ctx = kmemdup(src_ctx, sizeof(*src_ctx), GFP_KERNEL);
505 if (!ctx)
506 return -ENOMEM;
507
508 if (ctx->param_type == LEGACY_FS_INDIVIDUAL_PARAMS) {
509 ctx->legacy_data = kmemdup(src_ctx->legacy_data,
510 src_ctx->data_size, GFP_KERNEL);
511 if (!ctx->legacy_data) {
512 kfree(ctx);
513 return -ENOMEM;
514 }
515 }
516
517 fc->fs_private = ctx;
518 return 0;
519}
520
521/*
522 * Add a parameter to a legacy config. We build up a comma-separated list of
523 * options.
524 */
525static int legacy_parse_param(struct fs_context *fc, struct fs_parameter *param)
526{
527 struct legacy_fs_context *ctx = fc->fs_private;
528 unsigned int size = ctx->data_size;
529 size_t len = 0;
530 int ret;
531
532 ret = vfs_parse_fs_param_source(fc, param);
533 if (ret != -ENOPARAM)
534 return ret;
535
536 if (ctx->param_type == LEGACY_FS_MONOLITHIC_PARAMS)
537 return invalf(fc, "VFS: Legacy: Can't mix monolithic and individual options");
538
539 switch (param->type) {
540 case fs_value_is_string:
541 len = 1 + param->size;
542 fallthrough;
543 case fs_value_is_flag:
544 len += strlen(param->key);
545 break;
546 default:
547 return invalf(fc, "VFS: Legacy: Parameter type for '%s' not supported",
548 param->key);
549 }
550
551 if (len > PAGE_SIZE - 2 - size)
552 return invalf(fc, "VFS: Legacy: Cumulative options too large");
553 if (strchr(param->key, ',') ||
554 (param->type == fs_value_is_string &&
555 memchr(param->string, ',', param->size)))
556 return invalf(fc, "VFS: Legacy: Option '%s' contained comma",
557 param->key);
558 if (!ctx->legacy_data) {
559 ctx->legacy_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
560 if (!ctx->legacy_data)
561 return -ENOMEM;
562 }
563
564 ctx->legacy_data[size++] = ',';
565 len = strlen(param->key);
566 memcpy(ctx->legacy_data + size, param->key, len);
567 size += len;
568 if (param->type == fs_value_is_string) {
569 ctx->legacy_data[size++] = '=';
570 memcpy(ctx->legacy_data + size, param->string, param->size);
571 size += param->size;
572 }
573 ctx->legacy_data[size] = '\0';
574 ctx->data_size = size;
575 ctx->param_type = LEGACY_FS_INDIVIDUAL_PARAMS;
576 return 0;
577}
578
579/*
580 * Add monolithic mount data.
581 */
582static int legacy_parse_monolithic(struct fs_context *fc, void *data)
583{
584 struct legacy_fs_context *ctx = fc->fs_private;
585
586 if (ctx->param_type != LEGACY_FS_UNSET_PARAMS) {
587 pr_warn("VFS: Can't mix monolithic and individual options\n");
588 return -EINVAL;
589 }
590
591 ctx->legacy_data = data;
592 ctx->param_type = LEGACY_FS_MONOLITHIC_PARAMS;
593 if (!ctx->legacy_data)
594 return 0;
595
596 if (fc->fs_type->fs_flags & FS_BINARY_MOUNTDATA)
597 return 0;
598 return security_sb_eat_lsm_opts(ctx->legacy_data, &fc->security);
599}
600
601/*
602 * Get a mountable root with the legacy mount command.
603 */
604static int legacy_get_tree(struct fs_context *fc)
605{
606 struct legacy_fs_context *ctx = fc->fs_private;
607 struct super_block *sb;
608 struct dentry *root;
609
610 root = fc->fs_type->mount(fc->fs_type, fc->sb_flags,
611 fc->source, ctx->legacy_data);
612 if (IS_ERR(root))
613 return PTR_ERR(root);
614
615 sb = root->d_sb;
616 BUG_ON(!sb);
617
618 fc->root = root;
619 return 0;
620}
621
622/*
623 * Handle remount.
624 */
625static int legacy_reconfigure(struct fs_context *fc)
626{
627 struct legacy_fs_context *ctx = fc->fs_private;
628 struct super_block *sb = fc->root->d_sb;
629
630 if (!sb->s_op->remount_fs)
631 return 0;
632
633 return sb->s_op->remount_fs(sb, &fc->sb_flags,
634 ctx ? ctx->legacy_data : NULL);
635}
636
637const struct fs_context_operations legacy_fs_context_ops = {
638 .free = legacy_fs_context_free,
639 .dup = legacy_fs_context_dup,
640 .parse_param = legacy_parse_param,
641 .parse_monolithic = legacy_parse_monolithic,
642 .get_tree = legacy_get_tree,
643 .reconfigure = legacy_reconfigure,
644};
645
646/*
647 * Initialise a legacy context for a filesystem that doesn't support
648 * fs_context.
649 */
650static int legacy_init_fs_context(struct fs_context *fc)
651{
652 fc->fs_private = kzalloc(sizeof(struct legacy_fs_context), GFP_KERNEL);
653 if (!fc->fs_private)
654 return -ENOMEM;
655 fc->ops = &legacy_fs_context_ops;
656 return 0;
657}
658
659int parse_monolithic_mount_data(struct fs_context *fc, void *data)
660{
661 int (*monolithic_mount_data)(struct fs_context *, void *);
662
663 monolithic_mount_data = fc->ops->parse_monolithic;
664 if (!monolithic_mount_data)
665 monolithic_mount_data = generic_parse_monolithic;
666
667 return monolithic_mount_data(fc, data);
668}
669
670/*
671 * Clean up a context after performing an action on it and put it into a state
672 * from where it can be used to reconfigure a superblock.
673 *
674 * Note that here we do only the parts that can't fail; the rest is in
675 * finish_clean_context() below and in between those fs_context is marked
676 * FS_CONTEXT_AWAITING_RECONF. The reason for splitup is that after
677 * successful mount or remount we need to report success to userland.
678 * Trying to do full reinit (for the sake of possible subsequent remount)
679 * and failing to allocate memory would've put us into a nasty situation.
680 * So here we only discard the old state and reinitialization is left
681 * until we actually try to reconfigure.
682 */
683void vfs_clean_context(struct fs_context *fc)
684{
685 if (fc->need_free && fc->ops && fc->ops->free)
686 fc->ops->free(fc);
687 fc->need_free = false;
688 fc->fs_private = NULL;
689 fc->s_fs_info = NULL;
690 fc->sb_flags = 0;
691 security_free_mnt_opts(&fc->security);
692 kfree(fc->source);
693 fc->source = NULL;
694
695 fc->purpose = FS_CONTEXT_FOR_RECONFIGURE;
696 fc->phase = FS_CONTEXT_AWAITING_RECONF;
697}
698
699int finish_clean_context(struct fs_context *fc)
700{
701 int error;
702
703 if (fc->phase != FS_CONTEXT_AWAITING_RECONF)
704 return 0;
705
706 if (fc->fs_type->init_fs_context)
707 error = fc->fs_type->init_fs_context(fc);
708 else
709 error = legacy_init_fs_context(fc);
710 if (unlikely(error)) {
711 fc->phase = FS_CONTEXT_FAILED;
712 return error;
713 }
714 fc->need_free = true;
715 fc->phase = FS_CONTEXT_RECONF_PARAMS;
716 return 0;
717}