blob: 46966fd8bcf9f042e85d0b66134e59fbef83abfd [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * fs/libfs.c
4 * Library for filesystems writers.
5 */
6
Fabian Frederickac13a822014-06-04 16:06:27 -07007#include <linux/blkdev.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -05008#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/pagemap.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090010#include <linux/slab.h>
Ingo Molnar5b825c32017-02-02 17:54:15 +010011#include <linux/cred.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/mount.h>
13#include <linux/vfs.h>
npiggin@suse.de7bb46a62010-05-27 01:05:33 +100014#include <linux/quotaops.h>
Ingo Molnar7cf34c72006-03-23 03:00:36 -080015#include <linux/mutex.h>
Al Viro87dc8002013-09-16 10:30:04 -040016#include <linux/namei.h>
Christoph Hellwig25961102007-10-21 16:42:05 -070017#include <linux/exportfs.h>
Andrew Morton5ca14832022-09-09 13:57:41 -070018#include <linux/iversion.h>
Al Virod5aacad2009-06-07 14:56:44 -040019#include <linux/writeback.h>
Al Viroff01bb42011-09-16 02:31:11 -040020#include <linux/buffer_head.h> /* sync_mapping_buffers */
David Howells31d6d5c2019-03-25 16:38:23 +000021#include <linux/fs_context.h>
22#include <linux/pseudo_fs.h>
Al Viroa3d1e7e2019-11-18 09:43:10 -050023#include <linux/fsnotify.h>
Daniel Rosenbergc8438432020-07-08 02:12:35 -070024#include <linux/unicode.h>
25#include <linux/fscrypt.h>
Christian Braunerb28ddcc2024-02-19 16:30:57 +010026#include <linux/pidfs.h>
Ingo Molnar7cf34c72006-03-23 03:00:36 -080027
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080028#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Al Viroa4464db2011-07-07 15:03:58 -040030#include "internal.h"
31
Christian Braunerb74d24f2023-01-13 12:49:12 +010032int simple_getattr(struct mnt_idmap *idmap, const struct path *path,
Christian Brauner549c7292021-01-21 14:19:43 +010033 struct kstat *stat, u32 request_mask,
34 unsigned int query_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070035{
David Howellsa528d352017-01-31 16:46:22 +000036 struct inode *inode = d_inode(path->dentry);
Jeff Layton0d72b9282023-08-07 15:38:33 -040037 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030038 stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 return 0;
40}
Al Viro12f38872013-09-15 21:20:49 -040041EXPORT_SYMBOL(simple_getattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
David Howells726c3342006-06-23 02:02:58 -070043int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070044{
Amir Goldsteinae62bcb2023-10-23 17:30:49 +030045 u64 id = huge_encode_dev(dentry->d_sb->s_dev);
46
47 buf->f_fsid = u64_to_fsid(id);
David Howells726c3342006-06-23 02:02:58 -070048 buf->f_type = dentry->d_sb->s_magic;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030049 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 buf->f_namelen = NAME_MAX;
51 return 0;
52}
Al Viro12f38872013-09-15 21:20:49 -040053EXPORT_SYMBOL(simple_statfs);
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55/*
56 * Retaining negative dentries for an in-memory filesystem just wastes
57 * memory and lookup time: arrange for them to be deleted immediately.
58 */
Al Virob26d4cd2013-10-25 18:47:37 -040059int always_delete_dentry(const struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070060{
61 return 1;
62}
Al Virob26d4cd2013-10-25 18:47:37 -040063EXPORT_SYMBOL(always_delete_dentry);
64
65const struct dentry_operations simple_dentry_operations = {
66 .d_delete = always_delete_dentry,
67};
68EXPORT_SYMBOL(simple_dentry_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
70/*
71 * Lookup the data. This is trivial - if the dentry didn't already
72 * exist, we know it is negative. Set d_op to delete negative dentries.
73 */
Al Viro00cd8dd2012-06-10 17:13:09 -040074struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 if (dentry->d_name.len > NAME_MAX)
77 return ERR_PTR(-ENAMETOOLONG);
Al Viro74931da2013-07-14 17:43:25 +040078 if (!dentry->d_sb->s_d_op)
79 d_set_d_op(dentry, &simple_dentry_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 d_add(dentry, NULL);
81 return NULL;
82}
Al Viro12f38872013-09-15 21:20:49 -040083EXPORT_SYMBOL(simple_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Linus Torvalds1da177e2005-04-16 15:20:36 -070085int dcache_dir_open(struct inode *inode, struct file *file)
86{
Al Viroba65dc52016-06-10 11:32:47 -040087 file->private_data = d_alloc_cursor(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
89 return file->private_data ? 0 : -ENOMEM;
90}
Al Viro12f38872013-09-15 21:20:49 -040091EXPORT_SYMBOL(dcache_dir_open);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
93int dcache_dir_close(struct inode *inode, struct file *file)
94{
95 dput(file->private_data);
96 return 0;
97}
Al Viro12f38872013-09-15 21:20:49 -040098EXPORT_SYMBOL(dcache_dir_close);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Al Viro4f42c1b2016-06-06 19:37:13 -0400100/* parent is locked at least shared */
Al Virod4f4de52019-09-15 12:12:39 -0400101/*
102 * Returns an element of siblings' list.
103 * We are looking for <count>th positive after <p>; if
Al Viro26b6c982019-09-20 16:32:42 -0400104 * found, dentry is grabbed and returned to caller.
105 * If no such element exists, NULL is returned.
Al Virod4f4de52019-09-15 12:12:39 -0400106 */
Al Viro26b6c982019-09-20 16:32:42 -0400107static struct dentry *scan_positives(struct dentry *cursor,
Al Viroda549bd2023-11-07 02:00:39 -0500108 struct hlist_node **p,
Al Virod4f4de52019-09-15 12:12:39 -0400109 loff_t count,
Al Viro26b6c982019-09-20 16:32:42 -0400110 struct dentry *last)
Al Viro4f42c1b2016-06-06 19:37:13 -0400111{
Al Virod4f4de52019-09-15 12:12:39 -0400112 struct dentry *dentry = cursor->d_parent, *found = NULL;
Al Viro4f42c1b2016-06-06 19:37:13 -0400113
Al Virod4f4de52019-09-15 12:12:39 -0400114 spin_lock(&dentry->d_lock);
Al Viroda549bd2023-11-07 02:00:39 -0500115 while (*p) {
116 struct dentry *d = hlist_entry(*p, struct dentry, d_sib);
117 p = &d->d_sib.next;
Al Virod4f4de52019-09-15 12:12:39 -0400118 // we must at least skip cursors, to avoid livelocks
119 if (d->d_flags & DCACHE_DENTRY_CURSOR)
120 continue;
121 if (simple_positive(d) && !--count) {
122 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
123 if (simple_positive(d))
124 found = dget_dlock(d);
125 spin_unlock(&d->d_lock);
126 if (likely(found))
127 break;
128 count = 1;
129 }
130 if (need_resched()) {
Al Viroda549bd2023-11-07 02:00:39 -0500131 if (!hlist_unhashed(&cursor->d_sib))
132 __hlist_del(&cursor->d_sib);
133 hlist_add_behind(&cursor->d_sib, &d->d_sib);
134 p = &cursor->d_sib.next;
Al Virod4f4de52019-09-15 12:12:39 -0400135 spin_unlock(&dentry->d_lock);
136 cond_resched();
137 spin_lock(&dentry->d_lock);
Al Viro4f42c1b2016-06-06 19:37:13 -0400138 }
139 }
Al Virod4f4de52019-09-15 12:12:39 -0400140 spin_unlock(&dentry->d_lock);
Al Viro26b6c982019-09-20 16:32:42 -0400141 dput(last);
142 return found;
Al Viro4f42c1b2016-06-06 19:37:13 -0400143}
144
Andrew Morton965c8e52012-12-17 15:59:39 -0800145loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100147 struct dentry *dentry = file->f_path.dentry;
Andrew Morton965c8e52012-12-17 15:59:39 -0800148 switch (whence) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 case 1:
150 offset += file->f_pos;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -0500151 fallthrough;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 case 0:
153 if (offset >= 0)
154 break;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -0500155 fallthrough;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 return -EINVAL;
158 }
159 if (offset != file->f_pos) {
Al Virod4f4de52019-09-15 12:12:39 -0400160 struct dentry *cursor = file->private_data;
161 struct dentry *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Al Virod4f4de52019-09-15 12:12:39 -0400163 inode_lock_shared(dentry->d_inode);
164
Al Viro26b6c982019-09-20 16:32:42 -0400165 if (offset > 2)
Al Viroda549bd2023-11-07 02:00:39 -0500166 to = scan_positives(cursor, &dentry->d_children.first,
Al Viro26b6c982019-09-20 16:32:42 -0400167 offset - 2, NULL);
168 spin_lock(&dentry->d_lock);
Al Viroda549bd2023-11-07 02:00:39 -0500169 hlist_del_init(&cursor->d_sib);
Al Viro26b6c982019-09-20 16:32:42 -0400170 if (to)
Al Viroda549bd2023-11-07 02:00:39 -0500171 hlist_add_behind(&cursor->d_sib, &to->d_sib);
Al Viro26b6c982019-09-20 16:32:42 -0400172 spin_unlock(&dentry->d_lock);
Al Virod4f4de52019-09-15 12:12:39 -0400173 dput(to);
174
Al Viro26b6c982019-09-20 16:32:42 -0400175 file->f_pos = offset;
176
Al Virod4f4de52019-09-15 12:12:39 -0400177 inode_unlock_shared(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 return offset;
180}
Al Viro12f38872013-09-15 21:20:49 -0400181EXPORT_SYMBOL(dcache_dir_lseek);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183/*
184 * Directory is locked and all positive dentries in it are safe, since
185 * for ramfs-type trees they can't go away without unlink() or rmdir(),
186 * both impossible due to the lock on directory.
187 */
188
Al Viro5f99f4e2013-05-15 20:23:06 -0400189int dcache_readdir(struct file *file, struct dir_context *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
Al Viro5f99f4e2013-05-15 20:23:06 -0400191 struct dentry *dentry = file->f_path.dentry;
192 struct dentry *cursor = file->private_data;
Al Virod4f4de52019-09-15 12:12:39 -0400193 struct dentry *next = NULL;
Al Viroda549bd2023-11-07 02:00:39 -0500194 struct hlist_node **p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Al Viro5f99f4e2013-05-15 20:23:06 -0400196 if (!dir_emit_dots(file, ctx))
197 return 0;
Al Viro4f42c1b2016-06-06 19:37:13 -0400198
Al Viro5f99f4e2013-05-15 20:23:06 -0400199 if (ctx->pos == 2)
Al Viroda549bd2023-11-07 02:00:39 -0500200 p = &dentry->d_children.first;
Al Viro26b6c982019-09-20 16:32:42 -0400201 else
Al Viroda549bd2023-11-07 02:00:39 -0500202 p = &cursor->d_sib.next;
Al Virod4f4de52019-09-15 12:12:39 -0400203
Al Viro26b6c982019-09-20 16:32:42 -0400204 while ((next = scan_positives(cursor, p, 1, next)) != NULL) {
Al Viro5f99f4e2013-05-15 20:23:06 -0400205 if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
Jeff Layton364595a2023-03-30 06:41:43 -0400206 d_inode(next)->i_ino,
207 fs_umode_to_dtype(d_inode(next)->i_mode)))
Al Viro4f42c1b2016-06-06 19:37:13 -0400208 break;
Al Viro5f99f4e2013-05-15 20:23:06 -0400209 ctx->pos++;
Al Viroda549bd2023-11-07 02:00:39 -0500210 p = &next->d_sib.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 }
Al Virod4f4de52019-09-15 12:12:39 -0400212 spin_lock(&dentry->d_lock);
Al Viroda549bd2023-11-07 02:00:39 -0500213 hlist_del_init(&cursor->d_sib);
Al Viro26b6c982019-09-20 16:32:42 -0400214 if (next)
Al Viroda549bd2023-11-07 02:00:39 -0500215 hlist_add_before(&cursor->d_sib, &next->d_sib);
Al Virod4f4de52019-09-15 12:12:39 -0400216 spin_unlock(&dentry->d_lock);
217 dput(next);
218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 return 0;
220}
Al Viro12f38872013-09-15 21:20:49 -0400221EXPORT_SYMBOL(dcache_readdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
223ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
224{
225 return -EISDIR;
226}
Al Viro12f38872013-09-15 21:20:49 -0400227EXPORT_SYMBOL(generic_read_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800229const struct file_operations simple_dir_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 .open = dcache_dir_open,
231 .release = dcache_dir_close,
232 .llseek = dcache_dir_lseek,
233 .read = generic_read_dir,
Al Viro4e829012016-04-20 19:52:15 -0400234 .iterate_shared = dcache_readdir,
Christoph Hellwig1b061d92010-05-26 17:53:41 +0200235 .fsync = noop_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236};
Al Viro12f38872013-09-15 21:20:49 -0400237EXPORT_SYMBOL(simple_dir_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800239const struct inode_operations simple_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 .lookup = simple_lookup,
241};
Al Viro12f38872013-09-15 21:20:49 -0400242EXPORT_SYMBOL(simple_dir_inode_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Chuck Lever7beea722024-02-17 15:23:47 -0500244/* 0 is '.', 1 is '..', so always start with offset 2 or more */
245enum {
246 DIR_OFFSET_MIN = 2,
247};
248
Chuck Lever0e4a8622024-02-17 15:24:16 -0500249static void offset_set(struct dentry *dentry, long offset)
Chuck Lever6faddda692023-06-30 13:48:49 -0400250{
Chuck Lever0e4a8622024-02-17 15:24:16 -0500251 dentry->d_fsdata = (void *)offset;
Chuck Lever6faddda692023-06-30 13:48:49 -0400252}
253
Chuck Lever0e4a8622024-02-17 15:24:16 -0500254static long dentry2offset(struct dentry *dentry)
Chuck Lever6faddda692023-06-30 13:48:49 -0400255{
Chuck Lever0e4a8622024-02-17 15:24:16 -0500256 return (long)dentry->d_fsdata;
Chuck Lever6faddda692023-06-30 13:48:49 -0400257}
258
Chuck Lever0e4a8622024-02-17 15:24:16 -0500259static struct lock_class_key simple_offset_lock_class;
Chuck Leverbbaef792023-07-24 10:43:57 -0400260
Chuck Lever6faddda692023-06-30 13:48:49 -0400261/**
262 * simple_offset_init - initialize an offset_ctx
263 * @octx: directory offset map to be initialized
264 *
265 */
266void simple_offset_init(struct offset_ctx *octx)
267{
Chuck Lever0e4a8622024-02-17 15:24:16 -0500268 mt_init_flags(&octx->mt, MT_FLAGS_ALLOC_RANGE);
269 lockdep_set_class(&octx->mt.ma_lock, &simple_offset_lock_class);
Chuck Lever7beea722024-02-17 15:23:47 -0500270 octx->next_offset = DIR_OFFSET_MIN;
Chuck Lever6faddda692023-06-30 13:48:49 -0400271}
272
273/**
274 * simple_offset_add - Add an entry to a directory's offset map
275 * @octx: directory offset ctx to be updated
276 * @dentry: new dentry being added
277 *
Chuck Lever0e4a8622024-02-17 15:24:16 -0500278 * Returns zero on success. @octx and the dentry's offset are updated.
Chuck Lever6faddda692023-06-30 13:48:49 -0400279 * Otherwise, a negative errno value is returned.
280 */
281int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
282{
Chuck Lever0e4a8622024-02-17 15:24:16 -0500283 unsigned long offset;
Chuck Lever6faddda692023-06-30 13:48:49 -0400284 int ret;
285
286 if (dentry2offset(dentry) != 0)
287 return -EBUSY;
288
Chuck Lever0e4a8622024-02-17 15:24:16 -0500289 ret = mtree_alloc_cyclic(&octx->mt, &offset, dentry, DIR_OFFSET_MIN,
290 LONG_MAX, &octx->next_offset, GFP_KERNEL);
Chuck Lever6faddda692023-06-30 13:48:49 -0400291 if (ret < 0)
292 return ret;
293
294 offset_set(dentry, offset);
295 return 0;
296}
297
Chuck Lever23cdd0e2024-04-15 11:20:54 -0400298static int simple_offset_replace(struct offset_ctx *octx, struct dentry *dentry,
299 long offset)
300{
301 int ret;
302
303 ret = mtree_store(&octx->mt, offset, dentry, GFP_KERNEL);
304 if (ret)
305 return ret;
306 offset_set(dentry, offset);
307 return 0;
308}
309
Chuck Lever6faddda692023-06-30 13:48:49 -0400310/**
311 * simple_offset_remove - Remove an entry to a directory's offset map
312 * @octx: directory offset ctx to be updated
313 * @dentry: dentry being removed
314 *
315 */
316void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry)
317{
Chuck Lever0e4a8622024-02-17 15:24:16 -0500318 long offset;
Chuck Lever6faddda692023-06-30 13:48:49 -0400319
320 offset = dentry2offset(dentry);
321 if (offset == 0)
322 return;
323
Chuck Lever0e4a8622024-02-17 15:24:16 -0500324 mtree_erase(&octx->mt, offset);
Chuck Lever6faddda692023-06-30 13:48:49 -0400325 offset_set(dentry, 0);
326}
327
328/**
Chuck Leverecba88a2024-02-17 15:23:54 -0500329 * simple_offset_empty - Check if a dentry can be unlinked
330 * @dentry: dentry to be tested
331 *
332 * Returns 0 if @dentry is a non-empty directory; otherwise returns 1.
333 */
334int simple_offset_empty(struct dentry *dentry)
335{
336 struct inode *inode = d_inode(dentry);
337 struct offset_ctx *octx;
338 struct dentry *child;
339 unsigned long index;
340 int ret = 1;
341
342 if (!inode || !S_ISDIR(inode->i_mode))
343 return ret;
344
345 index = DIR_OFFSET_MIN;
346 octx = inode->i_op->get_offset_ctx(inode);
Chuck Lever0e4a8622024-02-17 15:24:16 -0500347 mt_for_each(&octx->mt, child, index, LONG_MAX) {
Chuck Leverecba88a2024-02-17 15:23:54 -0500348 spin_lock(&child->d_lock);
349 if (simple_positive(child)) {
350 spin_unlock(&child->d_lock);
351 ret = 0;
352 break;
353 }
354 spin_unlock(&child->d_lock);
355 }
356
357 return ret;
358}
359
360/**
Chuck Lever5a1a25b2024-04-15 11:20:55 -0400361 * simple_offset_rename - handle directory offsets for rename
362 * @old_dir: parent directory of source entry
363 * @old_dentry: dentry of source entry
364 * @new_dir: parent_directory of destination entry
365 * @new_dentry: dentry of destination
366 *
367 * Caller provides appropriate serialization.
368 *
Chuck Leverad191eb2024-04-15 11:20:56 -0400369 * User space expects the directory offset value of the replaced
370 * (new) directory entry to be unchanged after a rename.
371 *
Chuck Lever5a1a25b2024-04-15 11:20:55 -0400372 * Returns zero on success, a negative errno value on failure.
373 */
374int simple_offset_rename(struct inode *old_dir, struct dentry *old_dentry,
375 struct inode *new_dir, struct dentry *new_dentry)
376{
377 struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
378 struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
Chuck Leverad191eb2024-04-15 11:20:56 -0400379 long new_offset = dentry2offset(new_dentry);
Chuck Lever5a1a25b2024-04-15 11:20:55 -0400380
381 simple_offset_remove(old_ctx, old_dentry);
Chuck Leverad191eb2024-04-15 11:20:56 -0400382
383 if (new_offset) {
384 offset_set(new_dentry, 0);
385 return simple_offset_replace(new_ctx, old_dentry, new_offset);
386 }
Chuck Lever5a1a25b2024-04-15 11:20:55 -0400387 return simple_offset_add(new_ctx, old_dentry);
388}
389
390/**
Chuck Lever6faddda692023-06-30 13:48:49 -0400391 * simple_offset_rename_exchange - exchange rename with directory offsets
392 * @old_dir: parent of dentry being moved
393 * @old_dentry: dentry being moved
394 * @new_dir: destination parent
395 * @new_dentry: destination dentry
396 *
Chuck Lever23cdd0e2024-04-15 11:20:54 -0400397 * This API preserves the directory offset values. Caller provides
398 * appropriate serialization.
399 *
Chuck Lever6faddda692023-06-30 13:48:49 -0400400 * Returns zero on success. Otherwise a negative errno is returned and the
401 * rename is rolled back.
402 */
403int simple_offset_rename_exchange(struct inode *old_dir,
404 struct dentry *old_dentry,
405 struct inode *new_dir,
406 struct dentry *new_dentry)
407{
408 struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
409 struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
Chuck Lever0e4a8622024-02-17 15:24:16 -0500410 long old_index = dentry2offset(old_dentry);
411 long new_index = dentry2offset(new_dentry);
Chuck Lever6faddda692023-06-30 13:48:49 -0400412 int ret;
413
414 simple_offset_remove(old_ctx, old_dentry);
415 simple_offset_remove(new_ctx, new_dentry);
416
Chuck Lever23cdd0e2024-04-15 11:20:54 -0400417 ret = simple_offset_replace(new_ctx, old_dentry, new_index);
Chuck Lever6faddda692023-06-30 13:48:49 -0400418 if (ret)
419 goto out_restore;
420
Chuck Lever23cdd0e2024-04-15 11:20:54 -0400421 ret = simple_offset_replace(old_ctx, new_dentry, old_index);
Chuck Lever6faddda692023-06-30 13:48:49 -0400422 if (ret) {
423 simple_offset_remove(new_ctx, old_dentry);
424 goto out_restore;
425 }
426
427 ret = simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
428 if (ret) {
429 simple_offset_remove(new_ctx, old_dentry);
430 simple_offset_remove(old_ctx, new_dentry);
431 goto out_restore;
432 }
433 return 0;
434
435out_restore:
Chuck Lever23cdd0e2024-04-15 11:20:54 -0400436 (void)simple_offset_replace(old_ctx, old_dentry, old_index);
437 (void)simple_offset_replace(new_ctx, new_dentry, new_index);
Chuck Lever6faddda692023-06-30 13:48:49 -0400438 return ret;
439}
440
441/**
442 * simple_offset_destroy - Release offset map
443 * @octx: directory offset ctx that is about to be destroyed
444 *
445 * During fs teardown (eg. umount), a directory's offset map might still
446 * contain entries. xa_destroy() cleans out anything that remains.
447 */
448void simple_offset_destroy(struct offset_ctx *octx)
449{
Chuck Lever0e4a8622024-02-17 15:24:16 -0500450 mtree_destroy(&octx->mt);
Chuck Lever6faddda692023-06-30 13:48:49 -0400451}
452
yangerkun64a7ce72024-07-31 12:38:35 +0800453static int offset_dir_open(struct inode *inode, struct file *file)
454{
455 struct offset_ctx *ctx = inode->i_op->get_offset_ctx(inode);
456
457 file->private_data = (void *)ctx->next_offset;
458 return 0;
459}
460
Chuck Lever6faddda692023-06-30 13:48:49 -0400461/**
462 * offset_dir_llseek - Advance the read position of a directory descriptor
463 * @file: an open directory whose position is to be updated
464 * @offset: a byte offset
465 * @whence: enumerator describing the starting position for this update
466 *
467 * SEEK_END, SEEK_DATA, and SEEK_HOLE are not supported for directories.
468 *
469 * Returns the updated read position if successful; otherwise a
470 * negative errno is returned and the read position remains unchanged.
471 */
472static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
473{
yangerkun64a7ce72024-07-31 12:38:35 +0800474 struct inode *inode = file->f_inode;
475 struct offset_ctx *ctx = inode->i_op->get_offset_ctx(inode);
476
Chuck Lever6faddda692023-06-30 13:48:49 -0400477 switch (whence) {
478 case SEEK_CUR:
479 offset += file->f_pos;
480 fallthrough;
481 case SEEK_SET:
482 if (offset >= 0)
483 break;
484 fallthrough;
485 default:
486 return -EINVAL;
487 }
488
Chuck Lever796432e2023-11-19 18:56:17 -0500489 /* In this case, ->private_data is protected by f_pos_lock */
yangerkun64a7ce72024-07-31 12:38:35 +0800490 if (!offset)
491 file->private_data = (void *)ctx->next_offset;
Chuck Lever0e4a8622024-02-17 15:24:16 -0500492 return vfs_setpos(file, offset, LONG_MAX);
Chuck Lever6faddda692023-06-30 13:48:49 -0400493}
494
Chuck Lever3f6d8102024-02-17 15:23:40 -0500495static struct dentry *offset_find_next(struct offset_ctx *octx, loff_t offset)
Chuck Lever6faddda692023-06-30 13:48:49 -0400496{
Chuck Lever0e4a8622024-02-17 15:24:16 -0500497 MA_STATE(mas, &octx->mt, offset, offset);
Chuck Lever6faddda692023-06-30 13:48:49 -0400498 struct dentry *child, *found = NULL;
499
500 rcu_read_lock();
Chuck Lever0e4a8622024-02-17 15:24:16 -0500501 child = mas_find(&mas, LONG_MAX);
Chuck Lever6faddda692023-06-30 13:48:49 -0400502 if (!child)
503 goto out;
Chuck Lever2be4f052023-07-25 14:31:04 -0400504 spin_lock(&child->d_lock);
Chuck Lever6faddda692023-06-30 13:48:49 -0400505 if (simple_positive(child))
506 found = dget_dlock(child);
507 spin_unlock(&child->d_lock);
508out:
509 rcu_read_unlock();
510 return found;
511}
512
513static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry)
514{
Chuck Lever6faddda692023-06-30 13:48:49 -0400515 struct inode *inode = d_inode(dentry);
Chuck Lever0e4a8622024-02-17 15:24:16 -0500516 long offset = dentry2offset(dentry);
Chuck Lever6faddda692023-06-30 13:48:49 -0400517
518 return ctx->actor(ctx, dentry->d_name.name, dentry->d_name.len, offset,
519 inode->i_ino, fs_umode_to_dtype(inode->i_mode));
520}
521
yangerkun64a7ce72024-07-31 12:38:35 +0800522static void offset_iterate_dir(struct inode *inode, struct dir_context *ctx, long last_index)
Chuck Lever6faddda692023-06-30 13:48:49 -0400523{
Chuck Lever3f6d8102024-02-17 15:23:40 -0500524 struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
Chuck Lever6faddda692023-06-30 13:48:49 -0400525 struct dentry *dentry;
526
527 while (true) {
Chuck Lever3f6d8102024-02-17 15:23:40 -0500528 dentry = offset_find_next(octx, ctx->pos);
Chuck Lever6faddda692023-06-30 13:48:49 -0400529 if (!dentry)
yangerkun64a7ce72024-07-31 12:38:35 +0800530 return;
531
532 if (dentry2offset(dentry) >= last_index) {
533 dput(dentry);
534 return;
535 }
Chuck Lever6faddda692023-06-30 13:48:49 -0400536
537 if (!offset_dir_emit(ctx, dentry)) {
538 dput(dentry);
yangerkun64a7ce72024-07-31 12:38:35 +0800539 return;
Chuck Lever6faddda692023-06-30 13:48:49 -0400540 }
541
Chuck Lever3f6d8102024-02-17 15:23:40 -0500542 ctx->pos = dentry2offset(dentry) + 1;
Chuck Lever6faddda692023-06-30 13:48:49 -0400543 dput(dentry);
Chuck Lever6faddda692023-06-30 13:48:49 -0400544 }
545}
546
547/**
548 * offset_readdir - Emit entries starting at offset @ctx->pos
549 * @file: an open directory to iterate over
550 * @ctx: directory iteration context
551 *
552 * Caller must hold @file's i_rwsem to prevent insertion or removal of
553 * entries during this call.
554 *
555 * On entry, @ctx->pos contains an offset that represents the first entry
556 * to be read from the directory.
557 *
558 * The operation continues until there are no more entries to read, or
559 * until the ctx->actor indicates there is no more space in the caller's
560 * output buffer.
561 *
562 * On return, @ctx->pos contains an offset that will read the next entry
Chuck Lever2be4f052023-07-25 14:31:04 -0400563 * in this directory when offset_readdir() is called again with @ctx.
Chuck Lever6faddda692023-06-30 13:48:49 -0400564 *
565 * Return values:
566 * %0 - Complete
567 */
568static int offset_readdir(struct file *file, struct dir_context *ctx)
569{
570 struct dentry *dir = file->f_path.dentry;
yangerkun64a7ce72024-07-31 12:38:35 +0800571 long last_index = (long)file->private_data;
Chuck Lever6faddda692023-06-30 13:48:49 -0400572
573 lockdep_assert_held(&d_inode(dir)->i_rwsem);
574
575 if (!dir_emit_dots(file, ctx))
576 return 0;
577
yangerkun64a7ce72024-07-31 12:38:35 +0800578 offset_iterate_dir(d_inode(dir), ctx, last_index);
Chuck Lever6faddda692023-06-30 13:48:49 -0400579 return 0;
580}
581
582const struct file_operations simple_offset_dir_operations = {
yangerkun64a7ce72024-07-31 12:38:35 +0800583 .open = offset_dir_open,
Chuck Lever6faddda692023-06-30 13:48:49 -0400584 .llseek = offset_dir_llseek,
585 .iterate_shared = offset_readdir,
586 .read = generic_read_dir,
587 .fsync = noop_fsync,
588};
589
Al Viroa3d1e7e2019-11-18 09:43:10 -0500590static struct dentry *find_next_child(struct dentry *parent, struct dentry *prev)
591{
Al Viroda549bd2023-11-07 02:00:39 -0500592 struct dentry *child = NULL, *d;
Al Viroa3d1e7e2019-11-18 09:43:10 -0500593
594 spin_lock(&parent->d_lock);
Al Viroda549bd2023-11-07 02:00:39 -0500595 d = prev ? d_next_sibling(prev) : d_first_child(parent);
596 hlist_for_each_entry_from(d, d_sib) {
Al Viroa3d1e7e2019-11-18 09:43:10 -0500597 if (simple_positive(d)) {
598 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
599 if (simple_positive(d))
600 child = dget_dlock(d);
601 spin_unlock(&d->d_lock);
602 if (likely(child))
603 break;
604 }
605 }
606 spin_unlock(&parent->d_lock);
607 dput(prev);
608 return child;
609}
610
611void simple_recursive_removal(struct dentry *dentry,
612 void (*callback)(struct dentry *))
613{
614 struct dentry *this = dget(dentry);
615 while (true) {
616 struct dentry *victim = NULL, *child;
617 struct inode *inode = this->d_inode;
618
619 inode_lock(inode);
620 if (d_is_dir(this))
621 inode->i_flags |= S_DEAD;
622 while ((child = find_next_child(this, victim)) == NULL) {
623 // kill and ascend
624 // update metadata while it's still locked
Jeff Laytonf7f43852023-07-05 15:01:21 -0400625 inode_set_ctime_current(inode);
Al Viroa3d1e7e2019-11-18 09:43:10 -0500626 clear_nlink(inode);
627 inode_unlock(inode);
628 victim = this;
629 this = this->d_parent;
630 inode = this->d_inode;
631 inode_lock(inode);
632 if (simple_positive(victim)) {
633 d_invalidate(victim); // avoid lost mounts
634 if (d_is_dir(victim))
635 fsnotify_rmdir(inode, victim);
636 else
637 fsnotify_unlink(inode, victim);
638 if (callback)
639 callback(victim);
640 dput(victim); // unpin it
641 }
642 if (victim == dentry) {
Jeff Layton077c2122023-10-04 14:52:37 -0400643 inode_set_mtime_to_ts(inode,
644 inode_set_ctime_current(inode));
Al Viroa3d1e7e2019-11-18 09:43:10 -0500645 if (d_is_dir(dentry))
646 drop_nlink(inode);
647 inode_unlock(inode);
648 dput(dentry);
649 return;
650 }
651 }
652 inode_unlock(inode);
653 this = child;
654 }
655}
656EXPORT_SYMBOL(simple_recursive_removal);
657
Hugh Dickins759b9772007-03-05 00:30:28 -0800658static const struct super_operations simple_super_operations = {
659 .statfs = simple_statfs,
660};
661
David Howellsdb2c2462019-03-25 16:38:26 +0000662static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663{
David Howells31d6d5c2019-03-25 16:38:23 +0000664 struct pseudo_fs_context *ctx = fc->fs_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 struct inode *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
Jeff Layton89a4eb42009-08-18 14:11:08 -0700667 s->s_maxbytes = MAX_LFS_FILESIZE;
Alex Nixon3971e1a2008-07-29 22:33:03 -0700668 s->s_blocksize = PAGE_SIZE;
669 s->s_blocksize_bits = PAGE_SHIFT;
Al Viro8d9e46d2019-05-11 11:43:59 -0400670 s->s_magic = ctx->magic;
671 s->s_op = ctx->ops ?: &simple_super_operations;
672 s->s_xattr = ctx->xattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 s->s_time_gran = 1;
674 root = new_inode(s);
675 if (!root)
David Howellsdb2c2462019-03-25 16:38:26 +0000676 return -ENOMEM;
677
Jeff Layton1a1c9bb2007-05-08 00:32:31 -0700678 /*
679 * since this is the first inode, make it number 1. New inodes created
680 * after this must take care not to collide with it (by passing
681 * max_reserved of 1 to iunique).
682 */
683 root->i_ino = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
Jeff Layton077c2122023-10-04 14:52:37 -0400685 simple_inode_init_ts(root);
Al Viro8d9e46d2019-05-11 11:43:59 -0400686 s->s_root = d_make_root(root);
687 if (!s->s_root)
David Howellsdb2c2462019-03-25 16:38:26 +0000688 return -ENOMEM;
Al Viro8d9e46d2019-05-11 11:43:59 -0400689 s->s_d_op = ctx->dops;
David Howells31d6d5c2019-03-25 16:38:23 +0000690 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
Al Viro8d9e46d2019-05-11 11:43:59 -0400692
David Howellsdb2c2462019-03-25 16:38:26 +0000693static int pseudo_fs_get_tree(struct fs_context *fc)
694{
Al Viro2ac295d2019-06-01 20:48:55 -0400695 return get_tree_nodev(fc, pseudo_fs_fill_super);
David Howells31d6d5c2019-03-25 16:38:23 +0000696}
697
698static void pseudo_fs_free(struct fs_context *fc)
699{
700 kfree(fc->fs_private);
701}
702
703static const struct fs_context_operations pseudo_fs_context_ops = {
704 .free = pseudo_fs_free,
705 .get_tree = pseudo_fs_get_tree,
706};
707
708/*
709 * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
710 * will never be mountable)
711 */
712struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
713 unsigned long magic)
714{
715 struct pseudo_fs_context *ctx;
716
717 ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
718 if (likely(ctx)) {
719 ctx->magic = magic;
720 fc->fs_private = ctx;
721 fc->ops = &pseudo_fs_context_ops;
David Howellsdb2c2462019-03-25 16:38:26 +0000722 fc->sb_flags |= SB_NOUSER;
723 fc->global = true;
David Howells31d6d5c2019-03-25 16:38:23 +0000724 }
725 return ctx;
726}
727EXPORT_SYMBOL(init_pseudo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Stephen Boyd20955e82012-04-05 14:25:09 -0700729int simple_open(struct inode *inode, struct file *file)
730{
731 if (inode->i_private)
732 file->private_data = inode->i_private;
733 return 0;
734}
Al Viro12f38872013-09-15 21:20:49 -0400735EXPORT_SYMBOL(simple_open);
Stephen Boyd20955e82012-04-05 14:25:09 -0700736
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
738{
David Howellsdea655c2015-03-17 22:26:15 +0000739 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Jeff Layton077c2122023-10-04 14:52:37 -0400741 inode_set_mtime_to_ts(dir,
742 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
Dave Hansend8c76e62006-09-30 23:29:04 -0700743 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -0400744 ihold(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 dget(dentry);
746 d_instantiate(dentry, inode);
747 return 0;
748}
Al Viro12f38872013-09-15 21:20:49 -0400749EXPORT_SYMBOL(simple_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751int simple_empty(struct dentry *dentry)
752{
753 struct dentry *child;
754 int ret = 0;
755
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100756 spin_lock(&dentry->d_lock);
Al Viroda549bd2023-11-07 02:00:39 -0500757 hlist_for_each_entry(child, &dentry->d_children, d_sib) {
Nick Pigginda502952011-01-07 17:49:33 +1100758 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
759 if (simple_positive(child)) {
760 spin_unlock(&child->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 goto out;
Nick Pigginda502952011-01-07 17:49:33 +1100762 }
763 spin_unlock(&child->d_lock);
764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 ret = 1;
766out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100767 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 return ret;
769}
Al Viro12f38872013-09-15 21:20:49 -0400770EXPORT_SYMBOL(simple_empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
772int simple_unlink(struct inode *dir, struct dentry *dentry)
773{
David Howellsdea655c2015-03-17 22:26:15 +0000774 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Jeff Layton077c2122023-10-04 14:52:37 -0400776 inode_set_mtime_to_ts(dir,
777 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700778 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 dput(dentry);
780 return 0;
781}
Al Viro12f38872013-09-15 21:20:49 -0400782EXPORT_SYMBOL(simple_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
784int simple_rmdir(struct inode *dir, struct dentry *dentry)
785{
786 if (!simple_empty(dentry))
787 return -ENOTEMPTY;
788
David Howellsdea655c2015-03-17 22:26:15 +0000789 drop_nlink(d_inode(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 simple_unlink(dir, dentry);
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700791 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 return 0;
793}
Al Viro12f38872013-09-15 21:20:49 -0400794EXPORT_SYMBOL(simple_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Jeff Layton0c476792023-07-05 14:58:11 -0400796/**
797 * simple_rename_timestamp - update the various inode timestamps for rename
798 * @old_dir: old parent directory
799 * @old_dentry: dentry that is being renamed
800 * @new_dir: new parent directory
801 * @new_dentry: target for rename
802 *
803 * POSIX mandates that the old and new parent directories have their ctime and
804 * mtime updated, and that inodes of @old_dentry and @new_dentry (if any), have
805 * their ctime updated.
806 */
807void simple_rename_timestamp(struct inode *old_dir, struct dentry *old_dentry,
808 struct inode *new_dir, struct dentry *new_dentry)
809{
810 struct inode *newino = d_inode(new_dentry);
811
Jeff Layton077c2122023-10-04 14:52:37 -0400812 inode_set_mtime_to_ts(old_dir, inode_set_ctime_current(old_dir));
Jeff Layton0c476792023-07-05 14:58:11 -0400813 if (new_dir != old_dir)
Jeff Layton077c2122023-10-04 14:52:37 -0400814 inode_set_mtime_to_ts(new_dir,
815 inode_set_ctime_current(new_dir));
Jeff Layton0c476792023-07-05 14:58:11 -0400816 inode_set_ctime_current(d_inode(old_dentry));
817 if (newino)
818 inode_set_ctime_current(newino);
819}
820EXPORT_SYMBOL_GPL(simple_rename_timestamp);
821
Lorenz Bauer6429e462021-10-28 10:47:21 +0100822int simple_rename_exchange(struct inode *old_dir, struct dentry *old_dentry,
823 struct inode *new_dir, struct dentry *new_dentry)
824{
825 bool old_is_dir = d_is_dir(old_dentry);
826 bool new_is_dir = d_is_dir(new_dentry);
827
828 if (old_dir != new_dir && old_is_dir != new_is_dir) {
829 if (old_is_dir) {
830 drop_nlink(old_dir);
831 inc_nlink(new_dir);
832 } else {
833 drop_nlink(new_dir);
834 inc_nlink(old_dir);
835 }
836 }
Jeff Layton0c476792023-07-05 14:58:11 -0400837 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
Lorenz Bauer6429e462021-10-28 10:47:21 +0100838 return 0;
839}
840EXPORT_SYMBOL_GPL(simple_rename_exchange);
841
Christian Braunere18275a2023-01-13 12:49:17 +0100842int simple_rename(struct mnt_idmap *idmap, struct inode *old_dir,
Christian Brauner549c7292021-01-21 14:19:43 +0100843 struct dentry *old_dentry, struct inode *new_dir,
844 struct dentry *new_dentry, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845{
David Howellse36cb0b2015-01-29 12:02:35 +0000846 int they_are_dirs = d_is_dir(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Lorenz Bauer3871cb82021-10-28 10:47:22 +0100848 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
Miklos Szeredie0e0be82016-09-27 11:03:57 +0200849 return -EINVAL;
850
Lorenz Bauer3871cb82021-10-28 10:47:22 +0100851 if (flags & RENAME_EXCHANGE)
852 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 if (!simple_empty(new_dentry))
855 return -ENOTEMPTY;
856
David Howellsdea655c2015-03-17 22:26:15 +0000857 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 simple_unlink(new_dir, new_dentry);
Al Viro841590c2011-07-21 15:49:09 -0400859 if (they_are_dirs) {
David Howellsdea655c2015-03-17 22:26:15 +0000860 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700861 drop_nlink(old_dir);
Al Viro841590c2011-07-21 15:49:09 -0400862 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700864 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -0700865 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 }
867
Jeff Layton0c476792023-07-05 14:58:11 -0400868 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 return 0;
870}
Al Viro12f38872013-09-15 21:20:49 -0400871EXPORT_SYMBOL(simple_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000873/**
Christoph Hellwigeef23802010-06-04 11:30:01 +0200874 * simple_setattr - setattr for simple filesystem
Christian Braunerc1632a02023-01-13 12:49:11 +0100875 * @idmap: idmap of the target mount
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000876 * @dentry: dentry
877 * @iattr: iattr structure
878 *
879 * Returns 0 on success, -error on failure.
880 *
Christoph Hellwigeef23802010-06-04 11:30:01 +0200881 * simple_setattr is a simple ->setattr implementation without a proper
882 * implementation of size changes.
883 *
884 * It can either be used for in-memory filesystems or special files
885 * on simple regular filesystems. Anything that needs to change on-disk
886 * or wire state on size changes needs its own setattr method.
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000887 */
Christian Braunerc1632a02023-01-13 12:49:11 +0100888int simple_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
Christian Brauner549c7292021-01-21 14:19:43 +0100889 struct iattr *iattr)
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000890{
David Howellsdea655c2015-03-17 22:26:15 +0000891 struct inode *inode = d_inode(dentry);
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000892 int error;
893
Christian Braunerc1632a02023-01-13 12:49:11 +0100894 error = setattr_prepare(idmap, dentry, iattr);
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000895 if (error)
896 return error;
897
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200898 if (iattr->ia_valid & ATTR_SIZE)
899 truncate_setsize(inode, iattr->ia_size);
Christian Braunerc1632a02023-01-13 12:49:11 +0100900 setattr_copy(idmap, inode, iattr);
Christoph Hellwigeef23802010-06-04 11:30:01 +0200901 mark_inode_dirty(inode);
902 return 0;
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000903}
904EXPORT_SYMBOL(simple_setattr);
905
Matthew Wilcox (Oracle)a77f5802022-04-29 11:49:41 -0400906static int simple_read_folio(struct file *file, struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
Matthew Wilcox (Oracle)a77f5802022-04-29 11:49:41 -0400908 folio_zero_range(folio, 0, folio_size(folio));
909 flush_dcache_folio(folio);
910 folio_mark_uptodate(folio);
911 folio_unlock(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 return 0;
913}
914
Nick Pigginafddba42007-10-16 01:25:01 -0700915int simple_write_begin(struct file *file, struct address_space *mapping,
Matthew Wilcox (Oracle)9d6b0cd2022-02-22 14:31:43 -0500916 loff_t pos, unsigned len,
Matthew Wilcox (Oracle)1da86612024-07-15 14:24:01 -0400917 struct folio **foliop, void **fsdata)
Nick Pigginafddba42007-10-16 01:25:01 -0700918{
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100919 struct folio *folio;
Nick Pigginafddba42007-10-16 01:25:01 -0700920
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100921 folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN,
922 mapping_gfp_mask(mapping));
923 if (IS_ERR(folio))
924 return PTR_ERR(folio);
Nick Pigginafddba42007-10-16 01:25:01 -0700925
Matthew Wilcox (Oracle)1da86612024-07-15 14:24:01 -0400926 *foliop = folio;
Nick Pigginafddba42007-10-16 01:25:01 -0700927
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100928 if (!folio_test_uptodate(folio) && (len != folio_size(folio))) {
929 size_t from = offset_in_folio(folio, pos);
Nick Pigginafddba42007-10-16 01:25:01 -0700930
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100931 folio_zero_segments(folio, 0, from,
932 from + len, folio_size(folio));
Boaz Harrosh193cf4b2010-01-12 16:18:08 +0200933 }
934 return 0;
Nick Pigginafddba42007-10-16 01:25:01 -0700935}
Al Viro12f38872013-09-15 21:20:49 -0400936EXPORT_SYMBOL(simple_write_begin);
Nick Pigginafddba42007-10-16 01:25:01 -0700937
Boaz Harroshad2a722f2010-01-12 15:13:47 +0200938/**
939 * simple_write_end - .write_end helper for non-block-device FSes
Randy Dunlap8e88bfb2019-10-14 14:12:14 -0700940 * @file: See .write_end of address_space_operations
Boaz Harroshad2a722f2010-01-12 15:13:47 +0200941 * @mapping: "
942 * @pos: "
943 * @len: "
944 * @copied: "
Matthew Wilcox (Oracle)a2258002024-07-10 15:45:32 -0400945 * @folio: "
Boaz Harroshad2a722f2010-01-12 15:13:47 +0200946 * @fsdata: "
947 *
Matthew Wilcox (Oracle)a2258002024-07-10 15:45:32 -0400948 * simple_write_end does the minimum needed for updating a folio after
949 * writing is done. It has the same API signature as the .write_end of
Boaz Harroshad2a722f2010-01-12 15:13:47 +0200950 * address_space_operations vector. So it can just be set onto .write_end for
951 * FSes that don't need any other processing. i_mutex is assumed to be held.
952 * Block based filesystems should use generic_write_end().
953 * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
954 * is not called, so a filesystem that actually does store data in .write_inode
955 * should extend on what's done here with a call to mark_inode_dirty() in the
956 * case that i_size has changed.
Al Viro04fff642016-08-29 22:39:56 -0400957 *
Matthew Wilcox (Oracle)a77f5802022-04-29 11:49:41 -0400958 * Use *ONLY* with simple_read_folio()
Boaz Harroshad2a722f2010-01-12 15:13:47 +0200959 */
Christoph Hellwigc1e3dbe2021-06-28 19:36:09 -0700960static int simple_write_end(struct file *file, struct address_space *mapping,
Boaz Harroshad2a722f2010-01-12 15:13:47 +0200961 loff_t pos, unsigned len, unsigned copied,
Matthew Wilcox (Oracle)a2258002024-07-10 15:45:32 -0400962 struct folio *folio, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963{
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100964 struct inode *inode = folio->mapping->host;
Boaz Harroshad2a722f2010-01-12 15:13:47 +0200965 loff_t last_pos = pos + copied;
966
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100967 /* zero the stale part of the folio if we did a short copy */
968 if (!folio_test_uptodate(folio)) {
Al Viro04fff642016-08-29 22:39:56 -0400969 if (copied < len) {
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100970 size_t from = offset_in_folio(folio, pos);
Boaz Harroshad2a722f2010-01-12 15:13:47 +0200971
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100972 folio_zero_range(folio, from + copied, len - copied);
Al Viro04fff642016-08-29 22:39:56 -0400973 }
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100974 folio_mark_uptodate(folio);
Al Viro04fff642016-08-29 22:39:56 -0400975 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 /*
977 * No need to use i_size_read() here, the i_size
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800978 * cannot change under us because we hold the i_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 */
Boaz Harroshad2a722f2010-01-12 15:13:47 +0200980 if (last_pos > inode->i_size)
981 i_size_write(inode, last_pos);
982
Matthew Wilcox (Oracle)5522d9f2023-08-21 15:13:22 +0100983 folio_mark_dirty(folio);
984 folio_unlock(folio);
985 folio_put(folio);
Nick Pigginafddba42007-10-16 01:25:01 -0700986
987 return copied;
988}
Christoph Hellwigc1e3dbe2021-06-28 19:36:09 -0700989
990/*
991 * Provides ramfs-style behavior: data in the pagecache, but no writeback.
992 */
993const struct address_space_operations ram_aops = {
Matthew Wilcox (Oracle)a77f5802022-04-29 11:49:41 -0400994 .read_folio = simple_read_folio,
Christoph Hellwigc1e3dbe2021-06-28 19:36:09 -0700995 .write_begin = simple_write_begin,
996 .write_end = simple_write_end,
Matthew Wilcox (Oracle)46de8b972022-02-09 20:22:13 +0000997 .dirty_folio = noop_dirty_folio,
Christoph Hellwigc1e3dbe2021-06-28 19:36:09 -0700998};
999EXPORT_SYMBOL(ram_aops);
Nick Pigginafddba42007-10-16 01:25:01 -07001000
Jeff Layton1a1c9bb2007-05-08 00:32:31 -07001001/*
1002 * the inodes created here are not hashed. If you use iunique to generate
1003 * unique inode values later for this filesystem, then you must take care
1004 * to pass it an appropriate max_reserved value to avoid collisions.
1005 */
Roberto Sassu7d683a02010-06-03 11:58:28 +02001006int simple_fill_super(struct super_block *s, unsigned long magic,
Eric Biggerscda37122017-03-25 21:15:37 -07001007 const struct tree_descr *files)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 struct dentry *dentry;
1011 int i;
1012
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001013 s->s_blocksize = PAGE_SIZE;
1014 s->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 s->s_magic = magic;
Hugh Dickins759b9772007-03-05 00:30:28 -08001016 s->s_op = &simple_super_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 s->s_time_gran = 1;
1018
1019 inode = new_inode(s);
1020 if (!inode)
1021 return -ENOMEM;
Jeff Layton1a1c9bb2007-05-08 00:32:31 -07001022 /*
1023 * because the root inode is 1, the files array must not contain an
1024 * entry at index 1
1025 */
1026 inode->i_ino = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 inode->i_mode = S_IFDIR | 0755;
Jeff Layton077c2122023-10-04 14:52:37 -04001028 simple_inode_init_ts(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 inode->i_op = &simple_dir_inode_operations;
1030 inode->i_fop = &simple_dir_operations;
Miklos Szeredibfe86842011-10-28 14:13:29 +02001031 set_nlink(inode, 2);
Al Viro715cd662023-11-11 15:56:55 -05001032 s->s_root = d_make_root(inode);
1033 if (!s->s_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 for (i = 0; !files->name || files->name[0]; i++, files++) {
1036 if (!files->name)
1037 continue;
Jeff Layton1a1c9bb2007-05-08 00:32:31 -07001038
1039 /* warn if it tries to conflict with the root inode */
1040 if (unlikely(i == 1))
1041 printk(KERN_WARNING "%s: %s passed in a files array"
1042 "with an index of 1!\n", __func__,
1043 s->s_type->name);
1044
Al Viro715cd662023-11-11 15:56:55 -05001045 dentry = d_alloc_name(s->s_root, files->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 if (!dentry)
Al Viro715cd662023-11-11 15:56:55 -05001047 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 inode = new_inode(s);
Konstantin Khlebnikov32096ea2011-11-01 16:12:33 +03001049 if (!inode) {
1050 dput(dentry);
Al Viro715cd662023-11-11 15:56:55 -05001051 return -ENOMEM;
Konstantin Khlebnikov32096ea2011-11-01 16:12:33 +03001052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 inode->i_mode = S_IFREG | files->mode;
Jeff Layton077c2122023-10-04 14:52:37 -04001054 simple_inode_init_ts(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 inode->i_fop = files->ops;
1056 inode->i_ino = i;
1057 d_add(dentry, inode);
1058 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060}
Al Viro12f38872013-09-15 21:20:49 -04001061EXPORT_SYMBOL(simple_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
1063static DEFINE_SPINLOCK(pin_fs_lock);
1064
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04001065int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066{
1067 struct vfsmount *mnt = NULL;
1068 spin_lock(&pin_fs_lock);
1069 if (unlikely(!*mount)) {
1070 spin_unlock(&pin_fs_lock);
Linus Torvalds1751e8a2017-11-27 13:05:09 -08001071 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 if (IS_ERR(mnt))
1073 return PTR_ERR(mnt);
1074 spin_lock(&pin_fs_lock);
1075 if (!*mount)
1076 *mount = mnt;
1077 }
1078 mntget(*mount);
1079 ++*count;
1080 spin_unlock(&pin_fs_lock);
1081 mntput(mnt);
1082 return 0;
1083}
Al Viro12f38872013-09-15 21:20:49 -04001084EXPORT_SYMBOL(simple_pin_fs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
1086void simple_release_fs(struct vfsmount **mount, int *count)
1087{
1088 struct vfsmount *mnt;
1089 spin_lock(&pin_fs_lock);
1090 mnt = *mount;
1091 if (!--*count)
1092 *mount = NULL;
1093 spin_unlock(&pin_fs_lock);
1094 mntput(mnt);
1095}
Al Viro12f38872013-09-15 21:20:49 -04001096EXPORT_SYMBOL(simple_release_fs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
Akinobu Mita6d1029b2008-07-04 09:59:51 -07001098/**
1099 * simple_read_from_buffer - copy data from the buffer to user space
1100 * @to: the user space buffer to read to
1101 * @count: the maximum number of bytes to read
1102 * @ppos: the current position in the buffer
1103 * @from: the buffer to read from
1104 * @available: the size of the buffer
1105 *
1106 * The simple_read_from_buffer() function reads up to @count bytes from the
1107 * buffer @from at offset @ppos into the user space address starting at @to.
1108 *
1109 * On success, the number of bytes read is returned and the offset @ppos is
1110 * advanced by this number, or negative value is returned on error.
1111 **/
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
1113 const void *from, size_t available)
1114{
1115 loff_t pos = *ppos;
Steven Rostedt14be2742009-09-18 13:05:42 -07001116 size_t ret;
1117
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 if (pos < 0)
1119 return -EINVAL;
Steven Rostedt14be2742009-09-18 13:05:42 -07001120 if (pos >= available || !count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 return 0;
1122 if (count > available - pos)
1123 count = available - pos;
Steven Rostedt14be2742009-09-18 13:05:42 -07001124 ret = copy_to_user(to, from + pos, count);
1125 if (ret == count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 return -EFAULT;
Steven Rostedt14be2742009-09-18 13:05:42 -07001127 count -= ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 *ppos = pos + count;
1129 return count;
1130}
Al Viro12f38872013-09-15 21:20:49 -04001131EXPORT_SYMBOL(simple_read_from_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Akinobu Mita6d1029b2008-07-04 09:59:51 -07001133/**
Jiri Slaby6a727b42010-05-01 23:51:22 +02001134 * simple_write_to_buffer - copy data from user space to the buffer
1135 * @to: the buffer to write to
1136 * @available: the size of the buffer
1137 * @ppos: the current position in the buffer
1138 * @from: the user space buffer to read from
1139 * @count: the maximum number of bytes to read
1140 *
1141 * The simple_write_to_buffer() function reads up to @count bytes from the user
1142 * space address starting at @from into the buffer @to at offset @ppos.
1143 *
1144 * On success, the number of bytes written is returned and the offset @ppos is
1145 * advanced by this number, or negative value is returned on error.
1146 **/
1147ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
1148 const void __user *from, size_t count)
1149{
1150 loff_t pos = *ppos;
1151 size_t res;
1152
1153 if (pos < 0)
1154 return -EINVAL;
1155 if (pos >= available || !count)
1156 return 0;
1157 if (count > available - pos)
1158 count = available - pos;
1159 res = copy_from_user(to + pos, from, count);
1160 if (res == count)
1161 return -EFAULT;
1162 count -= res;
1163 *ppos = pos + count;
1164 return count;
1165}
Al Viro12f38872013-09-15 21:20:49 -04001166EXPORT_SYMBOL(simple_write_to_buffer);
Jiri Slaby6a727b42010-05-01 23:51:22 +02001167
1168/**
Akinobu Mita6d1029b2008-07-04 09:59:51 -07001169 * memory_read_from_buffer - copy data from the buffer
1170 * @to: the kernel space buffer to read to
1171 * @count: the maximum number of bytes to read
1172 * @ppos: the current position in the buffer
1173 * @from: the buffer to read from
1174 * @available: the size of the buffer
1175 *
1176 * The memory_read_from_buffer() function reads up to @count bytes from the
1177 * buffer @from at offset @ppos into the kernel space address starting at @to.
1178 *
1179 * On success, the number of bytes read is returned and the offset @ppos is
1180 * advanced by this number, or negative value is returned on error.
1181 **/
Akinobu Mita93b07112008-06-05 22:46:21 -07001182ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
1183 const void *from, size_t available)
1184{
1185 loff_t pos = *ppos;
1186
1187 if (pos < 0)
1188 return -EINVAL;
1189 if (pos >= available)
1190 return 0;
1191 if (count > available - pos)
1192 count = available - pos;
1193 memcpy(to, from + pos, count);
1194 *ppos = pos + count;
1195
1196 return count;
1197}
Al Viro12f38872013-09-15 21:20:49 -04001198EXPORT_SYMBOL(memory_read_from_buffer);
Akinobu Mita93b07112008-06-05 22:46:21 -07001199
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200/*
1201 * Transaction based IO.
1202 * The file expects a single write which triggers the transaction, and then
1203 * possibly a read which collects the result - which is stored in a
1204 * file-local buffer.
1205 */
Ingo Molnar76791ab2009-03-25 16:48:35 +01001206
1207void simple_transaction_set(struct file *file, size_t n)
1208{
1209 struct simple_transaction_argresp *ar = file->private_data;
1210
1211 BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
1212
1213 /*
1214 * The barrier ensures that ar->size will really remain zero until
1215 * ar->data is ready for reading.
1216 */
1217 smp_mb();
1218 ar->size = n;
1219}
Al Viro12f38872013-09-15 21:20:49 -04001220EXPORT_SYMBOL(simple_transaction_set);
Ingo Molnar76791ab2009-03-25 16:48:35 +01001221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
1223{
1224 struct simple_transaction_argresp *ar;
1225 static DEFINE_SPINLOCK(simple_transaction_lock);
1226
1227 if (size > SIMPLE_TRANSACTION_LIMIT - 1)
1228 return ERR_PTR(-EFBIG);
1229
1230 ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
1231 if (!ar)
1232 return ERR_PTR(-ENOMEM);
1233
1234 spin_lock(&simple_transaction_lock);
1235
1236 /* only one write allowed per open */
1237 if (file->private_data) {
1238 spin_unlock(&simple_transaction_lock);
1239 free_page((unsigned long)ar);
1240 return ERR_PTR(-EBUSY);
1241 }
1242
1243 file->private_data = ar;
1244
1245 spin_unlock(&simple_transaction_lock);
1246
1247 if (copy_from_user(ar->data, buf, size))
1248 return ERR_PTR(-EFAULT);
1249
1250 return ar->data;
1251}
Al Viro12f38872013-09-15 21:20:49 -04001252EXPORT_SYMBOL(simple_transaction_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
1255{
1256 struct simple_transaction_argresp *ar = file->private_data;
1257
1258 if (!ar)
1259 return 0;
1260 return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
1261}
Al Viro12f38872013-09-15 21:20:49 -04001262EXPORT_SYMBOL(simple_transaction_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264int simple_transaction_release(struct inode *inode, struct file *file)
1265{
1266 free_page((unsigned long)file->private_data);
1267 return 0;
1268}
Al Viro12f38872013-09-15 21:20:49 -04001269EXPORT_SYMBOL(simple_transaction_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001271/* Simple attribute files */
1272
1273struct simple_attr {
Christoph Hellwig8b88b092008-02-08 04:20:26 -08001274 int (*get)(void *, u64 *);
1275 int (*set)(void *, u64);
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001276 char get_buf[24]; /* enough to store a u64 and "\n\0" */
1277 char set_buf[24];
1278 void *data;
1279 const char *fmt; /* format for read operation */
Ingo Molnar7cf34c72006-03-23 03:00:36 -08001280 struct mutex mutex; /* protects access to these buffers */
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001281};
1282
1283/* simple_attr_open is called by an actual attribute open file operation
1284 * to set the attribute specific access operations. */
1285int simple_attr_open(struct inode *inode, struct file *file,
Christoph Hellwig8b88b092008-02-08 04:20:26 -08001286 int (*get)(void *, u64 *), int (*set)(void *, u64),
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001287 const char *fmt)
1288{
1289 struct simple_attr *attr;
1290
Eric Biggersa65cab72020-03-07 18:38:49 -08001291 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001292 if (!attr)
1293 return -ENOMEM;
1294
1295 attr->get = get;
1296 attr->set = set;
Theodore Ts'o8e18e292006-09-27 01:50:46 -07001297 attr->data = inode->i_private;
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001298 attr->fmt = fmt;
Ingo Molnar7cf34c72006-03-23 03:00:36 -08001299 mutex_init(&attr->mutex);
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001300
1301 file->private_data = attr;
1302
1303 return nonseekable_open(inode, file);
1304}
Al Viro12f38872013-09-15 21:20:49 -04001305EXPORT_SYMBOL_GPL(simple_attr_open);
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001306
Christoph Hellwig74bedc42008-02-08 04:20:28 -08001307int simple_attr_release(struct inode *inode, struct file *file)
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001308{
1309 kfree(file->private_data);
1310 return 0;
1311}
Al Viro12f38872013-09-15 21:20:49 -04001312EXPORT_SYMBOL_GPL(simple_attr_release); /* GPL-only? This? Really? */
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001313
1314/* read from the buffer that is filled with the get function */
1315ssize_t simple_attr_read(struct file *file, char __user *buf,
1316 size_t len, loff_t *ppos)
1317{
1318 struct simple_attr *attr;
1319 size_t size;
1320 ssize_t ret;
1321
1322 attr = file->private_data;
1323
1324 if (!attr->get)
1325 return -EACCES;
1326
Christoph Hellwig92613032008-02-08 04:20:27 -08001327 ret = mutex_lock_interruptible(&attr->mutex);
1328 if (ret)
1329 return ret;
1330
Eric Biggersa65cab72020-03-07 18:38:49 -08001331 if (*ppos && attr->get_buf[0]) {
1332 /* continued read */
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001333 size = strlen(attr->get_buf);
Eric Biggersa65cab72020-03-07 18:38:49 -08001334 } else {
1335 /* first read */
Christoph Hellwig8b88b092008-02-08 04:20:26 -08001336 u64 val;
1337 ret = attr->get(attr->data, &val);
1338 if (ret)
1339 goto out;
1340
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001341 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
Christoph Hellwig8b88b092008-02-08 04:20:26 -08001342 attr->fmt, (unsigned long long)val);
1343 }
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001344
1345 ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08001346out:
Ingo Molnar7cf34c72006-03-23 03:00:36 -08001347 mutex_unlock(&attr->mutex);
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001348 return ret;
1349}
Al Viro12f38872013-09-15 21:20:49 -04001350EXPORT_SYMBOL_GPL(simple_attr_read);
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001351
1352/* interpret the buffer as a number to call the set function with */
Akinobu Mita2e41f272022-09-20 02:24:16 +09001353static ssize_t simple_attr_write_xsigned(struct file *file, const char __user *buf,
1354 size_t len, loff_t *ppos, bool is_signed)
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001355{
1356 struct simple_attr *attr;
Yicong Yang488dac02020-11-21 22:17:19 -08001357 unsigned long long val;
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001358 size_t size;
1359 ssize_t ret;
1360
1361 attr = file->private_data;
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001362 if (!attr->set)
1363 return -EACCES;
1364
Christoph Hellwig92613032008-02-08 04:20:27 -08001365 ret = mutex_lock_interruptible(&attr->mutex);
1366 if (ret)
1367 return ret;
1368
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001369 ret = -EFAULT;
1370 size = min(sizeof(attr->set_buf) - 1, len);
1371 if (copy_from_user(attr->set_buf, buf, size))
1372 goto out;
1373
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001374 attr->set_buf[size] = '\0';
Akinobu Mita2e41f272022-09-20 02:24:16 +09001375 if (is_signed)
1376 ret = kstrtoll(attr->set_buf, 0, &val);
1377 else
1378 ret = kstrtoull(attr->set_buf, 0, &val);
Yicong Yang488dac02020-11-21 22:17:19 -08001379 if (ret)
1380 goto out;
Wu Fengguang05cc0ce2009-09-18 13:06:03 -07001381 ret = attr->set(attr->data, val);
1382 if (ret == 0)
1383 ret = len; /* on success, claim we got the whole input */
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001384out:
Ingo Molnar7cf34c72006-03-23 03:00:36 -08001385 mutex_unlock(&attr->mutex);
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001386 return ret;
1387}
Akinobu Mita2e41f272022-09-20 02:24:16 +09001388
1389ssize_t simple_attr_write(struct file *file, const char __user *buf,
1390 size_t len, loff_t *ppos)
1391{
1392 return simple_attr_write_xsigned(file, buf, len, ppos, false);
1393}
Al Viro12f38872013-09-15 21:20:49 -04001394EXPORT_SYMBOL_GPL(simple_attr_write);
Arnd Bergmannacaefc22005-05-18 14:40:59 +02001395
Akinobu Mita2e41f272022-09-20 02:24:16 +09001396ssize_t simple_attr_write_signed(struct file *file, const char __user *buf,
1397 size_t len, loff_t *ppos)
1398{
1399 return simple_attr_write_xsigned(file, buf, len, ppos, true);
1400}
1401EXPORT_SYMBOL_GPL(simple_attr_write_signed);
1402
Christoph Hellwig25961102007-10-21 16:42:05 -07001403/**
Amir Goldsteind9e5d922023-10-26 23:45:40 +03001404 * generic_encode_ino32_fh - generic export_operations->encode_fh function
1405 * @inode: the object to encode
1406 * @fh: where to store the file handle fragment
1407 * @max_len: maximum length to store there (in 4 byte units)
1408 * @parent: parent directory inode, if wanted
1409 *
1410 * This generic encode_fh function assumes that the 32 inode number
1411 * is suitable for locating an inode, and that the generation number
1412 * can be used to check that it is still valid. It places them in the
1413 * filehandle fragment where export_decode_fh expects to find them.
1414 */
1415int generic_encode_ino32_fh(struct inode *inode, __u32 *fh, int *max_len,
1416 struct inode *parent)
1417{
1418 struct fid *fid = (void *)fh;
1419 int len = *max_len;
1420 int type = FILEID_INO32_GEN;
1421
1422 if (parent && (len < 4)) {
1423 *max_len = 4;
1424 return FILEID_INVALID;
1425 } else if (len < 2) {
1426 *max_len = 2;
1427 return FILEID_INVALID;
1428 }
1429
1430 len = 2;
1431 fid->i32.ino = inode->i_ino;
1432 fid->i32.gen = inode->i_generation;
1433 if (parent) {
1434 fid->i32.parent_ino = parent->i_ino;
1435 fid->i32.parent_gen = parent->i_generation;
1436 len = 4;
1437 type = FILEID_INO32_GEN_PARENT;
1438 }
1439 *max_len = len;
1440 return type;
1441}
1442EXPORT_SYMBOL_GPL(generic_encode_ino32_fh);
1443
1444/**
Christoph Hellwig25961102007-10-21 16:42:05 -07001445 * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
1446 * @sb: filesystem to do the file handle conversion on
1447 * @fid: file handle to convert
1448 * @fh_len: length of the file handle in bytes
1449 * @fh_type: type of file handle
1450 * @get_inode: filesystem callback to retrieve inode
1451 *
1452 * This function decodes @fid as long as it has one of the well-known
1453 * Linux filehandle types and calls @get_inode on it to retrieve the
1454 * inode for the object specified in the file handle.
1455 */
1456struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
1457 int fh_len, int fh_type, struct inode *(*get_inode)
1458 (struct super_block *sb, u64 ino, u32 gen))
1459{
1460 struct inode *inode = NULL;
1461
1462 if (fh_len < 2)
1463 return NULL;
1464
1465 switch (fh_type) {
1466 case FILEID_INO32_GEN:
1467 case FILEID_INO32_GEN_PARENT:
1468 inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
1469 break;
1470 }
1471
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001472 return d_obtain_alias(inode);
Christoph Hellwig25961102007-10-21 16:42:05 -07001473}
1474EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1475
1476/**
Yanchuan Nianca186832012-09-05 16:31:29 +08001477 * generic_fh_to_parent - generic helper for the fh_to_parent export operation
Christoph Hellwig25961102007-10-21 16:42:05 -07001478 * @sb: filesystem to do the file handle conversion on
1479 * @fid: file handle to convert
1480 * @fh_len: length of the file handle in bytes
1481 * @fh_type: type of file handle
1482 * @get_inode: filesystem callback to retrieve inode
1483 *
1484 * This function decodes @fid as long as it has one of the well-known
1485 * Linux filehandle types and calls @get_inode on it to retrieve the
1486 * inode for the _parent_ object specified in the file handle if it
1487 * is specified in the file handle, or NULL otherwise.
1488 */
1489struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
1490 int fh_len, int fh_type, struct inode *(*get_inode)
1491 (struct super_block *sb, u64 ino, u32 gen))
1492{
1493 struct inode *inode = NULL;
1494
1495 if (fh_len <= 2)
1496 return NULL;
1497
1498 switch (fh_type) {
1499 case FILEID_INO32_GEN_PARENT:
1500 inode = get_inode(sb, fid->i32.parent_ino,
1501 (fh_len > 3 ? fid->i32.parent_gen : 0));
1502 break;
1503 }
1504
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001505 return d_obtain_alias(inode);
Christoph Hellwig25961102007-10-21 16:42:05 -07001506}
1507EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1508
Christoph Hellwig1b061d92010-05-26 17:53:41 +02001509/**
Fabian Frederickac13a822014-06-04 16:06:27 -07001510 * __generic_file_fsync - generic fsync implementation for simple filesystems
1511 *
Christoph Hellwig1b061d92010-05-26 17:53:41 +02001512 * @file: file to synchronize
Fabian Frederickac13a822014-06-04 16:06:27 -07001513 * @start: start offset in bytes
1514 * @end: end offset in bytes (inclusive)
Christoph Hellwig1b061d92010-05-26 17:53:41 +02001515 * @datasync: only synchronize essential metadata if true
1516 *
1517 * This is a generic implementation of the fsync method for simple
1518 * filesystems which track all non-inode metadata in the buffers list
1519 * hanging off the address_space structure.
1520 */
Fabian Frederickac13a822014-06-04 16:06:27 -07001521int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
1522 int datasync)
Al Virod5aacad2009-06-07 14:56:44 -04001523{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001524 struct inode *inode = file->f_mapping->host;
Al Virod5aacad2009-06-07 14:56:44 -04001525 int err;
1526 int ret;
1527
Jeff Layton383aa542017-07-06 07:02:29 -04001528 err = file_write_and_wait_range(file, start, end);
Josef Bacik02c24a82011-07-16 20:44:56 -04001529 if (err)
1530 return err;
1531
Al Viro59551022016-01-22 15:40:57 -05001532 inode_lock(inode);
Al Virod5aacad2009-06-07 14:56:44 -04001533 ret = sync_mapping_buffers(inode->i_mapping);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001534 if (!(inode->i_state & I_DIRTY_ALL))
Josef Bacik02c24a82011-07-16 20:44:56 -04001535 goto out;
Al Virod5aacad2009-06-07 14:56:44 -04001536 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
Josef Bacik02c24a82011-07-16 20:44:56 -04001537 goto out;
Al Virod5aacad2009-06-07 14:56:44 -04001538
Christoph Hellwigc37650162010-10-06 10:48:20 +02001539 err = sync_inode_metadata(inode, 1);
Al Virod5aacad2009-06-07 14:56:44 -04001540 if (ret == 0)
1541 ret = err;
Fabian Frederickac13a822014-06-04 16:06:27 -07001542
Josef Bacik02c24a82011-07-16 20:44:56 -04001543out:
Al Viro59551022016-01-22 15:40:57 -05001544 inode_unlock(inode);
Jeff Layton383aa542017-07-06 07:02:29 -04001545 /* check and advance again to catch errors after syncing out buffers */
1546 err = file_check_and_advance_wb_err(file);
1547 if (ret == 0)
1548 ret = err;
1549 return ret;
Al Virod5aacad2009-06-07 14:56:44 -04001550}
Fabian Frederickac13a822014-06-04 16:06:27 -07001551EXPORT_SYMBOL(__generic_file_fsync);
1552
1553/**
1554 * generic_file_fsync - generic fsync implementation for simple filesystems
1555 * with flush
1556 * @file: file to synchronize
1557 * @start: start offset in bytes
1558 * @end: end offset in bytes (inclusive)
1559 * @datasync: only synchronize essential metadata if true
1560 *
1561 */
1562
1563int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1564 int datasync)
1565{
1566 struct inode *inode = file->f_mapping->host;
1567 int err;
1568
1569 err = __generic_file_fsync(file, start, end, datasync);
1570 if (err)
1571 return err;
Christoph Hellwigc6bf3f02021-01-26 15:52:35 +01001572 return blkdev_issue_flush(inode->i_sb->s_bdev);
Fabian Frederickac13a822014-06-04 16:06:27 -07001573}
Christoph Hellwig1b061d92010-05-26 17:53:41 +02001574EXPORT_SYMBOL(generic_file_fsync);
1575
Patrick J. LoPresti30ca22c2010-07-22 15:03:41 -07001576/**
1577 * generic_check_addressable - Check addressability of file system
1578 * @blocksize_bits: log of file system block size
1579 * @num_blocks: number of blocks in file system
1580 *
1581 * Determine whether a file system with @num_blocks blocks (and a
1582 * block size of 2**@blocksize_bits) is addressable by the sector_t
1583 * and page cache of the system. Return 0 if so and -EFBIG otherwise.
1584 */
1585int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1586{
1587 u64 last_fs_block = num_blocks - 1;
Joel Beckera33f13e2010-08-16 12:10:17 -07001588 u64 last_fs_page =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001589 last_fs_block >> (PAGE_SHIFT - blocksize_bits);
Patrick J. LoPresti30ca22c2010-07-22 15:03:41 -07001590
1591 if (unlikely(num_blocks == 0))
1592 return 0;
1593
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001594 if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT))
Patrick J. LoPresti30ca22c2010-07-22 15:03:41 -07001595 return -EINVAL;
1596
Joel Beckera33f13e2010-08-16 12:10:17 -07001597 if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1598 (last_fs_page > (pgoff_t)(~0ULL))) {
Patrick J. LoPresti30ca22c2010-07-22 15:03:41 -07001599 return -EFBIG;
1600 }
1601 return 0;
1602}
1603EXPORT_SYMBOL(generic_check_addressable);
1604
Christoph Hellwig1b061d92010-05-26 17:53:41 +02001605/*
1606 * No-op implementation of ->fsync for in-memory filesystems.
1607 */
Josef Bacik02c24a82011-07-16 20:44:56 -04001608int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Christoph Hellwig1b061d92010-05-26 17:53:41 +02001609{
1610 return 0;
1611}
Christoph Hellwig1b061d92010-05-26 17:53:41 +02001612EXPORT_SYMBOL(noop_fsync);
Al Viro87dc8002013-09-16 10:30:04 -04001613
Dan Williamsf44c7762018-03-07 15:26:44 -08001614ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1615{
1616 /*
1617 * iomap based filesystems support direct I/O without need for
1618 * this callback. However, it still needs to be set in
1619 * inode->a_ops so that open/fcntl know that direct I/O is
1620 * generally supported.
1621 */
1622 return -EINVAL;
1623}
1624EXPORT_SYMBOL_GPL(noop_direct_IO);
1625
Al Virofceef392015-12-29 15:58:39 -05001626/* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1627void kfree_link(void *p)
Al Viro87dc8002013-09-16 10:30:04 -04001628{
Al Virofceef392015-12-29 15:58:39 -05001629 kfree(p);
Al Viro87dc8002013-09-16 10:30:04 -04001630}
Al Virofceef392015-12-29 15:58:39 -05001631EXPORT_SYMBOL(kfree_link);
Al Viro69878432013-10-02 22:35:11 -04001632
Al Viro69878432013-10-02 22:35:11 -04001633struct inode *alloc_anon_inode(struct super_block *s)
1634{
1635 static const struct address_space_operations anon_aops = {
Matthew Wilcox (Oracle)46de8b972022-02-09 20:22:13 +00001636 .dirty_folio = noop_dirty_folio,
Al Viro69878432013-10-02 22:35:11 -04001637 };
1638 struct inode *inode = new_inode_pseudo(s);
1639
1640 if (!inode)
1641 return ERR_PTR(-ENOMEM);
1642
1643 inode->i_ino = get_next_ino();
1644 inode->i_mapping->a_ops = &anon_aops;
1645
1646 /*
1647 * Mark the inode dirty from the very beginning,
1648 * that way it will never be moved to the dirty
1649 * list because mark_inode_dirty() will think
1650 * that it already _is_ on the dirty list.
1651 */
1652 inode->i_state = I_DIRTY;
1653 inode->i_mode = S_IRUSR | S_IWUSR;
1654 inode->i_uid = current_fsuid();
1655 inode->i_gid = current_fsgid();
1656 inode->i_flags |= S_PRIVATE;
Jeff Layton077c2122023-10-04 14:52:37 -04001657 simple_inode_init_ts(inode);
Al Viro69878432013-10-02 22:35:11 -04001658 return inode;
1659}
1660EXPORT_SYMBOL(alloc_anon_inode);
Jeff Layton1c994a02014-08-27 06:49:41 -04001661
1662/**
1663 * simple_nosetlease - generic helper for prohibiting leases
1664 * @filp: file pointer
1665 * @arg: type of lease to obtain
1666 * @flp: new lease supplied for insertion
Jeff Laytone6f5c782014-08-22 10:40:25 -04001667 * @priv: private data for lm_setup operation
Jeff Layton1c994a02014-08-27 06:49:41 -04001668 *
1669 * Generic helper for filesystems that do not wish to allow leases to be set.
1670 * All arguments are ignored and it just returns -EINVAL.
1671 */
1672int
Jeff Laytonc69ff402024-01-31 18:02:28 -05001673simple_nosetlease(struct file *filp, int arg, struct file_lease **flp,
Jeff Laytone6f5c782014-08-22 10:40:25 -04001674 void **priv)
Jeff Layton1c994a02014-08-27 06:49:41 -04001675{
1676 return -EINVAL;
1677}
1678EXPORT_SYMBOL(simple_nosetlease);
Al Viro61ba64f2015-05-02 09:54:06 -04001679
Eric Biggers6ee97062019-04-11 16:16:30 -07001680/**
1681 * simple_get_link - generic helper to get the target of "fast" symlinks
1682 * @dentry: not used here
1683 * @inode: the symlink inode
1684 * @done: not used here
1685 *
1686 * Generic helper for filesystems to use for symlink inodes where a pointer to
1687 * the symlink target is stored in ->i_link. NOTE: this isn't normally called,
1688 * since as an optimization the path lookup code uses any non-NULL ->i_link
1689 * directly, without calling ->get_link(). But ->get_link() still must be set,
1690 * to mark the inode_operations as being for a symlink.
1691 *
1692 * Return: the symlink target
1693 */
Al Viro6b255392015-11-17 10:20:54 -05001694const char *simple_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05001695 struct delayed_call *done)
Al Viro61ba64f2015-05-02 09:54:06 -04001696{
Al Viro6b255392015-11-17 10:20:54 -05001697 return inode->i_link;
Al Viro61ba64f2015-05-02 09:54:06 -04001698}
Al Viro6b255392015-11-17 10:20:54 -05001699EXPORT_SYMBOL(simple_get_link);
Al Viro61ba64f2015-05-02 09:54:06 -04001700
1701const struct inode_operations simple_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -05001702 .get_link = simple_get_link,
Al Viro61ba64f2015-05-02 09:54:06 -04001703};
1704EXPORT_SYMBOL(simple_symlink_inode_operations);
Eric W. Biedermanfbabfd02015-05-09 15:54:49 -05001705
1706/*
1707 * Operations for a permanently empty directory.
1708 */
1709static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1710{
1711 return ERR_PTR(-ENOENT);
1712}
1713
Christian Braunerb74d24f2023-01-13 12:49:12 +01001714static int empty_dir_getattr(struct mnt_idmap *idmap,
Christian Brauner549c7292021-01-21 14:19:43 +01001715 const struct path *path, struct kstat *stat,
David Howellsa528d352017-01-31 16:46:22 +00001716 u32 request_mask, unsigned int query_flags)
Eric W. Biedermanfbabfd02015-05-09 15:54:49 -05001717{
David Howellsa528d352017-01-31 16:46:22 +00001718 struct inode *inode = d_inode(path->dentry);
Jeff Layton0d72b9282023-08-07 15:38:33 -04001719 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
Eric W. Biedermanfbabfd02015-05-09 15:54:49 -05001720 return 0;
1721}
1722
Christian Braunerc1632a02023-01-13 12:49:11 +01001723static int empty_dir_setattr(struct mnt_idmap *idmap,
Christian Brauner549c7292021-01-21 14:19:43 +01001724 struct dentry *dentry, struct iattr *attr)
Eric W. Biedermanfbabfd02015-05-09 15:54:49 -05001725{
1726 return -EPERM;
1727}
1728
Eric W. Biedermanfbabfd02015-05-09 15:54:49 -05001729static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1730{
1731 return -EOPNOTSUPP;
1732}
1733
1734static const struct inode_operations empty_dir_inode_operations = {
1735 .lookup = empty_dir_lookup,
1736 .permission = generic_permission,
1737 .setattr = empty_dir_setattr,
1738 .getattr = empty_dir_getattr,
Eric W. Biedermanfbabfd02015-05-09 15:54:49 -05001739 .listxattr = empty_dir_listxattr,
1740};
1741
1742static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1743{
1744 /* An empty directory has two entries . and .. at offsets 0 and 1 */
1745 return generic_file_llseek_size(file, offset, whence, 2, 2);
1746}
1747
1748static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1749{
1750 dir_emit_dots(file, ctx);
1751 return 0;
1752}
1753
1754static const struct file_operations empty_dir_operations = {
1755 .llseek = empty_dir_llseek,
1756 .read = generic_read_dir,
Al Viroc51da202016-04-30 22:37:34 -04001757 .iterate_shared = empty_dir_readdir,
Eric W. Biedermanfbabfd02015-05-09 15:54:49 -05001758 .fsync = noop_fsync,
1759};
1760
1761
1762void make_empty_dir_inode(struct inode *inode)
1763{
1764 set_nlink(inode, 2);
1765 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1766 inode->i_uid = GLOBAL_ROOT_UID;
1767 inode->i_gid = GLOBAL_ROOT_GID;
1768 inode->i_rdev = 0;
Eric W. Biederman4b75de862015-08-12 15:00:12 -05001769 inode->i_size = 0;
Eric W. Biedermanfbabfd02015-05-09 15:54:49 -05001770 inode->i_blkbits = PAGE_SHIFT;
1771 inode->i_blocks = 0;
1772
1773 inode->i_op = &empty_dir_inode_operations;
Andreas Gruenbacherf5c24432016-09-29 17:48:41 +02001774 inode->i_opflags &= ~IOP_XATTR;
Eric W. Biedermanfbabfd02015-05-09 15:54:49 -05001775 inode->i_fop = &empty_dir_operations;
1776}
1777
1778bool is_empty_dir_inode(struct inode *inode)
1779{
1780 return (inode->i_fop == &empty_dir_operations) &&
1781 (inode->i_op == &empty_dir_inode_operations);
1782}
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001783
Christoph Hellwig5298d4b2022-01-18 07:56:14 +01001784#if IS_ENABLED(CONFIG_UNICODE)
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001785/**
1786 * generic_ci_d_compare - generic d_compare implementation for casefolding filesystems
1787 * @dentry: dentry whose name we are checking against
1788 * @len: len of name of dentry
1789 * @str: str pointer to name of dentry
1790 * @name: Name to compare against
1791 *
1792 * Return: 0 if names match, 1 if mismatch, or -ERRNO
1793 */
Eric Biggers794c43f2020-12-28 15:25:29 -08001794static int generic_ci_d_compare(const struct dentry *dentry, unsigned int len,
1795 const char *str, const struct qstr *name)
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001796{
Gabriel Krisman Bertazi0906fbb2024-01-24 15:13:40 -03001797 const struct dentry *parent;
1798 const struct inode *dir;
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001799 char strbuf[DNAME_INLINE_LEN];
Gabriel Krisman Bertazi0906fbb2024-01-24 15:13:40 -03001800 struct qstr qstr;
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001801
Gabriel Krisman Bertazi0906fbb2024-01-24 15:13:40 -03001802 /*
1803 * Attempt a case-sensitive match first. It is cheaper and
1804 * should cover most lookups, including all the sane
1805 * applications that expect a case-sensitive filesystem.
1806 *
1807 * This comparison is safe under RCU because the caller
1808 * guarantees the consistency between str and len. See
1809 * __d_lookup_rcu_op_compare() for details.
1810 */
1811 if (len == name->len && !memcmp(str, name->name, len))
1812 return 0;
1813
1814 parent = READ_ONCE(dentry->d_parent);
1815 dir = READ_ONCE(parent->d_inode);
Eric Biggersaf494af2023-08-14 11:29:03 -07001816 if (!dir || !IS_CASEFOLDED(dir))
Gabriel Krisman Bertazi0906fbb2024-01-24 15:13:40 -03001817 return 1;
1818
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001819 /*
1820 * If the dentry name is stored in-line, then it may be concurrently
1821 * modified by a rename. If this happens, the VFS will eventually retry
1822 * the lookup, so it doesn't matter what ->d_compare() returns.
1823 * However, it's unsafe to call utf8_strncasecmp() with an unstable
1824 * string. Therefore, we have to copy the name into a temporary buffer.
1825 */
1826 if (len <= DNAME_INLINE_LEN - 1) {
1827 memcpy(strbuf, str, len);
1828 strbuf[len] = 0;
Gabriel Krisman Bertazi0906fbb2024-01-24 15:13:40 -03001829 str = strbuf;
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001830 /* prevent compiler from optimizing out the temporary buffer */
1831 barrier();
1832 }
Gabriel Krisman Bertazi0906fbb2024-01-24 15:13:40 -03001833 qstr.len = len;
1834 qstr.name = str;
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001835
Gabriel Krisman Bertazi0906fbb2024-01-24 15:13:40 -03001836 return utf8_strncasecmp(dentry->d_sb->s_encoding, name, &qstr);
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001837}
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001838
1839/**
1840 * generic_ci_d_hash - generic d_hash implementation for casefolding filesystems
1841 * @dentry: dentry of the parent directory
1842 * @str: qstr of name whose hash we should fill in
1843 *
1844 * Return: 0 if hash was successful or unchanged, and -EINVAL on error
1845 */
Eric Biggers794c43f2020-12-28 15:25:29 -08001846static int generic_ci_d_hash(const struct dentry *dentry, struct qstr *str)
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001847{
1848 const struct inode *dir = READ_ONCE(dentry->d_inode);
1849 struct super_block *sb = dentry->d_sb;
1850 const struct unicode_map *um = sb->s_encoding;
Li zemingbae8bc42024-02-20 14:20:30 +08001851 int ret;
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001852
Eric Biggersaf494af2023-08-14 11:29:03 -07001853 if (!dir || !IS_CASEFOLDED(dir))
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001854 return 0;
1855
1856 ret = utf8_casefold_hash(um, dentry, str);
1857 if (ret < 0 && sb_has_strict_encoding(sb))
1858 return -EINVAL;
1859 return 0;
1860}
Daniel Rosenberg608af702020-11-19 06:09:02 +00001861
1862static const struct dentry_operations generic_ci_dentry_ops = {
1863 .d_hash = generic_ci_d_hash,
1864 .d_compare = generic_ci_d_compare,
Gabriel Krisman Bertazie6ca2882024-02-21 12:14:07 -05001865#ifdef CONFIG_FS_ENCRYPTION
1866 .d_revalidate = fscrypt_d_revalidate,
1867#endif
Daniel Rosenberg608af702020-11-19 06:09:02 +00001868};
Gabriel Krisman Bertazi6a79a4e2024-06-06 10:33:49 +03001869
1870/**
1871 * generic_ci_match() - Match a name (case-insensitively) with a dirent.
1872 * This is a filesystem helper for comparison with directory entries.
1873 * generic_ci_d_compare should be used in VFS' ->d_compare instead.
1874 *
1875 * @parent: Inode of the parent of the dirent under comparison
1876 * @name: name under lookup.
1877 * @folded_name: Optional pre-folded name under lookup
1878 * @de_name: Dirent name.
1879 * @de_name_len: dirent name length.
1880 *
1881 * Test whether a case-insensitive directory entry matches the filename
1882 * being searched. If @folded_name is provided, it is used instead of
1883 * recalculating the casefold of @name.
1884 *
1885 * Return: > 0 if the directory entry matches, 0 if it doesn't match, or
1886 * < 0 on error.
1887 */
1888int generic_ci_match(const struct inode *parent,
1889 const struct qstr *name,
1890 const struct qstr *folded_name,
1891 const u8 *de_name, u32 de_name_len)
1892{
1893 const struct super_block *sb = parent->i_sb;
1894 const struct unicode_map *um = sb->s_encoding;
1895 struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1896 struct qstr dirent = QSTR_INIT(de_name, de_name_len);
1897 int res = 0;
1898
1899 if (IS_ENCRYPTED(parent)) {
1900 const struct fscrypt_str encrypted_name =
1901 FSTR_INIT((u8 *) de_name, de_name_len);
1902
1903 if (WARN_ON_ONCE(!fscrypt_has_encryption_key(parent)))
1904 return -EINVAL;
1905
1906 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1907 if (!decrypted_name.name)
1908 return -ENOMEM;
1909 res = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1910 &decrypted_name);
1911 if (res < 0) {
1912 kfree(decrypted_name.name);
1913 return res;
1914 }
1915 dirent.name = decrypted_name.name;
1916 dirent.len = decrypted_name.len;
1917 }
1918
1919 /*
1920 * Attempt a case-sensitive match first. It is cheaper and
1921 * should cover most lookups, including all the sane
1922 * applications that expect a case-sensitive filesystem.
1923 */
1924
1925 if (dirent.len == name->len &&
1926 !memcmp(name->name, dirent.name, dirent.len))
1927 goto out;
1928
1929 if (folded_name->name)
1930 res = utf8_strncasecmp_folded(um, folded_name, &dirent);
1931 else
1932 res = utf8_strncasecmp(um, name, &dirent);
1933
1934out:
1935 kfree(decrypted_name.name);
1936 if (res < 0 && sb_has_strict_encoding(sb)) {
1937 pr_err_ratelimited("Directory contains filename that is invalid UTF-8");
1938 return 0;
1939 }
1940 return !res;
1941}
1942EXPORT_SYMBOL(generic_ci_match);
Daniel Rosenbergc8438432020-07-08 02:12:35 -07001943#endif
Daniel Rosenberg608af702020-11-19 06:09:02 +00001944
1945#ifdef CONFIG_FS_ENCRYPTION
1946static const struct dentry_operations generic_encrypted_dentry_ops = {
1947 .d_revalidate = fscrypt_d_revalidate,
1948};
1949#endif
1950
Daniel Rosenberg608af702020-11-19 06:09:02 +00001951/**
Gabriel Krisman Bertazi70dfe3f2024-02-21 12:14:08 -05001952 * generic_set_sb_d_ops - helper for choosing the set of
1953 * filesystem-wide dentry operations for the enabled features
1954 * @sb: superblock to be configured
Daniel Rosenberg608af702020-11-19 06:09:02 +00001955 *
Gabriel Krisman Bertazi70dfe3f2024-02-21 12:14:08 -05001956 * Filesystems supporting casefolding and/or fscrypt can call this
1957 * helper at mount-time to configure sb->s_d_op to best set of dentry
1958 * operations required for the enabled features. The helper must be
1959 * called after these have been configured, but before the root dentry
1960 * is created.
Daniel Rosenberg608af702020-11-19 06:09:02 +00001961 */
Gabriel Krisman Bertazi70dfe3f2024-02-21 12:14:08 -05001962void generic_set_sb_d_ops(struct super_block *sb)
Daniel Rosenberg608af702020-11-19 06:09:02 +00001963{
Christoph Hellwig5298d4b2022-01-18 07:56:14 +01001964#if IS_ENABLED(CONFIG_UNICODE)
Gabriel Krisman Bertazi70dfe3f2024-02-21 12:14:08 -05001965 if (sb->s_encoding) {
1966 sb->s_d_op = &generic_ci_dentry_ops;
Daniel Rosenberg608af702020-11-19 06:09:02 +00001967 return;
1968 }
1969#endif
1970#ifdef CONFIG_FS_ENCRYPTION
Gabriel Krisman Bertazi70dfe3f2024-02-21 12:14:08 -05001971 if (sb->s_cop) {
1972 sb->s_d_op = &generic_encrypted_dentry_ops;
Daniel Rosenberg608af702020-11-19 06:09:02 +00001973 return;
1974 }
1975#endif
1976}
Gabriel Krisman Bertazi70dfe3f2024-02-21 12:14:08 -05001977EXPORT_SYMBOL(generic_set_sb_d_ops);
Andrew Morton5ca14832022-09-09 13:57:41 -07001978
1979/**
1980 * inode_maybe_inc_iversion - increments i_version
1981 * @inode: inode with the i_version that should be updated
1982 * @force: increment the counter even if it's not necessary?
1983 *
1984 * Every time the inode is modified, the i_version field must be seen to have
1985 * changed by any observer.
1986 *
1987 * If "force" is set or the QUERIED flag is set, then ensure that we increment
1988 * the value, and clear the queried flag.
1989 *
1990 * In the common case where neither is set, then we can return "false" without
1991 * updating i_version.
1992 *
1993 * If this function returns false, and no other metadata has changed, then we
1994 * can avoid logging the metadata.
1995 */
1996bool inode_maybe_inc_iversion(struct inode *inode, bool force)
1997{
1998 u64 cur, new;
1999
2000 /*
2001 * The i_version field is not strictly ordered with any other inode
2002 * information, but the legacy inode_inc_iversion code used a spinlock
2003 * to serialize increments.
2004 *
Mateusz Guzikb381fbb2024-08-15 10:33:10 +02002005 * We add a full memory barrier to ensure that any de facto ordering
2006 * with other state is preserved (either implicitly coming from cmpxchg
2007 * or explicitly from smp_mb if we don't know upfront if we will execute
2008 * the former).
Andrew Morton5ca14832022-09-09 13:57:41 -07002009 *
Mateusz Guzikb381fbb2024-08-15 10:33:10 +02002010 * These barriers pair with inode_query_iversion().
Andrew Morton5ca14832022-09-09 13:57:41 -07002011 */
Andrew Morton5ca14832022-09-09 13:57:41 -07002012 cur = inode_peek_iversion_raw(inode);
Mateusz Guzikb381fbb2024-08-15 10:33:10 +02002013 if (!force && !(cur & I_VERSION_QUERIED)) {
2014 smp_mb();
2015 cur = inode_peek_iversion_raw(inode);
2016 }
2017
Andrew Morton5ca14832022-09-09 13:57:41 -07002018 do {
2019 /* If flag is clear then we needn't do anything */
2020 if (!force && !(cur & I_VERSION_QUERIED))
2021 return false;
2022
2023 /* Since lowest bit is flag, add 2 to avoid it */
2024 new = (cur & ~I_VERSION_QUERIED) + I_VERSION_INCREMENT;
2025 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
2026 return true;
2027}
2028EXPORT_SYMBOL(inode_maybe_inc_iversion);
Jeff Laytonc5bc1b32022-09-16 09:37:51 -04002029
2030/**
2031 * inode_query_iversion - read i_version for later use
2032 * @inode: inode from which i_version should be read
2033 *
2034 * Read the inode i_version counter. This should be used by callers that wish
2035 * to store the returned i_version for later comparison. This will guarantee
2036 * that a later query of the i_version will result in a different value if
2037 * anything has changed.
2038 *
2039 * In this implementation, we fetch the current value, set the QUERIED flag and
2040 * then try to swap it into place with a cmpxchg, if it wasn't already set. If
2041 * that fails, we try again with the newly fetched value from the cmpxchg.
2042 */
2043u64 inode_query_iversion(struct inode *inode)
2044{
2045 u64 cur, new;
Mateusz Guzikb381fbb2024-08-15 10:33:10 +02002046 bool fenced = false;
Jeff Laytonc5bc1b32022-09-16 09:37:51 -04002047
Mateusz Guzikb381fbb2024-08-15 10:33:10 +02002048 /*
2049 * Memory barriers (implicit in cmpxchg, explicit in smp_mb) pair with
2050 * inode_maybe_inc_iversion(), see that routine for more details.
2051 */
Jeff Laytonc5bc1b32022-09-16 09:37:51 -04002052 cur = inode_peek_iversion_raw(inode);
2053 do {
2054 /* If flag is already set, then no need to swap */
2055 if (cur & I_VERSION_QUERIED) {
Mateusz Guzikb381fbb2024-08-15 10:33:10 +02002056 if (!fenced)
2057 smp_mb();
Jeff Laytonc5bc1b32022-09-16 09:37:51 -04002058 break;
2059 }
2060
Mateusz Guzikb381fbb2024-08-15 10:33:10 +02002061 fenced = true;
Jeff Laytonc5bc1b32022-09-16 09:37:51 -04002062 new = cur | I_VERSION_QUERIED;
2063 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
2064 return cur >> I_VERSION_QUERIED_SHIFT;
2065}
2066EXPORT_SYMBOL(inode_query_iversion);
Christoph Hellwig44fff0f2023-06-01 16:59:01 +02002067
2068ssize_t direct_write_fallback(struct kiocb *iocb, struct iov_iter *iter,
2069 ssize_t direct_written, ssize_t buffered_written)
2070{
2071 struct address_space *mapping = iocb->ki_filp->f_mapping;
2072 loff_t pos = iocb->ki_pos - buffered_written;
2073 loff_t end = iocb->ki_pos - 1;
2074 int err;
2075
2076 /*
2077 * If the buffered write fallback returned an error, we want to return
2078 * the number of bytes which were written by direct I/O, or the error
2079 * code if that was zero.
2080 *
2081 * Note that this differs from normal direct-io semantics, which will
2082 * return -EFOO even if some bytes were written.
2083 */
2084 if (unlikely(buffered_written < 0)) {
2085 if (direct_written)
2086 return direct_written;
2087 return buffered_written;
2088 }
2089
2090 /*
2091 * We need to ensure that the page cache pages are written to disk and
2092 * invalidated to preserve the expected O_DIRECT semantics.
2093 */
2094 err = filemap_write_and_wait_range(mapping, pos, end);
2095 if (err < 0) {
2096 /*
2097 * We don't know how much we wrote, so just return the number of
2098 * bytes which were direct-written
2099 */
Al Viro82874742023-09-13 18:28:15 +02002100 iocb->ki_pos -= buffered_written;
Christoph Hellwig44fff0f2023-06-01 16:59:01 +02002101 if (direct_written)
2102 return direct_written;
2103 return err;
2104 }
2105 invalidate_mapping_pages(mapping, pos >> PAGE_SHIFT, end >> PAGE_SHIFT);
2106 return direct_written + buffered_written;
2107}
2108EXPORT_SYMBOL_GPL(direct_write_fallback);
Jeff Layton077c2122023-10-04 14:52:37 -04002109
2110/**
2111 * simple_inode_init_ts - initialize the timestamps for a new inode
2112 * @inode: inode to be initialized
2113 *
2114 * When a new inode is created, most filesystems set the timestamps to the
2115 * current time. Add a helper to do this.
2116 */
2117struct timespec64 simple_inode_init_ts(struct inode *inode)
2118{
2119 struct timespec64 ts = inode_set_ctime_current(inode);
2120
2121 inode_set_atime_to_ts(inode, ts);
2122 inode_set_mtime_to_ts(inode, ts);
2123 return ts;
2124}
2125EXPORT_SYMBOL(simple_inode_init_ts);
Christian Brauner07fd7c32024-02-18 14:50:13 +01002126
Christian Brauner4e32c252024-09-06 18:22:22 +02002127static inline struct dentry *get_stashed_dentry(struct dentry **stashed)
Christian Brauner07fd7c32024-02-18 14:50:13 +01002128{
2129 struct dentry *dentry;
2130
2131 guard(rcu)();
Christian Brauner4e32c252024-09-06 18:22:22 +02002132 dentry = rcu_dereference(*stashed);
Christian Brauner07fd7c32024-02-18 14:50:13 +01002133 if (!dentry)
2134 return NULL;
2135 if (!lockref_get_not_dead(&dentry->d_lockref))
2136 return NULL;
2137 return dentry;
2138}
2139
Christian Brauner2558e3b2024-02-21 09:59:51 +01002140static struct dentry *prepare_anon_dentry(struct dentry **stashed,
Christian Brauner159a0d92024-02-18 14:52:24 +01002141 struct super_block *sb,
Christian Brauner159a0d92024-02-18 14:52:24 +01002142 void *data)
Christian Brauner07fd7c32024-02-18 14:50:13 +01002143{
2144 struct dentry *dentry;
2145 struct inode *inode;
Christian Braunere9c52632024-03-01 10:26:03 +01002146 const struct stashed_operations *sops = sb->s_fs_info;
Christian Brauner9d9539d2024-03-12 10:39:44 +01002147 int ret;
Christian Brauner07fd7c32024-02-18 14:50:13 +01002148
2149 inode = new_inode_pseudo(sb);
2150 if (!inode) {
Christian Brauner9d9539d2024-03-12 10:39:44 +01002151 sops->put_data(data);
Christian Brauner07fd7c32024-02-18 14:50:13 +01002152 return ERR_PTR(-ENOMEM);
2153 }
2154
Christian Brauner07fd7c32024-02-18 14:50:13 +01002155 inode->i_flags |= S_IMMUTABLE;
Christian Braunere9c52632024-03-01 10:26:03 +01002156 inode->i_mode = S_IFREG;
Christian Brauner07fd7c32024-02-18 14:50:13 +01002157 simple_inode_init_ts(inode);
Christian Brauner9d9539d2024-03-12 10:39:44 +01002158
2159 ret = sops->init_inode(inode, data);
2160 if (ret < 0) {
2161 iput(inode);
2162 return ERR_PTR(ret);
2163 }
Christian Braunere9c52632024-03-01 10:26:03 +01002164
2165 /* Notice when this is changed. */
2166 WARN_ON_ONCE(!S_ISREG(inode->i_mode));
2167 WARN_ON_ONCE(!IS_IMMUTABLE(inode));
Christian Brauner07fd7c32024-02-18 14:50:13 +01002168
Christian Brauner9d9539d2024-03-12 10:39:44 +01002169 dentry = d_alloc_anon(sb);
2170 if (!dentry) {
2171 iput(inode);
2172 return ERR_PTR(-ENOMEM);
2173 }
2174
Christian Brauner2558e3b2024-02-21 09:59:51 +01002175 /* Store address of location where dentry's supposed to be stashed. */
2176 dentry->d_fsdata = stashed;
2177
Christian Brauner07fd7c32024-02-18 14:50:13 +01002178 /* @data is now owned by the fs */
2179 d_instantiate(dentry, inode);
Christian Brauner07fd7c32024-02-18 14:50:13 +01002180 return dentry;
2181}
2182
Christian Brauner159a0d92024-02-18 14:52:24 +01002183static struct dentry *stash_dentry(struct dentry **stashed,
2184 struct dentry *dentry)
2185{
2186 guard(rcu)();
2187 for (;;) {
2188 struct dentry *old;
2189
2190 /* Assume any old dentry was cleared out. */
2191 old = cmpxchg(stashed, NULL, dentry);
2192 if (likely(!old))
2193 return dentry;
2194
2195 /* Check if somebody else installed a reusable dentry. */
2196 if (lockref_get_not_dead(&old->d_lockref))
2197 return old;
2198
2199 /* There's an old dead dentry there, try to take it over. */
2200 if (likely(try_cmpxchg(stashed, &old, dentry)))
2201 return dentry;
2202 }
2203}
2204
Christian Brauner07fd7c32024-02-18 14:50:13 +01002205/**
2206 * path_from_stashed - create path from stashed or new dentry
2207 * @stashed: where to retrieve or stash dentry
Christian Brauner07fd7c32024-02-18 14:50:13 +01002208 * @mnt: mnt of the filesystems to use
Christian Brauner07fd7c32024-02-18 14:50:13 +01002209 * @data: data to store in inode->i_private
2210 * @path: path to create
2211 *
2212 * The function tries to retrieve a stashed dentry from @stashed. If the dentry
2213 * is still valid then it will be reused. If the dentry isn't able the function
Christian Brauner159a0d92024-02-18 14:52:24 +01002214 * will allocate a new dentry and inode. It will then check again whether it
2215 * can reuse an existing dentry in case one has been added in the meantime or
2216 * update @stashed with the newly added dentry.
Christian Brauner07fd7c32024-02-18 14:50:13 +01002217 *
2218 * Special-purpose helper for nsfs and pidfs.
2219 *
Christian Braunere9c52632024-03-01 10:26:03 +01002220 * Return: On success zero and on failure a negative error is returned.
Christian Brauner07fd7c32024-02-18 14:50:13 +01002221 */
Christian Brauner9d9539d2024-03-12 10:39:44 +01002222int path_from_stashed(struct dentry **stashed, struct vfsmount *mnt, void *data,
2223 struct path *path)
Christian Brauner07fd7c32024-02-18 14:50:13 +01002224{
2225 struct dentry *dentry;
Christian Braunere9c52632024-03-01 10:26:03 +01002226 const struct stashed_operations *sops = mnt->mnt_sb->s_fs_info;
Christian Brauner07fd7c32024-02-18 14:50:13 +01002227
Christian Brauner159a0d92024-02-18 14:52:24 +01002228 /* See if dentry can be reused. */
Christian Brauner4e32c252024-09-06 18:22:22 +02002229 path->dentry = get_stashed_dentry(stashed);
Christian Braunere9c52632024-03-01 10:26:03 +01002230 if (path->dentry) {
2231 sops->put_data(data);
Christian Brauner07fd7c32024-02-18 14:50:13 +01002232 goto out_path;
Christian Braunere9c52632024-03-01 10:26:03 +01002233 }
Christian Brauner07fd7c32024-02-18 14:50:13 +01002234
Christian Brauner159a0d92024-02-18 14:52:24 +01002235 /* Allocate a new dentry. */
Christian Brauner9d9539d2024-03-12 10:39:44 +01002236 dentry = prepare_anon_dentry(stashed, mnt->mnt_sb, data);
2237 if (IS_ERR(dentry))
Christian Brauner07fd7c32024-02-18 14:50:13 +01002238 return PTR_ERR(dentry);
Christian Brauner159a0d92024-02-18 14:52:24 +01002239
2240 /* Added a new dentry. @data is now owned by the filesystem. */
2241 path->dentry = stash_dentry(stashed, dentry);
2242 if (path->dentry != dentry)
2243 dput(dentry);
Christian Brauner07fd7c32024-02-18 14:50:13 +01002244
2245out_path:
Christian Brauner2558e3b2024-02-21 09:59:51 +01002246 WARN_ON_ONCE(path->dentry->d_fsdata != stashed);
2247 WARN_ON_ONCE(d_inode(path->dentry)->i_private != data);
Christian Brauner07fd7c32024-02-18 14:50:13 +01002248 path->mnt = mntget(mnt);
Christian Braunere9c52632024-03-01 10:26:03 +01002249 return 0;
Christian Brauner07fd7c32024-02-18 14:50:13 +01002250}
Christian Brauner2558e3b2024-02-21 09:59:51 +01002251
2252void stashed_dentry_prune(struct dentry *dentry)
2253{
2254 struct dentry **stashed = dentry->d_fsdata;
2255 struct inode *inode = d_inode(dentry);
2256
2257 if (WARN_ON_ONCE(!stashed))
2258 return;
2259
2260 if (!inode)
2261 return;
2262
2263 /*
2264 * Only replace our own @dentry as someone else might've
2265 * already cleared out @dentry and stashed their own
2266 * dentry in there.
2267 */
2268 cmpxchg(stashed, dentry, NULL);
2269}