| 1 | |
| 2 | |
| 3 | * Resizable virtual memory filesystem for Linux.
|
| 4 | *
|
| 5 | * Copyright (C) 2000 Linus Torvalds.
|
| 6 | * 2000 Transmeta Corp.
|
| 7 | * 2000-2001 Christoph Rohland
|
| 8 | * 2000-2001 SAP AG
|
| 9 | * 2002 Red Hat Inc.
|
| 10 | * Copyright (C) 2002-2011 Hugh Dickins.
|
| 11 | * Copyright (C) 2011 Google Inc.
|
| 12 | * Copyright (C) 2002-2005 VERITAS Software Corporation.
|
| 13 | * Copyright (C) 2004 Andi Kleen, SuSE Labs
|
| 14 | *
|
| 15 | * Extended attribute support for tmpfs:
|
| 16 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <[email protected]>
|
| 17 | * Copyright (c) 2004 Red Hat, Inc., James Morris <[email protected]>
|
| 18 | *
|
| 19 | * tiny-shmem:
|
| 20 | * Copyright (c) 2004, 2008 Matt Mackall <[email protected]>
|
| 21 |
|
| 22 |
|
| 23 | #include <linux/fs.h>
|
| 24 | #include <linux/init.h>
|
| 25 | #include <linux/vfs.h>
|
| 26 | #include <linux/mount.h>
|
| 27 | #include <linux/ramfs.h>
|
| 28 | #include <linux/pagemap.h>
|
| 29 | #include <linux/file.h>
|
| 30 | #include <linux/fileattr.h>
|
| 31 | #include <linux/filelock.h>
|
| 32 | #include <linux/mm.h>
|
| 33 | #include <linux/random.h>
|
| 34 | #include <linux/sched/signal.h>
|
| 35 | #include <linux/export.h>
|
| 36 | #include <linux/shmem_fs.h>
|
| 37 | #include <linux/swap.h>
|
| 38 | #include <linux/uio.h>
|
| 39 | #include <linux/hugetlb.h>
|
| 40 | #include <linux/fs_parser.h>
|
| 41 | #include <linux/swapfile.h>
|
| 42 | #include <linux/iversion.h>
|
| 43 | #include <linux/unicode.h>
|
| 44 | #include <linux/swap_ops.h>
|
| 45 | #include "swap.h"
|
| 46 |
|
| 47 | static struct vfsmount *shm_mnt __ro_after_init;
|
| 48 |
|
| 49 | #ifdef CONFIG_SHMEM
|
| 50 | |
| 51 | * This virtual memory filesystem is heavily based on the ramfs. It
|
| 52 | * extends ramfs by the ability to use swap and honor resource limits
|
| 53 | * which makes it a completely usable filesystem.
|
| 54 |
|
| 55 |
|
| 56 | #include <linux/xattr.h>
|
| 57 | #include <linux/exportfs.h>
|
| 58 | #include <linux/posix_acl.h>
|
| 59 | #include <linux/posix_acl_xattr.h>
|
| 60 | #include <linux/mman.h>
|
| 61 | #include <linux/string.h>
|
| 62 | #include <linux/slab.h>
|
| 63 | #include <linux/backing-dev.h>
|
| 64 | #include <linux/writeback.h>
|
| 65 | #include <linux/folio_batch.h>
|
| 66 | #include <linux/percpu_counter.h>
|
| 67 | #include <linux/falloc.h>
|
| 68 | #include <linux/splice.h>
|
| 69 | #include <linux/security.h>
|
| 70 | #include <linux/leafops.h>
|
| 71 | #include <linux/mempolicy.h>
|
| 72 | #include <linux/namei.h>
|
| 73 | #include <linux/ctype.h>
|
| 74 | #include <linux/migrate.h>
|
| 75 | #include <linux/highmem.h>
|
| 76 | #include <linux/seq_file.h>
|
| 77 | #include <linux/magic.h>
|
| 78 | #include <linux/syscalls.h>
|
| 79 | #include <linux/fcntl.h>
|
| 80 | #include <uapi/linux/memfd.h>
|
| 81 | #include <linux/rmap.h>
|
| 82 | #include <linux/uuid.h>
|
| 83 | #include <linux/quotaops.h>
|
| 84 | #include <linux/rcupdate_wait.h>
|
| 85 |
|
| 86 | #include <linux/uaccess.h>
|
| 87 |
|
| 88 | #include "internal.h"
|
| 89 |
|
| 90 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
|
| 91 |
|
| 92 |
|
| 93 | #define BOGO_DIRENT_SIZE 20
|
| 94 |
|
| 95 |
|
| 96 | #define BOGO_INODE_SIZE 1024
|
| 97 |
|
| 98 |
|
| 99 | #define SHORT_SYMLINK_LEN 128
|
| 100 |
|
| 101 | |
| 102 | * shmem_fallocate communicates with shmem_fault or shmem_writeout via
|
| 103 | * inode->i_private (with i_rwsem making sure that it has only one user at
|
| 104 | * a time): we would prefer not to enlarge the shmem inode just for that.
|
| 105 |
|
| 106 | struct shmem_falloc {
|
| 107 | wait_queue_head_t *waitq;
|
| 108 | pgoff_t start;
|
| 109 | pgoff_t next;
|
| 110 | pgoff_t nr_falloced;
|
| 111 | pgoff_t nr_unswapped;
|
| 112 | };
|
| 113 |
|
| 114 | struct shmem_options {
|
| 115 | unsigned long long blocks;
|
| 116 | unsigned long long inodes;
|
| 117 | struct mempolicy *mpol;
|
| 118 | kuid_t uid;
|
| 119 | kgid_t gid;
|
| 120 | umode_t mode;
|
| 121 | bool full_inums;
|
| 122 | int huge;
|
| 123 | int seen;
|
| 124 | bool noswap;
|
| 125 | unsigned short quota_types;
|
| 126 | struct shmem_quota_limits qlimits;
|
| 127 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 128 | struct unicode_map *encoding;
|
| 129 | bool strict_encoding;
|
| 130 | #endif
|
| 131 | #define SHMEM_SEEN_BLOCKS 1
|
| 132 | #define SHMEM_SEEN_INODES 2
|
| 133 | #define SHMEM_SEEN_HUGE 4
|
| 134 | #define SHMEM_SEEN_INUMS 8
|
| 135 | #define SHMEM_SEEN_QUOTA 16
|
| 136 | };
|
| 137 |
|
| 138 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 139 | static unsigned long huge_shmem_orders_always __read_mostly;
|
| 140 | static unsigned long huge_shmem_orders_madvise __read_mostly;
|
| 141 | static unsigned long huge_shmem_orders_inherit __read_mostly;
|
| 142 | static unsigned long huge_shmem_orders_within_size __read_mostly;
|
| 143 | static bool shmem_orders_configured __initdata;
|
| 144 | #endif
|
| 145 |
|
| 146 | #ifdef CONFIG_TMPFS
|
| 147 | static unsigned long shmem_default_max_blocks(void)
|
| 148 | {
|
| 149 | return totalram_pages() / 2;
|
| 150 | }
|
| 151 |
|
| 152 | static unsigned long shmem_default_max_inodes(void)
|
| 153 | {
|
| 154 | unsigned long nr_pages = totalram_pages();
|
| 155 |
|
| 156 | return min3(nr_pages - totalhigh_pages(), nr_pages / 2,
|
| 157 | ULONG_MAX / BOGO_INODE_SIZE);
|
| 158 | }
|
| 159 | #endif
|
| 160 |
|
| 161 | static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
|
| 162 | struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
|
| 163 | struct vm_fault *vmf, vm_fault_t *fault_type);
|
| 164 |
|
| 165 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
|
| 166 | {
|
| 167 | return sb->s_fs_info;
|
| 168 | }
|
| 169 |
|
| 170 | |
| 171 | * shmem_file_setup pre-accounts the whole fixed size of a VM object,
|
| 172 | * for shared memory and for shared anonymous (/dev/zero) mappings
|
| 173 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
|
| 174 | * consistent with the pre-accounting of private mappings ...
|
| 175 |
|
| 176 | static inline int shmem_acct_size(unsigned long flags, loff_t size)
|
| 177 | {
|
| 178 | return (flags & SHMEM_F_NORESERVE) ?
|
| 179 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
|
| 180 | }
|
| 181 |
|
| 182 | static inline void shmem_unacct_size(unsigned long flags, loff_t size)
|
| 183 | {
|
| 184 | if (!(flags & SHMEM_F_NORESERVE))
|
| 185 | vm_unacct_memory(VM_ACCT(size));
|
| 186 | }
|
| 187 |
|
| 188 | static inline int shmem_reacct_size(unsigned long flags,
|
| 189 | loff_t oldsize, loff_t newsize)
|
| 190 | {
|
| 191 | if (!(flags & SHMEM_F_NORESERVE)) {
|
| 192 | if (VM_ACCT(newsize) > VM_ACCT(oldsize))
|
| 193 | return security_vm_enough_memory_mm(current->mm,
|
| 194 | VM_ACCT(newsize) - VM_ACCT(oldsize));
|
| 195 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
|
| 196 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
|
| 197 | }
|
| 198 | return 0;
|
| 199 | }
|
| 200 |
|
| 201 | |
| 202 | * ... whereas tmpfs objects are accounted incrementally as
|
| 203 | * pages are allocated, in order to allow large sparse files.
|
| 204 | * shmem_get_folio reports shmem_acct_blocks failure as -ENOSPC not -ENOMEM,
|
| 205 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
|
| 206 |
|
| 207 | static inline int shmem_acct_blocks(unsigned long flags, long pages)
|
| 208 | {
|
| 209 | if (!(flags & SHMEM_F_NORESERVE))
|
| 210 | return 0;
|
| 211 |
|
| 212 | return security_vm_enough_memory_mm(current->mm,
|
| 213 | pages * VM_ACCT(PAGE_SIZE));
|
| 214 | }
|
| 215 |
|
| 216 | static inline void shmem_unacct_blocks(unsigned long flags, long pages)
|
| 217 | {
|
| 218 | if (flags & SHMEM_F_NORESERVE)
|
| 219 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
|
| 220 | }
|
| 221 |
|
| 222 | int shmem_inode_acct_blocks(struct inode *inode, long pages)
|
| 223 | {
|
| 224 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 225 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 226 | int err = -ENOSPC;
|
| 227 |
|
| 228 | if (shmem_acct_blocks(info->flags, pages))
|
| 229 | return err;
|
| 230 |
|
| 231 | might_sleep();
|
| 232 | if (sbinfo->max_blocks) {
|
| 233 | if (!percpu_counter_limited_add(&sbinfo->used_blocks,
|
| 234 | sbinfo->max_blocks, pages))
|
| 235 | goto unacct;
|
| 236 |
|
| 237 | err = dquot_alloc_block_nodirty(inode, pages);
|
| 238 | if (err) {
|
| 239 | percpu_counter_sub(&sbinfo->used_blocks, pages);
|
| 240 | goto unacct;
|
| 241 | }
|
| 242 | } else {
|
| 243 | err = dquot_alloc_block_nodirty(inode, pages);
|
| 244 | if (err)
|
| 245 | goto unacct;
|
| 246 | }
|
| 247 |
|
| 248 | return 0;
|
| 249 |
|
| 250 | unacct:
|
| 251 | shmem_unacct_blocks(info->flags, pages);
|
| 252 | return err;
|
| 253 | }
|
| 254 |
|
| 255 | static void shmem_inode_unacct_blocks(struct inode *inode, long pages)
|
| 256 | {
|
| 257 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 258 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 259 |
|
| 260 | might_sleep();
|
| 261 | dquot_free_block_nodirty(inode, pages);
|
| 262 |
|
| 263 | if (sbinfo->max_blocks)
|
| 264 | percpu_counter_sub(&sbinfo->used_blocks, pages);
|
| 265 | shmem_unacct_blocks(info->flags, pages);
|
| 266 | }
|
| 267 |
|
| 268 | static const struct super_operations shmem_ops;
|
| 269 | static const struct address_space_operations shmem_aops;
|
| 270 | static const struct file_operations shmem_file_operations;
|
| 271 | static const struct inode_operations shmem_inode_operations;
|
| 272 | static const struct inode_operations shmem_dir_inode_operations;
|
| 273 | static const struct inode_operations shmem_special_inode_operations;
|
| 274 | static const struct vm_operations_struct shmem_vm_ops;
|
| 275 | static const struct vm_operations_struct shmem_anon_vm_ops;
|
| 276 | static struct file_system_type shmem_fs_type;
|
| 277 |
|
| 278 | bool shmem_mapping(const struct address_space *mapping)
|
| 279 | {
|
| 280 | return mapping->a_ops == &shmem_aops;
|
| 281 | }
|
| 282 | EXPORT_SYMBOL_GPL(shmem_mapping);
|
| 283 |
|
| 284 | bool vma_is_anon_shmem(const struct vm_area_struct *vma)
|
| 285 | {
|
| 286 | return vma->vm_ops == &shmem_anon_vm_ops;
|
| 287 | }
|
| 288 |
|
| 289 | bool vma_is_shmem(const struct vm_area_struct *vma)
|
| 290 | {
|
| 291 | return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
|
| 292 | }
|
| 293 |
|
| 294 | static LIST_HEAD(shmem_swaplist);
|
| 295 | static DEFINE_SPINLOCK(shmem_swaplist_lock);
|
| 296 |
|
| 297 | #ifdef CONFIG_TMPFS_QUOTA
|
| 298 |
|
| 299 | static int shmem_enable_quotas(struct super_block *sb,
|
| 300 | unsigned short quota_types)
|
| 301 | {
|
| 302 | int type, err = 0;
|
| 303 |
|
| 304 | sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
|
| 305 | for (type = 0; type < SHMEM_MAXQUOTAS; type++) {
|
| 306 | if (!(quota_types & (1 << type)))
|
| 307 | continue;
|
| 308 | err = dquot_load_quota_sb(sb, type, QFMT_SHMEM,
|
| 309 | DQUOT_USAGE_ENABLED |
|
| 310 | DQUOT_LIMITS_ENABLED);
|
| 311 | if (err)
|
| 312 | goto out_err;
|
| 313 | }
|
| 314 | return 0;
|
| 315 |
|
| 316 | out_err:
|
| 317 | pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
|
| 318 | type, err);
|
| 319 | for (type--; type >= 0; type--)
|
| 320 | dquot_quota_off(sb, type);
|
| 321 | return err;
|
| 322 | }
|
| 323 |
|
| 324 | static void shmem_disable_quotas(struct super_block *sb)
|
| 325 | {
|
| 326 | int type;
|
| 327 |
|
| 328 | for (type = 0; type < SHMEM_MAXQUOTAS; type++)
|
| 329 | dquot_quota_off(sb, type);
|
| 330 | }
|
| 331 |
|
| 332 | static struct dquot __rcu **shmem_get_dquots(struct inode *inode)
|
| 333 | {
|
| 334 | return SHMEM_I(inode)->i_dquot;
|
| 335 | }
|
| 336 | #endif
|
| 337 |
|
| 338 | |
| 339 | * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
|
| 340 | * produces a novel ino for the newly allocated inode.
|
| 341 | *
|
| 342 | * It may also be called when making a hard link to permit the space needed by
|
| 343 | * each dentry. However, in that case, no new inode number is needed since that
|
| 344 | * internally draws from another pool of inode numbers (currently global
|
| 345 | * get_next_ino()). This case is indicated by passing NULL as inop.
|
| 346 |
|
| 347 | #define SHMEM_INO_BATCH 1024
|
| 348 | static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
|
| 349 | {
|
| 350 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
|
| 351 | ino_t ino;
|
| 352 |
|
| 353 | if (!(sb->s_flags & SB_KERNMOUNT)) {
|
| 354 | raw_spin_lock(&sbinfo->stat_lock);
|
| 355 | if (sbinfo->max_inodes) {
|
| 356 | if (sbinfo->free_ispace < BOGO_INODE_SIZE) {
|
| 357 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 358 | return -ENOSPC;
|
| 359 | }
|
| 360 | sbinfo->free_ispace -= BOGO_INODE_SIZE;
|
| 361 | }
|
| 362 | if (inop) {
|
| 363 | ino = sbinfo->next_ino++;
|
| 364 | if (unlikely(is_zero_ino(ino)))
|
| 365 | ino = sbinfo->next_ino++;
|
| 366 | if (unlikely(!sbinfo->full_inums &&
|
| 367 | ino > UINT_MAX)) {
|
| 368 | |
| 369 | * Emulate get_next_ino uint wraparound for
|
| 370 | * compatibility
|
| 371 |
|
| 372 | if (IS_ENABLED(CONFIG_64BIT))
|
| 373 | pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
|
| 374 | __func__, MINOR(sb->s_dev));
|
| 375 | sbinfo->next_ino = 1;
|
| 376 | ino = sbinfo->next_ino++;
|
| 377 | }
|
| 378 | *inop = ino;
|
| 379 | }
|
| 380 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 381 | } else if (inop) {
|
| 382 | |
| 383 | * __shmem_file_setup, one of our callers, is lock-free: it
|
| 384 | * doesn't hold stat_lock in shmem_reserve_inode since
|
| 385 | * max_inodes is always 0, and is called from potentially
|
| 386 | * unknown contexts. As such, use a per-cpu batched allocator
|
| 387 | * which doesn't require the per-sb stat_lock unless we are at
|
| 388 | * the batch boundary.
|
| 389 | *
|
| 390 | * We don't need to worry about inode{32,64} since SB_KERNMOUNT
|
| 391 | * shmem mounts are not exposed to userspace, so we don't need
|
| 392 | * to worry about things like glibc compatibility.
|
| 393 |
|
| 394 | ino_t *next_ino;
|
| 395 |
|
| 396 | next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
|
| 397 | ino = *next_ino;
|
| 398 | if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
|
| 399 | raw_spin_lock(&sbinfo->stat_lock);
|
| 400 | ino = sbinfo->next_ino;
|
| 401 | sbinfo->next_ino += SHMEM_INO_BATCH;
|
| 402 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 403 | if (unlikely(is_zero_ino(ino)))
|
| 404 | ino++;
|
| 405 | }
|
| 406 | *inop = ino;
|
| 407 | *next_ino = ++ino;
|
| 408 | put_cpu();
|
| 409 | }
|
| 410 |
|
| 411 | return 0;
|
| 412 | }
|
| 413 |
|
| 414 | static void shmem_free_inode(struct super_block *sb, size_t freed_ispace)
|
| 415 | {
|
| 416 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
|
| 417 | if (sbinfo->max_inodes) {
|
| 418 | raw_spin_lock(&sbinfo->stat_lock);
|
| 419 | sbinfo->free_ispace += BOGO_INODE_SIZE + freed_ispace;
|
| 420 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 421 | }
|
| 422 | }
|
| 423 |
|
| 424 | |
| 425 | * shmem_recalc_inode - recalculate the block usage of an inode
|
| 426 | * @inode: inode to recalc
|
| 427 | * @alloced: the change in number of pages allocated to inode
|
| 428 | * @swapped: the change in number of pages swapped from inode
|
| 429 | *
|
| 430 | * We have to calculate the free blocks since the mm can drop
|
| 431 | * undirtied hole pages behind our back.
|
| 432 | *
|
| 433 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
|
| 434 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
|
| 435 | *
|
| 436 | * Return: true if swapped was incremented from 0, for shmem_writeout().
|
| 437 |
|
| 438 | bool shmem_recalc_inode(struct inode *inode, long alloced, long swapped)
|
| 439 | {
|
| 440 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 441 | bool first_swapped = false;
|
| 442 | long freed;
|
| 443 |
|
| 444 | spin_lock(&info->lock);
|
| 445 | info->alloced += alloced;
|
| 446 | info->swapped += swapped;
|
| 447 | freed = info->alloced - info->swapped -
|
| 448 | READ_ONCE(inode->i_mapping->nrpages);
|
| 449 | |
| 450 | * Special case: whereas normally shmem_recalc_inode() is called
|
| 451 | * after i_mapping->nrpages has already been adjusted (up or down),
|
| 452 | * shmem_writeout() has to raise swapped before nrpages is lowered -
|
| 453 | * to stop a racing shmem_recalc_inode() from thinking that a page has
|
| 454 | * been freed. Compensate here, to avoid the need for a followup call.
|
| 455 |
|
| 456 | if (swapped > 0) {
|
| 457 | if (info->swapped == swapped)
|
| 458 | first_swapped = true;
|
| 459 | freed += swapped;
|
| 460 | }
|
| 461 | if (freed > 0)
|
| 462 | info->alloced -= freed;
|
| 463 | spin_unlock(&info->lock);
|
| 464 |
|
| 465 |
|
| 466 | if (freed > 0)
|
| 467 | shmem_inode_unacct_blocks(inode, freed);
|
| 468 | return first_swapped;
|
| 469 | }
|
| 470 |
|
| 471 | bool shmem_charge(struct inode *inode, long pages)
|
| 472 | {
|
| 473 | struct address_space *mapping = inode->i_mapping;
|
| 474 |
|
| 475 | if (shmem_inode_acct_blocks(inode, pages))
|
| 476 | return false;
|
| 477 |
|
| 478 |
|
| 479 | xa_lock_irq(&mapping->i_pages);
|
| 480 | mapping->nrpages += pages;
|
| 481 | xa_unlock_irq(&mapping->i_pages);
|
| 482 |
|
| 483 | shmem_recalc_inode(inode, pages, 0);
|
| 484 | return true;
|
| 485 | }
|
| 486 |
|
| 487 | void shmem_uncharge(struct inode *inode, long pages)
|
| 488 | {
|
| 489 |
|
| 490 |
|
| 491 |
|
| 492 | shmem_recalc_inode(inode, 0, 0);
|
| 493 | }
|
| 494 |
|
| 495 | |
| 496 | * Replace item expected in xarray by a new item, while holding xa_lock.
|
| 497 |
|
| 498 | static int shmem_replace_entry(struct address_space *mapping,
|
| 499 | pgoff_t index, void *expected, void *replacement)
|
| 500 | {
|
| 501 | XA_STATE(xas, &mapping->i_pages, index);
|
| 502 | void *item;
|
| 503 |
|
| 504 | VM_BUG_ON(!expected);
|
| 505 | VM_BUG_ON(!replacement);
|
| 506 | item = xas_load(&xas);
|
| 507 | if (item != expected)
|
| 508 | return -ENOENT;
|
| 509 | xas_store(&xas, replacement);
|
| 510 | return 0;
|
| 511 | }
|
| 512 |
|
| 513 | |
| 514 | * Sometimes, before we decide whether to proceed or to fail, we must check
|
| 515 | * that an entry was not already brought back or split by a racing thread.
|
| 516 | *
|
| 517 | * Checking folio is not enough: by the time a swapcache folio is locked, it
|
| 518 | * might be reused, and again be swapcache, using the same swap as before.
|
| 519 | * Returns the swap entry's order if it still presents, else returns -1.
|
| 520 |
|
| 521 | static int shmem_confirm_swap(struct address_space *mapping, pgoff_t index,
|
| 522 | swp_entry_t swap)
|
| 523 | {
|
| 524 | XA_STATE(xas, &mapping->i_pages, index);
|
| 525 | int ret = -1;
|
| 526 | void *entry;
|
| 527 |
|
| 528 | rcu_read_lock();
|
| 529 | do {
|
| 530 | entry = xas_load(&xas);
|
| 531 | if (entry == swp_to_radix_entry(swap))
|
| 532 | ret = xas_get_order(&xas);
|
| 533 | } while (xas_retry(&xas, entry));
|
| 534 | rcu_read_unlock();
|
| 535 | return ret;
|
| 536 | }
|
| 537 |
|
| 538 | |
| 539 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
|
| 540 | *
|
| 541 | * SHMEM_HUGE_NEVER:
|
| 542 | * disables huge pages for the mount;
|
| 543 | * SHMEM_HUGE_ALWAYS:
|
| 544 | * enables huge pages for the mount;
|
| 545 | * SHMEM_HUGE_WITHIN_SIZE:
|
| 546 | * only allocate huge pages if the page will be fully within i_size,
|
| 547 | * also respect madvise() hints;
|
| 548 | * SHMEM_HUGE_ADVISE:
|
| 549 | * only allocate huge pages if requested with madvise();
|
| 550 |
|
| 551 |
|
| 552 | #define SHMEM_HUGE_NEVER 0
|
| 553 | #define SHMEM_HUGE_ALWAYS 1
|
| 554 | #define SHMEM_HUGE_WITHIN_SIZE 2
|
| 555 | #define SHMEM_HUGE_ADVISE 3
|
| 556 |
|
| 557 | |
| 558 | * Special values.
|
| 559 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
|
| 560 | *
|
| 561 | * SHMEM_HUGE_DENY:
|
| 562 | * disables huge on shm_mnt and all mounts, for emergency use;
|
| 563 | * SHMEM_HUGE_FORCE:
|
| 564 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
|
| 565 | *
|
| 566 |
|
| 567 | #define SHMEM_HUGE_DENY (-1)
|
| 568 | #define SHMEM_HUGE_FORCE (-2)
|
| 569 |
|
| 570 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 571 |
|
| 572 |
|
| 573 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE_SHMEM_HUGE_NEVER)
|
| 574 | #define SHMEM_HUGE_DEFAULT SHMEM_HUGE_NEVER
|
| 575 | #elif defined(CONFIG_TRANSPARENT_HUGEPAGE_SHMEM_HUGE_ALWAYS)
|
| 576 | #define SHMEM_HUGE_DEFAULT SHMEM_HUGE_ALWAYS
|
| 577 | #elif defined(CONFIG_TRANSPARENT_HUGEPAGE_SHMEM_HUGE_WITHIN_SIZE)
|
| 578 | #define SHMEM_HUGE_DEFAULT SHMEM_HUGE_WITHIN_SIZE
|
| 579 | #elif defined(CONFIG_TRANSPARENT_HUGEPAGE_SHMEM_HUGE_ADVISE)
|
| 580 | #define SHMEM_HUGE_DEFAULT SHMEM_HUGE_ADVISE
|
| 581 | #else
|
| 582 | #define SHMEM_HUGE_DEFAULT SHMEM_HUGE_NEVER
|
| 583 | #endif
|
| 584 |
|
| 585 | static int shmem_huge __read_mostly = SHMEM_HUGE_DEFAULT;
|
| 586 |
|
| 587 | #undef SHMEM_HUGE_DEFAULT
|
| 588 |
|
| 589 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE_TMPFS_HUGE_NEVER)
|
| 590 | #define TMPFS_HUGE_DEFAULT SHMEM_HUGE_NEVER
|
| 591 | #elif defined(CONFIG_TRANSPARENT_HUGEPAGE_TMPFS_HUGE_ALWAYS)
|
| 592 | #define TMPFS_HUGE_DEFAULT SHMEM_HUGE_ALWAYS
|
| 593 | #elif defined(CONFIG_TRANSPARENT_HUGEPAGE_TMPFS_HUGE_WITHIN_SIZE)
|
| 594 | #define TMPFS_HUGE_DEFAULT SHMEM_HUGE_WITHIN_SIZE
|
| 595 | #elif defined(CONFIG_TRANSPARENT_HUGEPAGE_TMPFS_HUGE_ADVISE)
|
| 596 | #define TMPFS_HUGE_DEFAULT SHMEM_HUGE_ADVISE
|
| 597 | #else
|
| 598 | #define TMPFS_HUGE_DEFAULT SHMEM_HUGE_NEVER
|
| 599 | #endif
|
| 600 |
|
| 601 | static int tmpfs_huge __read_mostly = TMPFS_HUGE_DEFAULT;
|
| 602 |
|
| 603 | #undef TMPFS_HUGE_DEFAULT
|
| 604 |
|
| 605 | static unsigned int shmem_get_orders_within_size(struct inode *inode,
|
| 606 | unsigned long within_size_orders, pgoff_t index,
|
| 607 | loff_t write_end)
|
| 608 | {
|
| 609 | pgoff_t aligned_index;
|
| 610 | unsigned long order;
|
| 611 | loff_t i_size;
|
| 612 |
|
| 613 | order = highest_order(within_size_orders);
|
| 614 | while (within_size_orders) {
|
| 615 | aligned_index = round_up(index + 1, 1 << order);
|
| 616 | i_size = max(write_end, i_size_read(inode));
|
| 617 | i_size = round_up(i_size, PAGE_SIZE);
|
| 618 | if (i_size >> PAGE_SHIFT >= aligned_index)
|
| 619 | return within_size_orders;
|
| 620 |
|
| 621 | order = next_order(&within_size_orders, order);
|
| 622 | }
|
| 623 |
|
| 624 | return 0;
|
| 625 | }
|
| 626 |
|
| 627 | static unsigned int shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
|
| 628 | loff_t write_end, bool shmem_huge_force,
|
| 629 | struct vm_area_struct *vma,
|
| 630 | vm_flags_t vm_flags)
|
| 631 | {
|
| 632 | unsigned int maybe_pmd_order = HPAGE_PMD_ORDER > MAX_PAGECACHE_ORDER ?
|
| 633 | 0 : BIT(HPAGE_PMD_ORDER);
|
| 634 | unsigned long within_size_orders;
|
| 635 |
|
| 636 | if (!S_ISREG(inode->i_mode))
|
| 637 | return 0;
|
| 638 | if (shmem_huge == SHMEM_HUGE_DENY)
|
| 639 | return 0;
|
| 640 | if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
|
| 641 | return maybe_pmd_order;
|
| 642 |
|
| 643 | |
| 644 | * The huge order allocation for anon shmem is controlled through
|
| 645 | * the mTHP interface, so we still use PMD-sized huge order to
|
| 646 | * check whether global control is enabled.
|
| 647 | *
|
| 648 | * For tmpfs with 'huge=always' or 'huge=within_size' mount option,
|
| 649 | * we will always try PMD-sized order first. If that failed, it will
|
| 650 | * fall back to small large folios.
|
| 651 |
|
| 652 | switch (SHMEM_SB(inode->i_sb)->huge) {
|
| 653 | case SHMEM_HUGE_ALWAYS:
|
| 654 | return THP_ORDERS_ALL_FILE_DEFAULT;
|
| 655 | case SHMEM_HUGE_WITHIN_SIZE:
|
| 656 | within_size_orders = shmem_get_orders_within_size(inode,
|
| 657 | THP_ORDERS_ALL_FILE_DEFAULT, index, write_end);
|
| 658 | if (within_size_orders > 0)
|
| 659 | return within_size_orders;
|
| 660 |
|
| 661 | fallthrough;
|
| 662 | case SHMEM_HUGE_ADVISE:
|
| 663 | if (vm_flags & VM_HUGEPAGE)
|
| 664 | return THP_ORDERS_ALL_FILE_DEFAULT;
|
| 665 | fallthrough;
|
| 666 | default:
|
| 667 | return 0;
|
| 668 | }
|
| 669 | }
|
| 670 |
|
| 671 | static int shmem_parse_huge(const char *str)
|
| 672 | {
|
| 673 | int huge;
|
| 674 |
|
| 675 | if (!str)
|
| 676 | return -EINVAL;
|
| 677 |
|
| 678 | if (!strcmp(str, "never"))
|
| 679 | huge = SHMEM_HUGE_NEVER;
|
| 680 | else if (!strcmp(str, "always"))
|
| 681 | huge = SHMEM_HUGE_ALWAYS;
|
| 682 | else if (!strcmp(str, "within_size"))
|
| 683 | huge = SHMEM_HUGE_WITHIN_SIZE;
|
| 684 | else if (!strcmp(str, "advise"))
|
| 685 | huge = SHMEM_HUGE_ADVISE;
|
| 686 | else if (!strcmp(str, "deny"))
|
| 687 | huge = SHMEM_HUGE_DENY;
|
| 688 | else if (!strcmp(str, "force"))
|
| 689 | huge = SHMEM_HUGE_FORCE;
|
| 690 | else
|
| 691 | return -EINVAL;
|
| 692 |
|
| 693 | if (!has_transparent_hugepage() &&
|
| 694 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
|
| 695 | return -EINVAL;
|
| 696 |
|
| 697 |
|
| 698 | if (huge == SHMEM_HUGE_FORCE &&
|
| 699 | huge_shmem_orders_inherit != BIT(HPAGE_PMD_ORDER))
|
| 700 | return -EINVAL;
|
| 701 |
|
| 702 | return huge;
|
| 703 | }
|
| 704 |
|
| 705 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
|
| 706 | static const char *shmem_format_huge(int huge)
|
| 707 | {
|
| 708 | switch (huge) {
|
| 709 | case SHMEM_HUGE_NEVER:
|
| 710 | return "never";
|
| 711 | case SHMEM_HUGE_ALWAYS:
|
| 712 | return "always";
|
| 713 | case SHMEM_HUGE_WITHIN_SIZE:
|
| 714 | return "within_size";
|
| 715 | case SHMEM_HUGE_ADVISE:
|
| 716 | return "advise";
|
| 717 | case SHMEM_HUGE_DENY:
|
| 718 | return "deny";
|
| 719 | case SHMEM_HUGE_FORCE:
|
| 720 | return "force";
|
| 721 | default:
|
| 722 | VM_BUG_ON(1);
|
| 723 | return "bad_val";
|
| 724 | }
|
| 725 | }
|
| 726 | #endif
|
| 727 |
|
| 728 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
|
| 729 | struct shrink_control *sc, unsigned long nr_to_free)
|
| 730 | {
|
| 731 | LIST_HEAD(list), *pos, *next;
|
| 732 | struct inode *inode;
|
| 733 | struct shmem_inode_info *info;
|
| 734 | struct folio *folio;
|
| 735 | unsigned long batch = sc ? sc->nr_to_scan : 128;
|
| 736 | unsigned long split = 0, freed = 0;
|
| 737 |
|
| 738 | if (list_empty(&sbinfo->shrinklist))
|
| 739 | return SHRINK_STOP;
|
| 740 |
|
| 741 | spin_lock(&sbinfo->shrinklist_lock);
|
| 742 | list_for_each_safe(pos, next, &sbinfo->shrinklist) {
|
| 743 | info = list_entry(pos, struct shmem_inode_info, shrinklist);
|
| 744 |
|
| 745 |
|
| 746 | inode = igrab(&info->vfs_inode);
|
| 747 |
|
| 748 |
|
| 749 | if (!inode) {
|
| 750 | list_del_init(&info->shrinklist);
|
| 751 | goto next;
|
| 752 | }
|
| 753 |
|
| 754 | list_move(&info->shrinklist, &list);
|
| 755 | next:
|
| 756 | sbinfo->shrinklist_len--;
|
| 757 | if (!--batch)
|
| 758 | break;
|
| 759 | }
|
| 760 | spin_unlock(&sbinfo->shrinklist_lock);
|
| 761 |
|
| 762 | list_for_each_safe(pos, next, &list) {
|
| 763 | pgoff_t next, end;
|
| 764 | loff_t i_size;
|
| 765 | int ret;
|
| 766 |
|
| 767 | info = list_entry(pos, struct shmem_inode_info, shrinklist);
|
| 768 | inode = &info->vfs_inode;
|
| 769 |
|
| 770 | if (nr_to_free && freed >= nr_to_free)
|
| 771 | goto move_back;
|
| 772 |
|
| 773 | i_size = i_size_read(inode);
|
| 774 | folio = filemap_get_entry(inode->i_mapping, i_size / PAGE_SIZE);
|
| 775 | if (!folio || xa_is_value(folio))
|
| 776 | goto drop;
|
| 777 |
|
| 778 |
|
| 779 | if (!folio_test_large(folio)) {
|
| 780 | folio_put(folio);
|
| 781 | goto drop;
|
| 782 | }
|
| 783 |
|
| 784 |
|
| 785 | next = folio_next_index(folio);
|
| 786 | end = shmem_fallocend(inode, DIV_ROUND_UP(i_size, PAGE_SIZE));
|
| 787 | if (end <= folio->index || end >= next) {
|
| 788 | folio_put(folio);
|
| 789 | goto drop;
|
| 790 | }
|
| 791 |
|
| 792 | |
| 793 | * Move the inode on the list back to shrinklist if we failed
|
| 794 | * to lock the page at this time.
|
| 795 | *
|
| 796 | * Waiting for the lock may lead to deadlock in the
|
| 797 | * reclaim path.
|
| 798 |
|
| 799 | if (!folio_trylock(folio)) {
|
| 800 | folio_put(folio);
|
| 801 | goto move_back;
|
| 802 | }
|
| 803 |
|
| 804 | ret = split_folio(folio);
|
| 805 | folio_unlock(folio);
|
| 806 | folio_put(folio);
|
| 807 |
|
| 808 |
|
| 809 | if (ret)
|
| 810 | goto move_back;
|
| 811 |
|
| 812 | freed += next - end;
|
| 813 | split++;
|
| 814 | drop:
|
| 815 | list_del_init(&info->shrinklist);
|
| 816 | goto put;
|
| 817 | move_back:
|
| 818 | |
| 819 | * Make sure the inode is either on the global list or deleted
|
| 820 | * from any local list before iput() since it could be deleted
|
| 821 | * in another thread once we put the inode (then the local list
|
| 822 | * is corrupted).
|
| 823 |
|
| 824 | spin_lock(&sbinfo->shrinklist_lock);
|
| 825 | list_move(&info->shrinklist, &sbinfo->shrinklist);
|
| 826 | sbinfo->shrinklist_len++;
|
| 827 | spin_unlock(&sbinfo->shrinklist_lock);
|
| 828 | put:
|
| 829 | iput(inode);
|
| 830 | }
|
| 831 |
|
| 832 | return split;
|
| 833 | }
|
| 834 |
|
| 835 | static long shmem_unused_huge_scan(struct super_block *sb,
|
| 836 | struct shrink_control *sc)
|
| 837 | {
|
| 838 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
|
| 839 |
|
| 840 | if (!READ_ONCE(sbinfo->shrinklist_len))
|
| 841 | return SHRINK_STOP;
|
| 842 |
|
| 843 | return shmem_unused_huge_shrink(sbinfo, sc, 0);
|
| 844 | }
|
| 845 |
|
| 846 | static long shmem_unused_huge_count(struct super_block *sb,
|
| 847 | struct shrink_control *sc)
|
| 848 | {
|
| 849 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
|
| 850 |
|
| 851 | |
| 852 | * The per-superblock shrinklist is filesystem-global and does not
|
| 853 | * honour sc->memcg, so it is only meaningful on the global (kswapd or
|
| 854 | * root direct reclaim) shrink path. Skip the per-memcg iterations of
|
| 855 | * shrink_slab_memcg() to avoid queueing duplicate global work.
|
| 856 |
|
| 857 | if (!mem_cgroup_shrink_is_root(sc))
|
| 858 | return 0;
|
| 859 |
|
| 860 | return READ_ONCE(sbinfo->shrinklist_len);
|
| 861 | }
|
| 862 | #else
|
| 863 |
|
| 864 | #define shmem_huge SHMEM_HUGE_DENY
|
| 865 |
|
| 866 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
|
| 867 | struct shrink_control *sc, unsigned long nr_to_free)
|
| 868 | {
|
| 869 | return 0;
|
| 870 | }
|
| 871 |
|
| 872 | static unsigned int shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
|
| 873 | loff_t write_end, bool shmem_huge_force,
|
| 874 | struct vm_area_struct *vma,
|
| 875 | vm_flags_t vm_flags)
|
| 876 | {
|
| 877 | return 0;
|
| 878 | }
|
| 879 | #endif
|
| 880 |
|
| 881 | static void shmem_update_stats(struct folio *folio, int nr_pages)
|
| 882 | {
|
| 883 | if (folio_test_pmd_mappable(folio))
|
| 884 | lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr_pages);
|
| 885 | lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr_pages);
|
| 886 | lruvec_stat_mod_folio(folio, NR_SHMEM, nr_pages);
|
| 887 | }
|
| 888 |
|
| 889 | |
| 890 | * Somewhat like filemap_add_folio, but error if expected item has gone.
|
| 891 |
|
| 892 | int shmem_add_to_page_cache(struct folio *folio,
|
| 893 | struct address_space *mapping,
|
| 894 | pgoff_t index, void *expected, gfp_t gfp)
|
| 895 | {
|
| 896 | XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
|
| 897 | unsigned long nr = folio_nr_pages(folio);
|
| 898 | swp_entry_t iter, swap;
|
| 899 | void *entry;
|
| 900 |
|
| 901 | VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
|
| 902 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
|
| 903 | VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
|
| 904 |
|
| 905 | folio_ref_add(folio, nr);
|
| 906 | folio->mapping = mapping;
|
| 907 | folio->index = index;
|
| 908 |
|
| 909 | gfp &= GFP_RECLAIM_MASK;
|
| 910 | folio_throttle_swaprate(folio, gfp);
|
| 911 | swap = radix_to_swp_entry(expected);
|
| 912 |
|
| 913 | do {
|
| 914 | iter = swap;
|
| 915 | xas_lock_irq(&xas);
|
| 916 | xas_for_each_conflict(&xas, entry) {
|
| 917 | |
| 918 | * The range must either be empty, or filled with
|
| 919 | * expected swap entries. Shmem swap entries are never
|
| 920 | * partially freed without split of both entry and
|
| 921 | * folio, so there shouldn't be any holes.
|
| 922 |
|
| 923 | if (!expected || entry != swp_to_radix_entry(iter)) {
|
| 924 | xas_set_err(&xas, -EEXIST);
|
| 925 | goto unlock;
|
| 926 | }
|
| 927 | iter.val += 1 << xas_get_order(&xas);
|
| 928 | }
|
| 929 | if (expected && iter.val - nr != swap.val) {
|
| 930 | xas_set_err(&xas, -EEXIST);
|
| 931 | goto unlock;
|
| 932 | }
|
| 933 | xas_store(&xas, folio);
|
| 934 | if (xas_error(&xas))
|
| 935 | goto unlock;
|
| 936 | shmem_update_stats(folio, nr);
|
| 937 | mapping->nrpages += nr;
|
| 938 | unlock:
|
| 939 | xas_unlock_irq(&xas);
|
| 940 | } while (xas_nomem(&xas, gfp));
|
| 941 |
|
| 942 | if (xas_error(&xas)) {
|
| 943 | folio->mapping = NULL;
|
| 944 | folio_ref_sub(folio, nr);
|
| 945 | return xas_error(&xas);
|
| 946 | }
|
| 947 |
|
| 948 | return 0;
|
| 949 | }
|
| 950 |
|
| 951 | |
| 952 | * Somewhat like filemap_remove_folio, but substitutes swap for @folio.
|
| 953 |
|
| 954 | static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
|
| 955 | {
|
| 956 | struct address_space *mapping = folio->mapping;
|
| 957 | long nr = folio_nr_pages(folio);
|
| 958 | int error;
|
| 959 |
|
| 960 | xa_lock_irq(&mapping->i_pages);
|
| 961 | error = shmem_replace_entry(mapping, folio->index, folio, radswap);
|
| 962 | folio->mapping = NULL;
|
| 963 | mapping->nrpages -= nr;
|
| 964 | shmem_update_stats(folio, -nr);
|
| 965 | xa_unlock_irq(&mapping->i_pages);
|
| 966 | folio_put_refs(folio, nr);
|
| 967 | BUG_ON(error);
|
| 968 | }
|
| 969 |
|
| 970 | |
| 971 | * Remove swap entry from page cache, free the swap and its page cache. Returns
|
| 972 | * the number of pages being freed. 0 means entry not found in XArray (0 pages
|
| 973 | * being freed).
|
| 974 |
|
| 975 | static long shmem_free_swap(struct address_space *mapping,
|
| 976 | pgoff_t index, pgoff_t end, void *radswap)
|
| 977 | {
|
| 978 | XA_STATE(xas, &mapping->i_pages, index);
|
| 979 | unsigned int nr_pages = 0;
|
| 980 | pgoff_t base;
|
| 981 | void *entry;
|
| 982 |
|
| 983 | xas_lock_irq(&xas);
|
| 984 | entry = xas_load(&xas);
|
| 985 | if (entry == radswap) {
|
| 986 | nr_pages = 1 << xas_get_order(&xas);
|
| 987 | base = round_down(xas.xa_index, nr_pages);
|
| 988 | if (base < index || base + nr_pages - 1 > end)
|
| 989 | nr_pages = 0;
|
| 990 | else
|
| 991 | xas_store(&xas, NULL);
|
| 992 | }
|
| 993 | xas_unlock_irq(&xas);
|
| 994 |
|
| 995 | if (nr_pages)
|
| 996 | swap_put_entries_direct(radix_to_swp_entry(radswap), nr_pages);
|
| 997 |
|
| 998 | return nr_pages;
|
| 999 | }
|
| 1000 |
|
| 1001 | |
| 1002 | * Determine (in bytes) how many of the shmem object's pages mapped by the
|
| 1003 | * given offsets are swapped out.
|
| 1004 | *
|
| 1005 | * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
|
| 1006 | * as long as the inode doesn't go away and racy results are not a problem.
|
| 1007 |
|
| 1008 | unsigned long shmem_partial_swap_usage(struct address_space *mapping,
|
| 1009 | pgoff_t start, pgoff_t end)
|
| 1010 | {
|
| 1011 | XA_STATE(xas, &mapping->i_pages, start);
|
| 1012 | struct folio *folio;
|
| 1013 | unsigned long swapped = 0;
|
| 1014 | unsigned long max = end - 1;
|
| 1015 |
|
| 1016 | rcu_read_lock();
|
| 1017 | xas_for_each(&xas, folio, max) {
|
| 1018 | if (xas_retry(&xas, folio))
|
| 1019 | continue;
|
| 1020 | if (xa_is_value(folio))
|
| 1021 | swapped += 1 << xas_get_order(&xas);
|
| 1022 | if (xas.xa_index == max)
|
| 1023 | break;
|
| 1024 | if (need_resched()) {
|
| 1025 | xas_pause(&xas);
|
| 1026 | cond_resched_rcu();
|
| 1027 | }
|
| 1028 | }
|
| 1029 | rcu_read_unlock();
|
| 1030 |
|
| 1031 | return swapped << PAGE_SHIFT;
|
| 1032 | }
|
| 1033 |
|
| 1034 | |
| 1035 | * Determine (in bytes) how many of the shmem object's pages mapped by the
|
| 1036 | * given vma is swapped out.
|
| 1037 | *
|
| 1038 | * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
|
| 1039 | * as long as the inode doesn't go away and racy results are not a problem.
|
| 1040 |
|
| 1041 | unsigned long shmem_swap_usage(struct vm_area_struct *vma)
|
| 1042 | {
|
| 1043 | struct inode *inode = file_inode(vma->vm_file);
|
| 1044 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 1045 | struct address_space *mapping = inode->i_mapping;
|
| 1046 | const pgoff_t pgoff = vma_start_pgoff(vma);
|
| 1047 | const pgoff_t pgoff_end = vma_end_pgoff(vma);
|
| 1048 | unsigned long swapped;
|
| 1049 |
|
| 1050 |
|
| 1051 | swapped = READ_ONCE(info->swapped);
|
| 1052 |
|
| 1053 | |
| 1054 | * The easier cases are when the shmem object has nothing in swap, or
|
| 1055 | * the vma maps it whole. Then we can simply use the stats that we
|
| 1056 | * already track.
|
| 1057 |
|
| 1058 | if (!swapped)
|
| 1059 | return 0;
|
| 1060 |
|
| 1061 | if (!pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
|
| 1062 | return swapped << PAGE_SHIFT;
|
| 1063 |
|
| 1064 |
|
| 1065 | return shmem_partial_swap_usage(mapping, pgoff, pgoff_end);
|
| 1066 | }
|
| 1067 |
|
| 1068 | |
| 1069 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
|
| 1070 |
|
| 1071 | void shmem_unlock_mapping(struct address_space *mapping)
|
| 1072 | {
|
| 1073 | struct folio_batch fbatch;
|
| 1074 | pgoff_t index = 0;
|
| 1075 |
|
| 1076 | folio_batch_init(&fbatch);
|
| 1077 | |
| 1078 | * Minor point, but we might as well stop if someone else SHM_LOCKs it.
|
| 1079 |
|
| 1080 | while (!mapping_unevictable(mapping) &&
|
| 1081 | filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
|
| 1082 | check_move_unevictable_folios(&fbatch);
|
| 1083 | folio_batch_release(&fbatch);
|
| 1084 | cond_resched();
|
| 1085 | }
|
| 1086 | }
|
| 1087 |
|
| 1088 | static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
|
| 1089 | {
|
| 1090 | struct folio *folio;
|
| 1091 |
|
| 1092 | |
| 1093 | * At first avoid shmem_get_folio(,,,SGP_READ): that fails
|
| 1094 | * beyond i_size, and reports fallocated folios as holes.
|
| 1095 |
|
| 1096 | folio = filemap_get_entry(inode->i_mapping, index);
|
| 1097 | if (!folio)
|
| 1098 | return folio;
|
| 1099 | if (!xa_is_value(folio)) {
|
| 1100 | folio_lock(folio);
|
| 1101 | if (folio->mapping == inode->i_mapping)
|
| 1102 | return folio;
|
| 1103 |
|
| 1104 | folio_unlock(folio);
|
| 1105 | folio_put(folio);
|
| 1106 | }
|
| 1107 | |
| 1108 | * But read a folio back from swap if any of it is within i_size
|
| 1109 | * (although in some cases this is just a waste of time).
|
| 1110 |
|
| 1111 | folio = NULL;
|
| 1112 | shmem_get_folio(inode, index, 0, &folio, SGP_READ);
|
| 1113 | return folio;
|
| 1114 | }
|
| 1115 |
|
| 1116 | |
| 1117 | * Remove range of pages and swap entries from page cache, and free them.
|
| 1118 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
|
| 1119 |
|
| 1120 | static void shmem_undo_range(struct inode *inode, loff_t lstart, uoff_t lend,
|
| 1121 | bool unfalloc)
|
| 1122 | {
|
| 1123 | struct address_space *mapping = inode->i_mapping;
|
| 1124 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 1125 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
|
| 1126 | pgoff_t end = (lend + 1) >> PAGE_SHIFT;
|
| 1127 | struct folio_batch fbatch;
|
| 1128 | pgoff_t indices[FOLIO_BATCH_SIZE];
|
| 1129 | struct folio *folio;
|
| 1130 | bool same_folio;
|
| 1131 | long nr_swaps_freed = 0;
|
| 1132 | pgoff_t index;
|
| 1133 | int i;
|
| 1134 |
|
| 1135 | if (lend == -1)
|
| 1136 | end = -1;
|
| 1137 |
|
| 1138 | if (info->fallocend > start && info->fallocend <= end && !unfalloc)
|
| 1139 | info->fallocend = start;
|
| 1140 |
|
| 1141 | folio_batch_init(&fbatch);
|
| 1142 | index = start;
|
| 1143 | while (index < end && find_lock_entries(mapping, &index, end - 1,
|
| 1144 | &fbatch, indices)) {
|
| 1145 | for (i = 0; i < folio_batch_count(&fbatch); i++) {
|
| 1146 | folio = fbatch.folios[i];
|
| 1147 |
|
| 1148 | if (xa_is_value(folio)) {
|
| 1149 | if (unfalloc)
|
| 1150 | continue;
|
| 1151 | nr_swaps_freed += shmem_free_swap(mapping, indices[i],
|
| 1152 | end - 1, folio);
|
| 1153 | continue;
|
| 1154 | }
|
| 1155 |
|
| 1156 | if (!unfalloc || !folio_test_uptodate(folio))
|
| 1157 | truncate_inode_folio(mapping, folio);
|
| 1158 | folio_unlock(folio);
|
| 1159 | }
|
| 1160 | folio_batch_remove_exceptionals(&fbatch);
|
| 1161 | folio_batch_release(&fbatch);
|
| 1162 | cond_resched();
|
| 1163 | }
|
| 1164 |
|
| 1165 | |
| 1166 | * When undoing a failed fallocate, we want none of the partial folio
|
| 1167 | * zeroing and splitting below, but shall want to truncate the whole
|
| 1168 | * folio when !uptodate indicates that it was added by this fallocate,
|
| 1169 | * even when [lstart, lend] covers only a part of the folio.
|
| 1170 |
|
| 1171 | if (unfalloc)
|
| 1172 | goto whole_folios;
|
| 1173 |
|
| 1174 | same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
|
| 1175 | folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
|
| 1176 | if (folio) {
|
| 1177 | same_folio = lend < folio_next_pos(folio);
|
| 1178 | folio_mark_dirty(folio);
|
| 1179 | if (!truncate_inode_partial_folio(folio, lstart, lend)) {
|
| 1180 | start = folio_next_index(folio);
|
| 1181 | if (same_folio)
|
| 1182 | end = folio->index;
|
| 1183 | }
|
| 1184 | folio_unlock(folio);
|
| 1185 | folio_put(folio);
|
| 1186 | folio = NULL;
|
| 1187 | }
|
| 1188 |
|
| 1189 | if (!same_folio)
|
| 1190 | folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
|
| 1191 | if (folio) {
|
| 1192 | folio_mark_dirty(folio);
|
| 1193 | if (!truncate_inode_partial_folio(folio, lstart, lend))
|
| 1194 | end = folio->index;
|
| 1195 | folio_unlock(folio);
|
| 1196 | folio_put(folio);
|
| 1197 | }
|
| 1198 |
|
| 1199 | whole_folios:
|
| 1200 |
|
| 1201 | index = start;
|
| 1202 | while (index < end) {
|
| 1203 | cond_resched();
|
| 1204 |
|
| 1205 | if (!find_get_entries(mapping, &index, end - 1, &fbatch,
|
| 1206 | indices)) {
|
| 1207 |
|
| 1208 | if (index == start || end != -1)
|
| 1209 | break;
|
| 1210 |
|
| 1211 | index = start;
|
| 1212 | continue;
|
| 1213 | }
|
| 1214 | for (i = 0; i < folio_batch_count(&fbatch); i++) {
|
| 1215 | folio = fbatch.folios[i];
|
| 1216 |
|
| 1217 | if (xa_is_value(folio)) {
|
| 1218 | int order;
|
| 1219 | long swaps_freed;
|
| 1220 |
|
| 1221 | if (unfalloc)
|
| 1222 | continue;
|
| 1223 | swaps_freed = shmem_free_swap(mapping, indices[i],
|
| 1224 | end - 1, folio);
|
| 1225 | if (!swaps_freed) {
|
| 1226 | pgoff_t base = indices[i];
|
| 1227 |
|
| 1228 | order = shmem_confirm_swap(mapping, indices[i],
|
| 1229 | radix_to_swp_entry(folio));
|
| 1230 | |
| 1231 | * If found a large swap entry cross the end or start
|
| 1232 | * border, skip it as the truncate_inode_partial_folio
|
| 1233 | * above should have at least zerod its content once.
|
| 1234 |
|
| 1235 | if (order > 0) {
|
| 1236 | base = round_down(base, 1 << order);
|
| 1237 | if (base < start || base + (1 << order) > end)
|
| 1238 | continue;
|
| 1239 | }
|
| 1240 |
|
| 1241 | index = base;
|
| 1242 | break;
|
| 1243 | }
|
| 1244 | nr_swaps_freed += swaps_freed;
|
| 1245 | continue;
|
| 1246 | }
|
| 1247 |
|
| 1248 | folio_lock(folio);
|
| 1249 |
|
| 1250 | if (!unfalloc || !folio_test_uptodate(folio)) {
|
| 1251 | if (folio_mapping(folio) != mapping) {
|
| 1252 |
|
| 1253 | folio_unlock(folio);
|
| 1254 | index = indices[i];
|
| 1255 | break;
|
| 1256 | }
|
| 1257 | VM_BUG_ON_FOLIO(folio_test_writeback(folio),
|
| 1258 | folio);
|
| 1259 |
|
| 1260 | if (!folio_test_large(folio)) {
|
| 1261 | truncate_inode_folio(mapping, folio);
|
| 1262 | } else if (truncate_inode_partial_folio(folio, lstart, lend)) {
|
| 1263 | |
| 1264 | * If we split a page, reset the loop so
|
| 1265 | * that we pick up the new sub pages.
|
| 1266 | * Otherwise the THP was entirely
|
| 1267 | * dropped or the target range was
|
| 1268 | * zeroed, so just continue the loop as
|
| 1269 | * is.
|
| 1270 |
|
| 1271 | if (!folio_test_large(folio)) {
|
| 1272 | folio_unlock(folio);
|
| 1273 | index = start;
|
| 1274 | break;
|
| 1275 | }
|
| 1276 | }
|
| 1277 | }
|
| 1278 | folio_unlock(folio);
|
| 1279 | }
|
| 1280 | folio_batch_remove_exceptionals(&fbatch);
|
| 1281 | folio_batch_release(&fbatch);
|
| 1282 | }
|
| 1283 |
|
| 1284 | shmem_recalc_inode(inode, 0, -nr_swaps_freed);
|
| 1285 | }
|
| 1286 |
|
| 1287 | void shmem_truncate_range(struct inode *inode, loff_t lstart, uoff_t lend)
|
| 1288 | {
|
| 1289 | shmem_undo_range(inode, lstart, lend, false);
|
| 1290 | inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
|
| 1291 | inode_inc_iversion(inode);
|
| 1292 | }
|
| 1293 | EXPORT_SYMBOL_GPL(shmem_truncate_range);
|
| 1294 |
|
| 1295 | static int shmem_getattr(const struct mnt_idmap *idmap,
|
| 1296 | const struct path *path, struct kstat *stat,
|
| 1297 | u32 request_mask, unsigned int query_flags)
|
| 1298 | {
|
| 1299 | struct inode *inode = path->dentry->d_inode;
|
| 1300 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 1301 |
|
| 1302 |
|
| 1303 | if (data_race(info->alloced - info->swapped != inode->i_mapping->nrpages))
|
| 1304 | shmem_recalc_inode(inode, 0, 0);
|
| 1305 |
|
| 1306 | if (info->fsflags & FS_APPEND_FL)
|
| 1307 | stat->attributes |= STATX_ATTR_APPEND;
|
| 1308 | if (info->fsflags & FS_IMMUTABLE_FL)
|
| 1309 | stat->attributes |= STATX_ATTR_IMMUTABLE;
|
| 1310 | if (info->fsflags & FS_NODUMP_FL)
|
| 1311 | stat->attributes |= STATX_ATTR_NODUMP;
|
| 1312 | stat->attributes_mask |= (STATX_ATTR_APPEND |
|
| 1313 | STATX_ATTR_IMMUTABLE |
|
| 1314 | STATX_ATTR_NODUMP);
|
| 1315 | generic_fillattr(idmap, request_mask, inode, stat);
|
| 1316 |
|
| 1317 | if (shmem_huge_global_enabled(inode, 0, 0, false, NULL, 0))
|
| 1318 | stat->blksize = HPAGE_PMD_SIZE;
|
| 1319 |
|
| 1320 | if (request_mask & STATX_BTIME) {
|
| 1321 | stat->result_mask |= STATX_BTIME;
|
| 1322 | stat->btime.tv_sec = info->i_crtime.tv_sec;
|
| 1323 | stat->btime.tv_nsec = info->i_crtime.tv_nsec;
|
| 1324 | }
|
| 1325 |
|
| 1326 | return 0;
|
| 1327 | }
|
| 1328 |
|
| 1329 | static int shmem_setattr(const struct mnt_idmap *idmap,
|
| 1330 | struct dentry *dentry, struct iattr *attr)
|
| 1331 | {
|
| 1332 | struct inode *inode = d_inode(dentry);
|
| 1333 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 1334 | int error;
|
| 1335 | bool update_mtime = false;
|
| 1336 | bool update_ctime = true;
|
| 1337 |
|
| 1338 | error = setattr_prepare(idmap, dentry, attr);
|
| 1339 | if (error)
|
| 1340 | return error;
|
| 1341 |
|
| 1342 | if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
|
| 1343 | if ((inode->i_mode ^ attr->ia_mode) & 0111) {
|
| 1344 | return -EPERM;
|
| 1345 | }
|
| 1346 | }
|
| 1347 |
|
| 1348 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
|
| 1349 | loff_t oldsize = inode->i_size;
|
| 1350 | loff_t newsize = attr->ia_size;
|
| 1351 |
|
| 1352 |
|
| 1353 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
|
| 1354 | (newsize > oldsize && (info->seals & F_SEAL_GROW)))
|
| 1355 | return -EPERM;
|
| 1356 |
|
| 1357 | if (newsize != oldsize) {
|
| 1358 | if (info->flags & SHMEM_F_MAPPING_FROZEN)
|
| 1359 | return -EPERM;
|
| 1360 | error = shmem_reacct_size(SHMEM_I(inode)->flags,
|
| 1361 | oldsize, newsize);
|
| 1362 | if (error)
|
| 1363 | return error;
|
| 1364 | i_size_write(inode, newsize);
|
| 1365 | update_mtime = true;
|
| 1366 | } else {
|
| 1367 | update_ctime = false;
|
| 1368 | }
|
| 1369 | if (newsize <= oldsize) {
|
| 1370 | loff_t holebegin = round_up(newsize, PAGE_SIZE);
|
| 1371 | if (oldsize > holebegin)
|
| 1372 | unmap_mapping_range(inode->i_mapping,
|
| 1373 | holebegin, 0, 1);
|
| 1374 | if (info->alloced)
|
| 1375 | shmem_truncate_range(inode,
|
| 1376 | newsize, (loff_t)-1);
|
| 1377 |
|
| 1378 | if (oldsize > holebegin)
|
| 1379 | unmap_mapping_range(inode->i_mapping,
|
| 1380 | holebegin, 0, 1);
|
| 1381 | }
|
| 1382 | }
|
| 1383 |
|
| 1384 | if (is_quota_modification(idmap, inode, attr)) {
|
| 1385 | error = dquot_initialize(inode);
|
| 1386 | if (error)
|
| 1387 | return error;
|
| 1388 | }
|
| 1389 |
|
| 1390 |
|
| 1391 | if (i_uid_needs_update(idmap, attr, inode) ||
|
| 1392 | i_gid_needs_update(idmap, attr, inode)) {
|
| 1393 | error = dquot_transfer(idmap, inode, attr);
|
| 1394 | if (error)
|
| 1395 | return error;
|
| 1396 | }
|
| 1397 |
|
| 1398 | setattr_copy(idmap, inode, attr);
|
| 1399 | if (attr->ia_valid & ATTR_MODE)
|
| 1400 | error = posix_acl_chmod(idmap, dentry, inode->i_mode);
|
| 1401 | if (!error && update_ctime) {
|
| 1402 | inode_set_ctime_current(inode);
|
| 1403 | if (update_mtime)
|
| 1404 | inode_set_mtime_to_ts(inode, inode_get_ctime(inode));
|
| 1405 | inode_inc_iversion(inode);
|
| 1406 | }
|
| 1407 | return error;
|
| 1408 | }
|
| 1409 |
|
| 1410 | static void shmem_evict_inode(struct inode *inode)
|
| 1411 | {
|
| 1412 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 1413 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 1414 | size_t freed = 0;
|
| 1415 |
|
| 1416 | if (shmem_mapping(inode->i_mapping)) {
|
| 1417 | shmem_unacct_size(info->flags, inode->i_size);
|
| 1418 | inode->i_size = 0;
|
| 1419 | mapping_set_exiting(inode->i_mapping);
|
| 1420 | shmem_truncate_range(inode, 0, (loff_t)-1);
|
| 1421 | if (!list_empty(&info->shrinklist)) {
|
| 1422 | spin_lock(&sbinfo->shrinklist_lock);
|
| 1423 | if (!list_empty(&info->shrinklist)) {
|
| 1424 | list_del_init(&info->shrinklist);
|
| 1425 | sbinfo->shrinklist_len--;
|
| 1426 | }
|
| 1427 | spin_unlock(&sbinfo->shrinklist_lock);
|
| 1428 | }
|
| 1429 | while (!list_empty(&info->swaplist)) {
|
| 1430 |
|
| 1431 | wait_var_event(&info->stop_eviction,
|
| 1432 | !atomic_read(&info->stop_eviction));
|
| 1433 | spin_lock(&shmem_swaplist_lock);
|
| 1434 |
|
| 1435 | if (!atomic_read(&info->stop_eviction))
|
| 1436 | list_del_init(&info->swaplist);
|
| 1437 | spin_unlock(&shmem_swaplist_lock);
|
| 1438 | }
|
| 1439 | }
|
| 1440 |
|
| 1441 | simple_xattrs_free(&sbinfo->xa_cache, &info->xattrs, sbinfo->max_inodes ? &freed : NULL);
|
| 1442 |
|
| 1443 | shmem_free_inode(inode->i_sb, freed);
|
| 1444 | if (inode->i_blocks)
|
| 1445 | pr_warn("%s: ino=%llu i_blocks=%llu alloced=%lu swapped=%lu nrpages=%lu\n",
|
| 1446 | __func__, inode->i_ino, inode->i_blocks,
|
| 1447 | info->alloced, info->swapped, inode->i_mapping->nrpages);
|
| 1448 | clear_inode(inode);
|
| 1449 | #ifdef CONFIG_TMPFS_QUOTA
|
| 1450 | dquot_free_inode(inode);
|
| 1451 | dquot_drop(inode);
|
| 1452 | #endif
|
| 1453 | }
|
| 1454 |
|
| 1455 | static unsigned int shmem_find_swap_entries(struct address_space *mapping,
|
| 1456 | pgoff_t start, struct folio_batch *fbatch,
|
| 1457 | pgoff_t *indices, unsigned int type)
|
| 1458 | {
|
| 1459 | XA_STATE(xas, &mapping->i_pages, start);
|
| 1460 | struct folio *folio;
|
| 1461 | swp_entry_t entry;
|
| 1462 |
|
| 1463 | rcu_read_lock();
|
| 1464 | xas_for_each(&xas, folio, ULONG_MAX) {
|
| 1465 | if (xas_retry(&xas, folio))
|
| 1466 | continue;
|
| 1467 |
|
| 1468 | if (!xa_is_value(folio))
|
| 1469 | continue;
|
| 1470 |
|
| 1471 | entry = radix_to_swp_entry(folio);
|
| 1472 | |
| 1473 | * swapin error entries can be found in the mapping. But they're
|
| 1474 | * deliberately ignored here as we've done everything we can do.
|
| 1475 |
|
| 1476 | if (swp_type(entry) != type)
|
| 1477 | continue;
|
| 1478 |
|
| 1479 | indices[folio_batch_count(fbatch)] = xas.xa_index;
|
| 1480 | if (!folio_batch_add(fbatch, folio))
|
| 1481 | break;
|
| 1482 |
|
| 1483 | if (need_resched()) {
|
| 1484 | xas_pause(&xas);
|
| 1485 | cond_resched_rcu();
|
| 1486 | }
|
| 1487 | }
|
| 1488 | rcu_read_unlock();
|
| 1489 |
|
| 1490 | return folio_batch_count(fbatch);
|
| 1491 | }
|
| 1492 |
|
| 1493 | |
| 1494 | * Move the swapped pages for an inode to page cache. Returns the count
|
| 1495 | * of pages swapped in, or the error in case of failure.
|
| 1496 |
|
| 1497 | static int shmem_unuse_swap_entries(struct inode *inode,
|
| 1498 | struct folio_batch *fbatch, pgoff_t *indices)
|
| 1499 | {
|
| 1500 | int i = 0;
|
| 1501 | int ret = 0;
|
| 1502 | int error = 0;
|
| 1503 | struct address_space *mapping = inode->i_mapping;
|
| 1504 |
|
| 1505 | for (i = 0; i < folio_batch_count(fbatch); i++) {
|
| 1506 | struct folio *folio = fbatch->folios[i];
|
| 1507 |
|
| 1508 | error = shmem_swapin_folio(inode, indices[i], &folio, SGP_CACHE,
|
| 1509 | mapping_gfp_mask(mapping), NULL, NULL);
|
| 1510 | if (error == 0) {
|
| 1511 | folio_unlock(folio);
|
| 1512 | folio_put(folio);
|
| 1513 | ret++;
|
| 1514 | }
|
| 1515 | if (error == -ENOMEM)
|
| 1516 | break;
|
| 1517 | error = 0;
|
| 1518 | }
|
| 1519 | return error ? error : ret;
|
| 1520 | }
|
| 1521 |
|
| 1522 | |
| 1523 | * If swap found in inode, free it and move page from swapcache to filecache.
|
| 1524 |
|
| 1525 | static int shmem_unuse_inode(struct inode *inode, unsigned int type)
|
| 1526 | {
|
| 1527 | struct address_space *mapping = inode->i_mapping;
|
| 1528 | pgoff_t start = 0;
|
| 1529 | struct folio_batch fbatch;
|
| 1530 | pgoff_t indices[FOLIO_BATCH_SIZE];
|
| 1531 | int ret = 0;
|
| 1532 |
|
| 1533 | do {
|
| 1534 | folio_batch_init(&fbatch);
|
| 1535 | if (!shmem_find_swap_entries(mapping, start, &fbatch,
|
| 1536 | indices, type)) {
|
| 1537 | ret = 0;
|
| 1538 | break;
|
| 1539 | }
|
| 1540 |
|
| 1541 | ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
|
| 1542 | if (ret < 0)
|
| 1543 | break;
|
| 1544 |
|
| 1545 | start = indices[folio_batch_count(&fbatch) - 1];
|
| 1546 | } while (true);
|
| 1547 |
|
| 1548 | return ret;
|
| 1549 | }
|
| 1550 |
|
| 1551 | |
| 1552 | * Read all the shared memory data that resides in the swap
|
| 1553 | * device 'type' back into memory, so the swap device can be
|
| 1554 | * unused.
|
| 1555 |
|
| 1556 | int shmem_unuse(unsigned int type)
|
| 1557 | {
|
| 1558 | struct shmem_inode_info *info, *next;
|
| 1559 | int error = 0;
|
| 1560 |
|
| 1561 | if (list_empty(&shmem_swaplist))
|
| 1562 | return 0;
|
| 1563 |
|
| 1564 | spin_lock(&shmem_swaplist_lock);
|
| 1565 | start_over:
|
| 1566 | list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
|
| 1567 | if (!info->swapped) {
|
| 1568 | list_del_init(&info->swaplist);
|
| 1569 | continue;
|
| 1570 | }
|
| 1571 | |
| 1572 | * Drop the swaplist mutex while searching the inode for swap;
|
| 1573 | * but before doing so, make sure shmem_evict_inode() will not
|
| 1574 | * remove placeholder inode from swaplist, nor let it be freed
|
| 1575 | * (igrab() would protect from unlink, but not from unmount).
|
| 1576 |
|
| 1577 | atomic_inc(&info->stop_eviction);
|
| 1578 | spin_unlock(&shmem_swaplist_lock);
|
| 1579 |
|
| 1580 | error = shmem_unuse_inode(&info->vfs_inode, type);
|
| 1581 | cond_resched();
|
| 1582 |
|
| 1583 | spin_lock(&shmem_swaplist_lock);
|
| 1584 | if (atomic_dec_and_test(&info->stop_eviction))
|
| 1585 | wake_up_var(&info->stop_eviction);
|
| 1586 | if (error)
|
| 1587 | break;
|
| 1588 | if (list_empty(&info->swaplist))
|
| 1589 | goto start_over;
|
| 1590 | next = list_next_entry(info, swaplist);
|
| 1591 | if (!info->swapped)
|
| 1592 | list_del_init(&info->swaplist);
|
| 1593 | }
|
| 1594 | spin_unlock(&shmem_swaplist_lock);
|
| 1595 |
|
| 1596 | return error;
|
| 1597 | }
|
| 1598 |
|
| 1599 | |
| 1600 | * shmem_writeout - Write the folio to swap
|
| 1601 | * @ctx: swap I/O context
|
| 1602 | * @folio: The folio to write
|
| 1603 | * @folio_list: list to put back folios on split
|
| 1604 | *
|
| 1605 | * Move the folio from the page cache to the swap cache.
|
| 1606 |
|
| 1607 | int shmem_writeout(struct swap_io_ctx *ctx, struct folio *folio,
|
| 1608 | struct list_head *folio_list)
|
| 1609 | {
|
| 1610 | struct address_space *mapping = folio->mapping;
|
| 1611 | struct inode *inode = mapping->host;
|
| 1612 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 1613 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 1614 | pgoff_t index;
|
| 1615 | int nr_pages;
|
| 1616 | bool split = false;
|
| 1617 |
|
| 1618 | if ((info->flags & SHMEM_F_LOCKED) || sbinfo->noswap)
|
| 1619 | goto redirty;
|
| 1620 |
|
| 1621 | if (!total_swap_pages)
|
| 1622 | goto redirty;
|
| 1623 |
|
| 1624 | |
| 1625 | * If CONFIG_THP_SWAP is not enabled, the large folio should be
|
| 1626 | * split when swapping.
|
| 1627 | *
|
| 1628 | * And shrinkage of pages beyond i_size does not split swap, so
|
| 1629 | * swapout of a large folio crossing i_size needs to split too
|
| 1630 | * (unless fallocate has been used to preallocate beyond EOF).
|
| 1631 |
|
| 1632 | if (folio_test_large(folio)) {
|
| 1633 | index = shmem_fallocend(inode,
|
| 1634 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE));
|
| 1635 | if ((index > folio->index && index < folio_next_index(folio)) ||
|
| 1636 | !IS_ENABLED(CONFIG_THP_SWAP))
|
| 1637 | split = true;
|
| 1638 | }
|
| 1639 |
|
| 1640 | if (split) {
|
| 1641 | int order;
|
| 1642 |
|
| 1643 | try_split:
|
| 1644 | order = folio_order(folio);
|
| 1645 |
|
| 1646 | folio_test_set_dirty(folio);
|
| 1647 | if (split_folio_to_list(folio, folio_list))
|
| 1648 | goto redirty;
|
| 1649 |
|
| 1650 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 1651 | if (order >= HPAGE_PMD_ORDER) {
|
| 1652 | count_memcg_folio_events(folio, THP_SWPOUT_FALLBACK, 1);
|
| 1653 | count_vm_event(THP_SWPOUT_FALLBACK);
|
| 1654 | }
|
| 1655 | #endif
|
| 1656 | count_mthp_stat(order, MTHP_STAT_SWPOUT_FALLBACK);
|
| 1657 |
|
| 1658 | folio_clear_dirty(folio);
|
| 1659 | }
|
| 1660 |
|
| 1661 | index = folio->index;
|
| 1662 | nr_pages = folio_nr_pages(folio);
|
| 1663 |
|
| 1664 | |
| 1665 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
|
| 1666 | * value into swapfile.c, the only way we can correctly account for a
|
| 1667 | * fallocated folio arriving here is now to initialize it and write it.
|
| 1668 | *
|
| 1669 | * That's okay for a folio already fallocated earlier, but if we have
|
| 1670 | * not yet completed the fallocation, then (a) we want to keep track
|
| 1671 | * of this folio in case we have to undo it, and (b) it may not be a
|
| 1672 | * good idea to continue anyway, once we're pushing into swap. So
|
| 1673 | * reactivate the folio, and let shmem_fallocate() quit when too many.
|
| 1674 |
|
| 1675 | if (!folio_test_uptodate(folio)) {
|
| 1676 | if (READ_ONCE(inode->i_private)) {
|
| 1677 | struct shmem_falloc *shmem_falloc;
|
| 1678 | spin_lock(&inode->i_lock);
|
| 1679 | shmem_falloc = inode->i_private;
|
| 1680 | if (shmem_falloc &&
|
| 1681 | !shmem_falloc->waitq &&
|
| 1682 | index >= shmem_falloc->start &&
|
| 1683 | index < shmem_falloc->next)
|
| 1684 | shmem_falloc->nr_unswapped += nr_pages;
|
| 1685 | else
|
| 1686 | shmem_falloc = NULL;
|
| 1687 | spin_unlock(&inode->i_lock);
|
| 1688 | if (shmem_falloc)
|
| 1689 | goto redirty;
|
| 1690 | }
|
| 1691 | folio_zero_range(folio, 0, folio_size(folio));
|
| 1692 | flush_dcache_folio(folio);
|
| 1693 | folio_mark_uptodate(folio);
|
| 1694 | }
|
| 1695 |
|
| 1696 | if (!folio_alloc_swap(folio)) {
|
| 1697 | bool first_swapped = shmem_recalc_inode(inode, 0, nr_pages);
|
| 1698 | int error;
|
| 1699 |
|
| 1700 | |
| 1701 | * Add inode to shmem_unuse()'s list of swapped-out inodes,
|
| 1702 | * if it's not already there. Do it now before the folio is
|
| 1703 | * removed from page cache, when its pagelock no longer
|
| 1704 | * protects the inode from eviction. And do it now, after
|
| 1705 | * we've incremented swapped, because shmem_unuse() will
|
| 1706 | * prune a !swapped inode from the swaplist.
|
| 1707 |
|
| 1708 | if (first_swapped) {
|
| 1709 | spin_lock(&shmem_swaplist_lock);
|
| 1710 | if (list_empty(&info->swaplist))
|
| 1711 | list_add(&info->swaplist, &shmem_swaplist);
|
| 1712 | spin_unlock(&shmem_swaplist_lock);
|
| 1713 | }
|
| 1714 |
|
| 1715 | folio_dup_swap(folio, NULL);
|
| 1716 | shmem_delete_from_page_cache(folio, swp_to_radix_entry(folio->swap));
|
| 1717 |
|
| 1718 | BUG_ON(folio_mapped(folio));
|
| 1719 | error = swap_writeout(ctx, folio);
|
| 1720 | if (error != AOP_WRITEPAGE_ACTIVATE) {
|
| 1721 |
|
| 1722 | return error;
|
| 1723 | }
|
| 1724 |
|
| 1725 | |
| 1726 | * The intention here is to avoid holding on to the swap when
|
| 1727 | * zswap was unable to compress and unable to writeback; but
|
| 1728 | * it will be appropriate if other reactivate cases are added.
|
| 1729 |
|
| 1730 | error = shmem_add_to_page_cache(folio, mapping, index,
|
| 1731 | swp_to_radix_entry(folio->swap),
|
| 1732 | __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN);
|
| 1733 |
|
| 1734 | if (!error) {
|
| 1735 | shmem_recalc_inode(inode, 0, -nr_pages);
|
| 1736 | folio_put_swap(folio, NULL);
|
| 1737 | }
|
| 1738 |
|
| 1739 | |
| 1740 | * The swap_cache_del_folio() below could be left for
|
| 1741 | * shrink_folio_list()'s folio_free_swap() to dispose of;
|
| 1742 | * but I'm a little nervous about letting this folio out of
|
| 1743 | * shmem_writeout() in a hybrid half-tmpfs-half-swap state
|
| 1744 | * e.g. folio_mapping(folio) might give an unexpected answer.
|
| 1745 |
|
| 1746 | swap_cache_del_folio(folio);
|
| 1747 | goto redirty;
|
| 1748 | }
|
| 1749 | if (nr_pages > 1)
|
| 1750 | goto try_split;
|
| 1751 | redirty:
|
| 1752 | folio_mark_dirty(folio);
|
| 1753 | return AOP_WRITEPAGE_ACTIVATE;
|
| 1754 | }
|
| 1755 |
|
| 1756 | int shmem_write_folio(struct folio *folio)
|
| 1757 | {
|
| 1758 | struct swap_io_ctx ctx = {};
|
| 1759 | int err;
|
| 1760 |
|
| 1761 | err = shmem_writeout(&ctx, folio, NULL);
|
| 1762 | swap_write_submit(&ctx);
|
| 1763 | return err;
|
| 1764 | }
|
| 1765 | EXPORT_SYMBOL_GPL(shmem_write_folio);
|
| 1766 |
|
| 1767 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
|
| 1768 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
|
| 1769 | {
|
| 1770 | char buffer[64];
|
| 1771 |
|
| 1772 | if (!mpol || mpol->mode == MPOL_DEFAULT)
|
| 1773 | return;
|
| 1774 |
|
| 1775 | mpol_to_str(buffer, sizeof(buffer), mpol);
|
| 1776 |
|
| 1777 | seq_printf(seq, ",mpol=%s", buffer);
|
| 1778 | }
|
| 1779 |
|
| 1780 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
|
| 1781 | {
|
| 1782 | struct mempolicy *mpol = NULL;
|
| 1783 | if (sbinfo->mpol) {
|
| 1784 | raw_spin_lock(&sbinfo->stat_lock);
|
| 1785 | mpol = sbinfo->mpol;
|
| 1786 | mpol_get(mpol);
|
| 1787 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 1788 | }
|
| 1789 | return mpol;
|
| 1790 | }
|
| 1791 | #else
|
| 1792 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
|
| 1793 | {
|
| 1794 | }
|
| 1795 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
|
| 1796 | {
|
| 1797 | return NULL;
|
| 1798 | }
|
| 1799 | #endif
|
| 1800 |
|
| 1801 | static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
|
| 1802 | pgoff_t index, unsigned int order, pgoff_t *ilx);
|
| 1803 |
|
| 1804 | static struct folio *shmem_swapin_cluster(swp_entry_t swap, gfp_t gfp,
|
| 1805 | struct shmem_inode_info *info, pgoff_t index)
|
| 1806 | {
|
| 1807 | struct mempolicy *mpol;
|
| 1808 | pgoff_t ilx;
|
| 1809 | struct folio *folio;
|
| 1810 |
|
| 1811 | mpol = shmem_get_pgoff_policy(info, index, 0, &ilx);
|
| 1812 | folio = swap_cluster_readahead(swap, gfp, mpol, ilx);
|
| 1813 | mpol_cond_put(mpol);
|
| 1814 |
|
| 1815 | return folio;
|
| 1816 | }
|
| 1817 |
|
| 1818 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 1819 | bool shmem_hpage_pmd_enabled(void)
|
| 1820 | {
|
| 1821 | if (shmem_huge == SHMEM_HUGE_DENY)
|
| 1822 | return false;
|
| 1823 | if (test_bit(HPAGE_PMD_ORDER, &huge_shmem_orders_always))
|
| 1824 | return true;
|
| 1825 | if (test_bit(HPAGE_PMD_ORDER, &huge_shmem_orders_madvise))
|
| 1826 | return true;
|
| 1827 | if (test_bit(HPAGE_PMD_ORDER, &huge_shmem_orders_within_size))
|
| 1828 | return true;
|
| 1829 | if (test_bit(HPAGE_PMD_ORDER, &huge_shmem_orders_inherit) &&
|
| 1830 | shmem_huge != SHMEM_HUGE_NEVER)
|
| 1831 | return true;
|
| 1832 |
|
| 1833 | return false;
|
| 1834 | }
|
| 1835 |
|
| 1836 | unsigned long shmem_allowable_huge_orders(struct inode *inode,
|
| 1837 | struct vm_area_struct *vma, pgoff_t index,
|
| 1838 | loff_t write_end, bool shmem_huge_force)
|
| 1839 | {
|
| 1840 | unsigned long mask = READ_ONCE(huge_shmem_orders_always);
|
| 1841 | unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
|
| 1842 | vm_flags_t vm_flags = vma ? vma->vm_flags : 0;
|
| 1843 | unsigned int global_orders;
|
| 1844 |
|
| 1845 | if (thp_disabled_by_hw() || (vma && vma_thp_disabled(vma, vm_flags, shmem_huge_force)))
|
| 1846 | return 0;
|
| 1847 |
|
| 1848 | global_orders = shmem_huge_global_enabled(inode, index, write_end,
|
| 1849 | shmem_huge_force, vma, vm_flags);
|
| 1850 | |
| 1851 | * Tmpfs huge pages allocation or forced collapse ignores
|
| 1852 | * sysfs configs.
|
| 1853 |
|
| 1854 | if (!vma || !vma_is_anon_shmem(vma) || shmem_huge_force)
|
| 1855 | return global_orders;
|
| 1856 |
|
| 1857 | |
| 1858 | * Following the 'deny' semantics of the top level, force the huge
|
| 1859 | * option off from all mounts.
|
| 1860 |
|
| 1861 | if (shmem_huge == SHMEM_HUGE_DENY)
|
| 1862 | return 0;
|
| 1863 |
|
| 1864 | |
| 1865 | * Only allow inherit orders if the top-level value is 'force', which
|
| 1866 | * means non-PMD sized THP can not override 'huge' mount option now.
|
| 1867 |
|
| 1868 | if (shmem_huge == SHMEM_HUGE_FORCE)
|
| 1869 | return READ_ONCE(huge_shmem_orders_inherit);
|
| 1870 |
|
| 1871 |
|
| 1872 | mask |= shmem_get_orders_within_size(inode, within_size_orders, index, 0);
|
| 1873 |
|
| 1874 | if (vm_flags & VM_HUGEPAGE)
|
| 1875 | mask |= READ_ONCE(huge_shmem_orders_madvise);
|
| 1876 |
|
| 1877 | if (global_orders > 0)
|
| 1878 | mask |= READ_ONCE(huge_shmem_orders_inherit);
|
| 1879 |
|
| 1880 | return THP_ORDERS_ALL_FILE_DEFAULT & mask;
|
| 1881 | }
|
| 1882 |
|
| 1883 | static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
|
| 1884 | struct address_space *mapping, pgoff_t index,
|
| 1885 | unsigned long orders)
|
| 1886 | {
|
| 1887 | struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
|
| 1888 | pgoff_t aligned_index;
|
| 1889 | unsigned long pages;
|
| 1890 | int order;
|
| 1891 |
|
| 1892 | if (vma) {
|
| 1893 | orders = thp_vma_suitable_orders(vma, vmf->address, orders);
|
| 1894 | if (!orders)
|
| 1895 | return 0;
|
| 1896 | }
|
| 1897 |
|
| 1898 |
|
| 1899 | order = highest_order(orders);
|
| 1900 | while (orders) {
|
| 1901 | pages = 1UL << order;
|
| 1902 | aligned_index = round_down(index, pages);
|
| 1903 | |
| 1904 | * Check for conflict before waiting on a huge allocation.
|
| 1905 | * Conflict might be that a huge page has just been allocated
|
| 1906 | * and added to page cache by a racing thread, or that there
|
| 1907 | * is already at least one small page in the huge extent.
|
| 1908 | * Be careful to retry when appropriate, but not forever!
|
| 1909 | * Elsewhere -EEXIST would be the right code, but not here.
|
| 1910 |
|
| 1911 | if (!xa_find(&mapping->i_pages, &aligned_index,
|
| 1912 | aligned_index + pages - 1, XA_PRESENT))
|
| 1913 | break;
|
| 1914 | order = next_order(&orders, order);
|
| 1915 | }
|
| 1916 |
|
| 1917 | return orders;
|
| 1918 | }
|
| 1919 | #else
|
| 1920 | static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
|
| 1921 | struct address_space *mapping, pgoff_t index,
|
| 1922 | unsigned long orders)
|
| 1923 | {
|
| 1924 | return 0;
|
| 1925 | }
|
| 1926 | #endif
|
| 1927 |
|
| 1928 | static struct folio *shmem_alloc_folio(gfp_t gfp, int order,
|
| 1929 | struct shmem_inode_info *info, pgoff_t index)
|
| 1930 | {
|
| 1931 | struct mempolicy *mpol;
|
| 1932 | pgoff_t ilx;
|
| 1933 | struct folio *folio;
|
| 1934 |
|
| 1935 | mpol = shmem_get_pgoff_policy(info, index, order, &ilx);
|
| 1936 | folio = folio_alloc_mpol(gfp, order, mpol, ilx, numa_node_id());
|
| 1937 | mpol_cond_put(mpol);
|
| 1938 |
|
| 1939 | return folio;
|
| 1940 | }
|
| 1941 |
|
| 1942 | static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,
|
| 1943 | gfp_t gfp, struct inode *inode, pgoff_t index,
|
| 1944 | struct mm_struct *fault_mm, unsigned long orders)
|
| 1945 | {
|
| 1946 | struct address_space *mapping = inode->i_mapping;
|
| 1947 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 1948 | unsigned long suitable_orders = 0;
|
| 1949 | struct folio *folio = NULL;
|
| 1950 | pgoff_t aligned_index;
|
| 1951 | long pages;
|
| 1952 | int error, order;
|
| 1953 |
|
| 1954 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
|
| 1955 | orders = 0;
|
| 1956 |
|
| 1957 | if (orders > 0) {
|
| 1958 | suitable_orders = shmem_suitable_orders(inode, vmf,
|
| 1959 | mapping, index, orders);
|
| 1960 |
|
| 1961 | order = highest_order(suitable_orders);
|
| 1962 | while (suitable_orders) {
|
| 1963 | pages = 1UL << order;
|
| 1964 | aligned_index = round_down(index, pages);
|
| 1965 | folio = shmem_alloc_folio(gfp, order, info, aligned_index);
|
| 1966 | if (folio) {
|
| 1967 | index = aligned_index;
|
| 1968 | goto allocated;
|
| 1969 | }
|
| 1970 |
|
| 1971 | if (pages == HPAGE_PMD_NR)
|
| 1972 | count_vm_event(THP_FILE_FALLBACK);
|
| 1973 | count_mthp_stat(order, MTHP_STAT_SHMEM_FALLBACK);
|
| 1974 | order = next_order(&suitable_orders, order);
|
| 1975 | }
|
| 1976 | } else {
|
| 1977 | pages = 1;
|
| 1978 | folio = shmem_alloc_folio(gfp, 0, info, index);
|
| 1979 | }
|
| 1980 | if (!folio)
|
| 1981 | return ERR_PTR(-ENOMEM);
|
| 1982 |
|
| 1983 | allocated:
|
| 1984 | __folio_set_locked(folio);
|
| 1985 | __folio_set_swapbacked(folio);
|
| 1986 |
|
| 1987 | gfp &= GFP_RECLAIM_MASK;
|
| 1988 | error = mem_cgroup_charge(folio, fault_mm, gfp);
|
| 1989 | if (error) {
|
| 1990 | if (xa_find(&mapping->i_pages, &index,
|
| 1991 | index + pages - 1, XA_PRESENT)) {
|
| 1992 | error = -EEXIST;
|
| 1993 | } else if (pages > 1) {
|
| 1994 | if (pages == HPAGE_PMD_NR) {
|
| 1995 | count_vm_event(THP_FILE_FALLBACK);
|
| 1996 | count_vm_event(THP_FILE_FALLBACK_CHARGE);
|
| 1997 | }
|
| 1998 | count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK);
|
| 1999 | count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK_CHARGE);
|
| 2000 | }
|
| 2001 | goto unlock;
|
| 2002 | }
|
| 2003 |
|
| 2004 | error = shmem_add_to_page_cache(folio, mapping, index, NULL, gfp);
|
| 2005 | if (error)
|
| 2006 | goto unlock;
|
| 2007 |
|
| 2008 | error = shmem_inode_acct_blocks(inode, pages);
|
| 2009 | if (error) {
|
| 2010 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 2011 | long freed;
|
| 2012 | |
| 2013 | * Try to reclaim some space by splitting a few
|
| 2014 | * large folios beyond i_size on the filesystem.
|
| 2015 |
|
| 2016 | shmem_unused_huge_shrink(sbinfo, NULL, pages);
|
| 2017 | |
| 2018 | * And do a shmem_recalc_inode() to account for freed pages:
|
| 2019 | * except our folio is there in cache, so not quite balanced.
|
| 2020 |
|
| 2021 | spin_lock(&info->lock);
|
| 2022 | freed = pages + info->alloced - info->swapped -
|
| 2023 | READ_ONCE(mapping->nrpages);
|
| 2024 | if (freed > 0)
|
| 2025 | info->alloced -= freed;
|
| 2026 | spin_unlock(&info->lock);
|
| 2027 | if (freed > 0)
|
| 2028 | shmem_inode_unacct_blocks(inode, freed);
|
| 2029 | error = shmem_inode_acct_blocks(inode, pages);
|
| 2030 | if (error) {
|
| 2031 | filemap_remove_folio(folio);
|
| 2032 | goto unlock;
|
| 2033 | }
|
| 2034 | }
|
| 2035 |
|
| 2036 | shmem_recalc_inode(inode, pages, 0);
|
| 2037 | folio_add_lru(folio);
|
| 2038 | return folio;
|
| 2039 |
|
| 2040 | unlock:
|
| 2041 | folio_unlock(folio);
|
| 2042 | folio_put(folio);
|
| 2043 | return ERR_PTR(error);
|
| 2044 | }
|
| 2045 |
|
| 2046 | static struct folio *shmem_swap_alloc_folio(struct inode *inode,
|
| 2047 | struct vm_fault *vmf, pgoff_t index,
|
| 2048 | swp_entry_t entry, int order, gfp_t gfp)
|
| 2049 | {
|
| 2050 | pgoff_t ilx;
|
| 2051 | struct folio *folio;
|
| 2052 | struct mempolicy *mpol;
|
| 2053 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 2054 |
|
| 2055 | if ((vmf && unlikely(userfaultfd_armed(vmf->vma))) ||
|
| 2056 | !zswap_never_enabled())
|
| 2057 | order = 0;
|
| 2058 |
|
| 2059 | again:
|
| 2060 | mpol = shmem_get_pgoff_policy(info, index, order, &ilx);
|
| 2061 | folio = swapin_sync(entry, gfp, BIT(order), vmf, mpol, ilx);
|
| 2062 | mpol_cond_put(mpol);
|
| 2063 |
|
| 2064 | if (!IS_ERR(folio))
|
| 2065 | return folio;
|
| 2066 |
|
| 2067 | if (order) {
|
| 2068 | order = 0;
|
| 2069 | goto again;
|
| 2070 | }
|
| 2071 |
|
| 2072 | return folio;
|
| 2073 | }
|
| 2074 |
|
| 2075 | |
| 2076 | * When a page is moved from swapcache to shmem filecache (either by the
|
| 2077 | * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
|
| 2078 | * shmem_unuse_inode()), it may have been read in earlier from swap, in
|
| 2079 | * ignorance of the mapping it belongs to. If that mapping has special
|
| 2080 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
|
| 2081 | * we may need to copy to a suitable page before moving to filecache.
|
| 2082 | *
|
| 2083 | * In a future release, this may well be extended to respect cpuset and
|
| 2084 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
|
| 2085 | * but for now it is a simple matter of zone.
|
| 2086 |
|
| 2087 | static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
|
| 2088 | {
|
| 2089 | return folio_zonenum(folio) > gfp_zone(gfp);
|
| 2090 | }
|
| 2091 |
|
| 2092 | static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
|
| 2093 | struct shmem_inode_info *info, pgoff_t index,
|
| 2094 | struct vm_area_struct *vma)
|
| 2095 | {
|
| 2096 | struct swap_cluster_info *ci;
|
| 2097 | struct folio *new, *old = *foliop;
|
| 2098 | swp_entry_t entry = old->swap;
|
| 2099 | int nr_pages = folio_nr_pages(old);
|
| 2100 | int error = 0;
|
| 2101 |
|
| 2102 | |
| 2103 | * We have arrived here because our zones are constrained, so don't
|
| 2104 | * limit chance of success by further cpuset and node constraints.
|
| 2105 |
|
| 2106 | gfp &= ~GFP_CONSTRAINT_MASK;
|
| 2107 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 2108 | if (nr_pages > 1) {
|
| 2109 | gfp_t huge_gfp = vma_thp_gfp_mask(vma);
|
| 2110 |
|
| 2111 | gfp = thp_shmem_limit_gfp_mask(huge_gfp, gfp);
|
| 2112 | }
|
| 2113 | #endif
|
| 2114 |
|
| 2115 | new = shmem_alloc_folio(gfp, folio_order(old), info, index);
|
| 2116 | if (!new)
|
| 2117 | return -ENOMEM;
|
| 2118 |
|
| 2119 | folio_ref_add(new, nr_pages);
|
| 2120 | folio_copy(new, old);
|
| 2121 | flush_dcache_folio(new);
|
| 2122 |
|
| 2123 | __folio_set_locked(new);
|
| 2124 | __folio_set_swapbacked(new);
|
| 2125 | folio_mark_uptodate(new);
|
| 2126 | new->swap = entry;
|
| 2127 | folio_set_swapcache(new);
|
| 2128 |
|
| 2129 | ci = swap_cluster_get_and_lock_irq(old);
|
| 2130 | __swap_cache_replace_folio(ci, old, new);
|
| 2131 | mem_cgroup_replace_folio(old, new);
|
| 2132 | shmem_update_stats(new, nr_pages);
|
| 2133 | shmem_update_stats(old, -nr_pages);
|
| 2134 | swap_cluster_unlock_irq(ci);
|
| 2135 |
|
| 2136 | folio_add_lru(new);
|
| 2137 | *foliop = new;
|
| 2138 |
|
| 2139 | folio_clear_swapcache(old);
|
| 2140 | old->private = NULL;
|
| 2141 |
|
| 2142 | folio_unlock(old);
|
| 2143 | |
| 2144 | * The old folio are removed from swap cache, drop the 'nr_pages'
|
| 2145 | * reference, as well as one temporary reference getting from swap
|
| 2146 | * cache.
|
| 2147 |
|
| 2148 | folio_put_refs(old, nr_pages + 1);
|
| 2149 | return error;
|
| 2150 | }
|
| 2151 |
|
| 2152 | static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
|
| 2153 | struct folio *folio, swp_entry_t swap)
|
| 2154 | {
|
| 2155 | struct address_space *mapping = inode->i_mapping;
|
| 2156 | swp_entry_t swapin_error;
|
| 2157 | void *old;
|
| 2158 | int nr_pages;
|
| 2159 |
|
| 2160 | swapin_error = make_poisoned_swp_entry();
|
| 2161 | old = xa_cmpxchg_irq(&mapping->i_pages, index,
|
| 2162 | swp_to_radix_entry(swap),
|
| 2163 | swp_to_radix_entry(swapin_error), 0);
|
| 2164 | if (old != swp_to_radix_entry(swap))
|
| 2165 | return;
|
| 2166 |
|
| 2167 | nr_pages = folio_nr_pages(folio);
|
| 2168 | folio_wait_writeback(folio);
|
| 2169 | folio_put_swap(folio, NULL);
|
| 2170 | swap_cache_del_folio(folio);
|
| 2171 | |
| 2172 | * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
|
| 2173 | * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
|
| 2174 | * in shmem_evict_inode().
|
| 2175 |
|
| 2176 | shmem_recalc_inode(inode, -nr_pages, -nr_pages);
|
| 2177 | }
|
| 2178 |
|
| 2179 | static int shmem_split_large_entry(struct inode *inode, pgoff_t index,
|
| 2180 | swp_entry_t swap, gfp_t gfp)
|
| 2181 | {
|
| 2182 | struct address_space *mapping = inode->i_mapping;
|
| 2183 | XA_STATE_ORDER(xas, &mapping->i_pages, index, 0);
|
| 2184 | int split_order = 0;
|
| 2185 | int i;
|
| 2186 |
|
| 2187 |
|
| 2188 | gfp &= GFP_RECLAIM_MASK;
|
| 2189 |
|
| 2190 | for (;;) {
|
| 2191 | void *old = NULL;
|
| 2192 | int cur_order;
|
| 2193 | pgoff_t swap_index;
|
| 2194 |
|
| 2195 | xas_lock_irq(&xas);
|
| 2196 | old = xas_load(&xas);
|
| 2197 | if (!xa_is_value(old) || swp_to_radix_entry(swap) != old) {
|
| 2198 | xas_set_err(&xas, -EEXIST);
|
| 2199 | goto unlock;
|
| 2200 | }
|
| 2201 |
|
| 2202 | cur_order = xas_get_order(&xas);
|
| 2203 | if (!cur_order)
|
| 2204 | goto unlock;
|
| 2205 |
|
| 2206 |
|
| 2207 | swap_index = round_down(index, 1 << cur_order);
|
| 2208 | split_order = xas_try_split_min_order(cur_order);
|
| 2209 |
|
| 2210 | while (cur_order > 0) {
|
| 2211 | pgoff_t aligned_index =
|
| 2212 | round_down(index, 1 << cur_order);
|
| 2213 | pgoff_t swap_offset = aligned_index - swap_index;
|
| 2214 |
|
| 2215 | xas_set_order(&xas, index, split_order);
|
| 2216 | xas_try_split(&xas, old, cur_order);
|
| 2217 | if (xas_error(&xas))
|
| 2218 | goto unlock;
|
| 2219 |
|
| 2220 | |
| 2221 | * Re-set the swap entry after splitting, and the swap
|
| 2222 | * offset of the original large entry must be continuous.
|
| 2223 |
|
| 2224 | for (i = 0; i < 1 << cur_order;
|
| 2225 | i += (1 << split_order)) {
|
| 2226 | swp_entry_t tmp;
|
| 2227 |
|
| 2228 | tmp = swp_entry(swp_type(swap),
|
| 2229 | swp_offset(swap) + swap_offset +
|
| 2230 | i);
|
| 2231 | __xa_store(&mapping->i_pages, aligned_index + i,
|
| 2232 | swp_to_radix_entry(tmp), 0);
|
| 2233 | }
|
| 2234 | cur_order = split_order;
|
| 2235 | split_order = xas_try_split_min_order(split_order);
|
| 2236 | }
|
| 2237 |
|
| 2238 | unlock:
|
| 2239 | xas_unlock_irq(&xas);
|
| 2240 |
|
| 2241 | if (!xas_nomem(&xas, gfp))
|
| 2242 | break;
|
| 2243 | }
|
| 2244 |
|
| 2245 | if (xas_error(&xas))
|
| 2246 | return xas_error(&xas);
|
| 2247 |
|
| 2248 | return 0;
|
| 2249 | }
|
| 2250 |
|
| 2251 | |
| 2252 | * Swap in the folio pointed to by *foliop.
|
| 2253 | * Caller has to make sure that *foliop contains a valid swapped folio.
|
| 2254 | * Returns 0 and the folio in foliop if success. On failure, returns the
|
| 2255 | * error code and NULL in *foliop.
|
| 2256 |
|
| 2257 | static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
|
| 2258 | struct folio **foliop, enum sgp_type sgp,
|
| 2259 | gfp_t gfp, struct vm_fault *vmf,
|
| 2260 | vm_fault_t *fault_type)
|
| 2261 | {
|
| 2262 | struct address_space *mapping = inode->i_mapping;
|
| 2263 | struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
|
| 2264 | struct mm_struct *fault_mm = vmf ? vmf->vma->vm_mm : NULL;
|
| 2265 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 2266 | swp_entry_t swap;
|
| 2267 | softleaf_t index_entry;
|
| 2268 | struct swap_info_struct *si;
|
| 2269 | struct folio *folio = NULL;
|
| 2270 | int error, nr_pages, order;
|
| 2271 | pgoff_t offset;
|
| 2272 |
|
| 2273 | VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
|
| 2274 | index_entry = radix_to_swp_entry(*foliop);
|
| 2275 | swap = index_entry;
|
| 2276 | *foliop = NULL;
|
| 2277 |
|
| 2278 | if (softleaf_is_poison_marker(index_entry))
|
| 2279 | return -EIO;
|
| 2280 |
|
| 2281 | si = get_swap_device(index_entry);
|
| 2282 | order = shmem_confirm_swap(mapping, index, index_entry);
|
| 2283 | if (unlikely(!si)) {
|
| 2284 | if (order < 0)
|
| 2285 | return -EEXIST;
|
| 2286 | else
|
| 2287 | return -EINVAL;
|
| 2288 | }
|
| 2289 | if (unlikely(order < 0)) {
|
| 2290 | put_swap_device(si);
|
| 2291 | return -EEXIST;
|
| 2292 | }
|
| 2293 |
|
| 2294 |
|
| 2295 | if (order) {
|
| 2296 | offset = index - round_down(index, 1 << order);
|
| 2297 | swap = swp_entry(swp_type(swap), swp_offset(swap) + offset);
|
| 2298 | }
|
| 2299 |
|
| 2300 |
|
| 2301 | folio = swap_cache_get_folio(swap);
|
| 2302 | if (!folio) {
|
| 2303 | if (data_race(si->flags & SWP_SYNCHRONOUS_IO)) {
|
| 2304 |
|
| 2305 | folio = shmem_swap_alloc_folio(inode, vmf, index,
|
| 2306 | swap, order, gfp);
|
| 2307 | } else {
|
| 2308 |
|
| 2309 | folio = shmem_swapin_cluster(swap, gfp, info, index);
|
| 2310 | }
|
| 2311 | if (IS_ERR_OR_NULL(folio)) {
|
| 2312 | if (IS_ERR(folio))
|
| 2313 | error = PTR_ERR(folio);
|
| 2314 | else
|
| 2315 | error = -ENOMEM;
|
| 2316 | folio = NULL;
|
| 2317 | goto failed;
|
| 2318 | }
|
| 2319 | if (fault_type) {
|
| 2320 | *fault_type |= VM_FAULT_MAJOR;
|
| 2321 | count_vm_event(PGMAJFAULT);
|
| 2322 | count_memcg_event_mm(fault_mm, PGMAJFAULT);
|
| 2323 | }
|
| 2324 | } else {
|
| 2325 | swap_update_readahead(folio, NULL, 0);
|
| 2326 | }
|
| 2327 |
|
| 2328 | if (order > folio_order(folio)) {
|
| 2329 | |
| 2330 | * Swapin may get smaller folios due to various reasons:
|
| 2331 | * It may fallback to order 0 due to memory pressure or race,
|
| 2332 | * swap readahead may swap in order 0 folios into swapcache
|
| 2333 | * asynchronously, while the shmem mapping can still stores
|
| 2334 | * large swap entries. In such cases, we should split the
|
| 2335 | * large swap entry to prevent possible data corruption.
|
| 2336 |
|
| 2337 | error = shmem_split_large_entry(inode, index, index_entry, gfp);
|
| 2338 | if (error)
|
| 2339 | goto failed_nolock;
|
| 2340 | }
|
| 2341 |
|
| 2342 | |
| 2343 | * If the folio is large, round down swap and index by folio size.
|
| 2344 | * No matter what race occurs, the swap layer ensures we either get
|
| 2345 | * a valid folio that has its swap entry aligned by size, or a
|
| 2346 | * temporarily invalid one which we'll abort very soon and retry.
|
| 2347 | *
|
| 2348 | * shmem_add_to_page_cache ensures the whole range contains expected
|
| 2349 | * entries and prevents any corruption, so any race split is fine
|
| 2350 | * too, it will succeed as long as the entries are still there.
|
| 2351 |
|
| 2352 | nr_pages = folio_nr_pages(folio);
|
| 2353 | if (nr_pages > 1) {
|
| 2354 | swap.val = round_down(swap.val, nr_pages);
|
| 2355 | index = round_down(index, nr_pages);
|
| 2356 | }
|
| 2357 |
|
| 2358 | |
| 2359 | * We have to do this with the folio locked to prevent races.
|
| 2360 | * The shmem_confirm_swap below only checks if the first swap
|
| 2361 | * entry matches the folio, that's enough to ensure the folio
|
| 2362 | * is not used outside of shmem, as shmem swap entries
|
| 2363 | * and swap cache folios are never partially freed.
|
| 2364 |
|
| 2365 | folio_lock(folio);
|
| 2366 | if (!folio_matches_swap_entry(folio, swap) ||
|
| 2367 | shmem_confirm_swap(mapping, index, swap) < 0) {
|
| 2368 | error = -EEXIST;
|
| 2369 | goto unlock;
|
| 2370 | }
|
| 2371 | if (!folio_test_uptodate(folio)) {
|
| 2372 | error = -EIO;
|
| 2373 | goto failed;
|
| 2374 | }
|
| 2375 | folio_wait_writeback(folio);
|
| 2376 |
|
| 2377 | |
| 2378 | * Some architectures may have to restore extra metadata to the
|
| 2379 | * folio after reading from swap.
|
| 2380 |
|
| 2381 | arch_swap_restore(folio_swap(swap, folio), folio);
|
| 2382 |
|
| 2383 | if (shmem_should_replace_folio(folio, gfp)) {
|
| 2384 | error = shmem_replace_folio(&folio, gfp, info, index, vma);
|
| 2385 | if (error)
|
| 2386 | goto failed;
|
| 2387 | }
|
| 2388 |
|
| 2389 | error = shmem_add_to_page_cache(folio, mapping, index,
|
| 2390 | swp_to_radix_entry(swap), gfp);
|
| 2391 | if (error)
|
| 2392 | goto failed;
|
| 2393 |
|
| 2394 | shmem_recalc_inode(inode, 0, -nr_pages);
|
| 2395 |
|
| 2396 | if (sgp == SGP_WRITE)
|
| 2397 | folio_mark_accessed(folio);
|
| 2398 |
|
| 2399 | folio_put_swap(folio, NULL);
|
| 2400 | swap_cache_del_folio(folio);
|
| 2401 | folio_mark_dirty(folio);
|
| 2402 | put_swap_device(si);
|
| 2403 |
|
| 2404 | *foliop = folio;
|
| 2405 | return 0;
|
| 2406 | failed:
|
| 2407 | if (shmem_confirm_swap(mapping, index, swap) < 0)
|
| 2408 | error = -EEXIST;
|
| 2409 | if (error == -EIO)
|
| 2410 | shmem_set_folio_swapin_error(inode, index, folio, swap);
|
| 2411 | unlock:
|
| 2412 | if (folio)
|
| 2413 | folio_unlock(folio);
|
| 2414 | failed_nolock:
|
| 2415 | if (folio)
|
| 2416 | folio_put(folio);
|
| 2417 | put_swap_device(si);
|
| 2418 |
|
| 2419 | return error;
|
| 2420 | }
|
| 2421 |
|
| 2422 | |
| 2423 | * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
|
| 2424 | *
|
| 2425 | * If we allocate a new one we do not mark it dirty. That's up to the
|
| 2426 | * vm. If we swap it in we mark it dirty since we also free the swap
|
| 2427 | * entry since a page cannot live in both the swap and page cache.
|
| 2428 | *
|
| 2429 | * vmf and fault_type are only supplied by shmem_fault: otherwise they are NULL.
|
| 2430 |
|
| 2431 | static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
|
| 2432 | loff_t write_end, struct folio **foliop, enum sgp_type sgp,
|
| 2433 | gfp_t gfp, struct vm_fault *vmf, vm_fault_t *fault_type)
|
| 2434 | {
|
| 2435 | struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
|
| 2436 | struct mm_struct *fault_mm;
|
| 2437 | struct folio *folio;
|
| 2438 | int error;
|
| 2439 | bool alloced;
|
| 2440 | unsigned long orders = 0;
|
| 2441 |
|
| 2442 | if (WARN_ON_ONCE(!shmem_mapping(inode->i_mapping)))
|
| 2443 | return -EINVAL;
|
| 2444 |
|
| 2445 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
|
| 2446 | return -EFBIG;
|
| 2447 | repeat:
|
| 2448 | if (sgp <= SGP_CACHE &&
|
| 2449 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode))
|
| 2450 | return -EINVAL;
|
| 2451 |
|
| 2452 | alloced = false;
|
| 2453 | fault_mm = vma ? vma->vm_mm : NULL;
|
| 2454 |
|
| 2455 | folio = filemap_get_entry(inode->i_mapping, index);
|
| 2456 | if (folio && vma && userfaultfd_minor(vma)) {
|
| 2457 | if (!xa_is_value(folio))
|
| 2458 | folio_put(folio);
|
| 2459 | *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
|
| 2460 | return 0;
|
| 2461 | }
|
| 2462 |
|
| 2463 | if (xa_is_value(folio)) {
|
| 2464 | error = shmem_swapin_folio(inode, index, &folio,
|
| 2465 | sgp, gfp, vmf, fault_type);
|
| 2466 | if (error == -EEXIST)
|
| 2467 | goto repeat;
|
| 2468 |
|
| 2469 | *foliop = folio;
|
| 2470 | return error;
|
| 2471 | }
|
| 2472 |
|
| 2473 | if (folio) {
|
| 2474 | folio_lock(folio);
|
| 2475 |
|
| 2476 |
|
| 2477 | if (unlikely(folio->mapping != inode->i_mapping)) {
|
| 2478 | folio_unlock(folio);
|
| 2479 | folio_put(folio);
|
| 2480 | goto repeat;
|
| 2481 | }
|
| 2482 | if (sgp == SGP_WRITE)
|
| 2483 | folio_mark_accessed(folio);
|
| 2484 | if (folio_test_uptodate(folio))
|
| 2485 | goto out;
|
| 2486 |
|
| 2487 | if (sgp != SGP_READ)
|
| 2488 | goto clear;
|
| 2489 | folio_unlock(folio);
|
| 2490 | folio_put(folio);
|
| 2491 | }
|
| 2492 |
|
| 2493 | |
| 2494 | * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
|
| 2495 | * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
|
| 2496 |
|
| 2497 | *foliop = NULL;
|
| 2498 | if (sgp == SGP_READ)
|
| 2499 | return 0;
|
| 2500 | if (sgp == SGP_NOALLOC)
|
| 2501 | return -ENOENT;
|
| 2502 |
|
| 2503 | |
| 2504 | * Fast cache lookup and swap lookup did not find it: allocate.
|
| 2505 |
|
| 2506 |
|
| 2507 | if (vma && userfaultfd_missing(vma)) {
|
| 2508 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
|
| 2509 | return 0;
|
| 2510 | }
|
| 2511 |
|
| 2512 |
|
| 2513 | orders = shmem_allowable_huge_orders(inode, vma, index, write_end, false);
|
| 2514 | if (orders > 0) {
|
| 2515 | gfp_t huge_gfp;
|
| 2516 |
|
| 2517 | huge_gfp = vma_thp_gfp_mask(vma);
|
| 2518 | huge_gfp = thp_shmem_limit_gfp_mask(huge_gfp, gfp);
|
| 2519 | folio = shmem_alloc_and_add_folio(vmf, huge_gfp,
|
| 2520 | inode, index, fault_mm, orders);
|
| 2521 | if (!IS_ERR(folio)) {
|
| 2522 | if (folio_test_pmd_mappable(folio))
|
| 2523 | count_vm_event(THP_FILE_ALLOC);
|
| 2524 | count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_ALLOC);
|
| 2525 | goto alloced;
|
| 2526 | }
|
| 2527 | if (PTR_ERR(folio) == -EEXIST)
|
| 2528 | goto repeat;
|
| 2529 | }
|
| 2530 |
|
| 2531 | folio = shmem_alloc_and_add_folio(vmf, gfp, inode, index, fault_mm, 0);
|
| 2532 | if (IS_ERR(folio)) {
|
| 2533 | error = PTR_ERR(folio);
|
| 2534 | if (error == -EEXIST)
|
| 2535 | goto repeat;
|
| 2536 | folio = NULL;
|
| 2537 | goto unlock;
|
| 2538 | }
|
| 2539 |
|
| 2540 | alloced:
|
| 2541 | alloced = true;
|
| 2542 | if (folio_test_large(folio) &&
|
| 2543 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
|
| 2544 | folio_next_index(folio)) {
|
| 2545 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 2546 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 2547 | |
| 2548 | * Part of the large folio is beyond i_size: subject
|
| 2549 | * to shrink under memory pressure.
|
| 2550 |
|
| 2551 | spin_lock(&sbinfo->shrinklist_lock);
|
| 2552 | |
| 2553 | * _careful to defend against unlocked access to
|
| 2554 | * ->shrink_list in shmem_unused_huge_shrink()
|
| 2555 |
|
| 2556 | if (list_empty_careful(&info->shrinklist)) {
|
| 2557 | list_add_tail(&info->shrinklist,
|
| 2558 | &sbinfo->shrinklist);
|
| 2559 | sbinfo->shrinklist_len++;
|
| 2560 | }
|
| 2561 | spin_unlock(&sbinfo->shrinklist_lock);
|
| 2562 | }
|
| 2563 |
|
| 2564 | if (sgp == SGP_WRITE)
|
| 2565 | folio_set_referenced(folio);
|
| 2566 | |
| 2567 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
|
| 2568 |
|
| 2569 | if (sgp == SGP_FALLOC)
|
| 2570 | sgp = SGP_WRITE;
|
| 2571 | clear:
|
| 2572 | |
| 2573 | * Let SGP_WRITE caller clear ends if write does not fill folio;
|
| 2574 | * but SGP_FALLOC on a folio fallocated earlier must initialize
|
| 2575 | * it now, lest undo on failure cancel our earlier guarantee.
|
| 2576 |
|
| 2577 | if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
|
| 2578 | long i, n = folio_nr_pages(folio);
|
| 2579 |
|
| 2580 | for (i = 0; i < n; i++)
|
| 2581 | clear_highpage(folio_page(folio, i));
|
| 2582 | flush_dcache_folio(folio);
|
| 2583 | folio_mark_uptodate(folio);
|
| 2584 | }
|
| 2585 |
|
| 2586 |
|
| 2587 | if (sgp <= SGP_CACHE &&
|
| 2588 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
|
| 2589 | error = -EINVAL;
|
| 2590 | goto unlock;
|
| 2591 | }
|
| 2592 | out:
|
| 2593 | *foliop = folio;
|
| 2594 | return 0;
|
| 2595 |
|
| 2596 | |
| 2597 | * Error recovery.
|
| 2598 |
|
| 2599 | unlock:
|
| 2600 | if (alloced)
|
| 2601 | filemap_remove_folio(folio);
|
| 2602 | shmem_recalc_inode(inode, 0, 0);
|
| 2603 | if (folio) {
|
| 2604 | folio_unlock(folio);
|
| 2605 | folio_put(folio);
|
| 2606 | }
|
| 2607 | return error;
|
| 2608 | }
|
| 2609 |
|
| 2610 | |
| 2611 | * shmem_get_folio - find, and lock a shmem folio.
|
| 2612 | * @inode: inode to search
|
| 2613 | * @index: the page index.
|
| 2614 | * @write_end: end of a write, could extend inode size
|
| 2615 | * @foliop: pointer to the folio if found
|
| 2616 | * @sgp: SGP_* flags to control behavior
|
| 2617 | *
|
| 2618 | * Looks up the page cache entry at @inode & @index. If a folio is
|
| 2619 | * present, it is returned locked with an increased refcount.
|
| 2620 | *
|
| 2621 | * If the caller modifies data in the folio, it must call folio_mark_dirty()
|
| 2622 | * before unlocking the folio to ensure that the folio is not reclaimed.
|
| 2623 | * There is no need to reserve space before calling folio_mark_dirty().
|
| 2624 | *
|
| 2625 | * When no folio is found, the behavior depends on @sgp:
|
| 2626 | * - for SGP_READ, *@foliop is %NULL and 0 is returned
|
| 2627 | * - for SGP_NOALLOC, *@foliop is %NULL and -ENOENT is returned
|
| 2628 | * - for all other flags a new folio is allocated, inserted into the
|
| 2629 | * page cache and returned locked in @foliop.
|
| 2630 | *
|
| 2631 | * Context: May sleep.
|
| 2632 | * Return: 0 if successful, else a negative error code.
|
| 2633 |
|
| 2634 | int shmem_get_folio(struct inode *inode, pgoff_t index, loff_t write_end,
|
| 2635 | struct folio **foliop, enum sgp_type sgp)
|
| 2636 | {
|
| 2637 | return shmem_get_folio_gfp(inode, index, write_end, foliop, sgp,
|
| 2638 | mapping_gfp_mask(inode->i_mapping), NULL, NULL);
|
| 2639 | }
|
| 2640 | EXPORT_SYMBOL_GPL(shmem_get_folio);
|
| 2641 |
|
| 2642 | |
| 2643 | * This is like autoremove_wake_function, but it removes the wait queue
|
| 2644 | * entry unconditionally - even if something else had already woken the
|
| 2645 | * target.
|
| 2646 |
|
| 2647 | static int synchronous_wake_function(wait_queue_entry_t *wait,
|
| 2648 | unsigned int mode, int sync, void *key)
|
| 2649 | {
|
| 2650 | int ret = default_wake_function(wait, mode, sync, key);
|
| 2651 | list_del_init(&wait->entry);
|
| 2652 | return ret;
|
| 2653 | }
|
| 2654 |
|
| 2655 | |
| 2656 | * Trinity finds that probing a hole which tmpfs is punching can
|
| 2657 | * prevent the hole-punch from ever completing: which in turn
|
| 2658 | * locks writers out with its hold on i_rwsem. So refrain from
|
| 2659 | * faulting pages into the hole while it's being punched. Although
|
| 2660 | * shmem_undo_range() does remove the additions, it may be unable to
|
| 2661 | * keep up, as each new page needs its own unmap_mapping_range() call,
|
| 2662 | * and the i_mmap tree grows ever slower to scan if new vmas are added.
|
| 2663 | *
|
| 2664 | * It does not matter if we sometimes reach this check just before the
|
| 2665 | * hole-punch begins, so that one fault then races with the punch:
|
| 2666 | * we just need to make racing faults a rare case.
|
| 2667 | *
|
| 2668 | * The implementation below would be much simpler if we just used a
|
| 2669 | * standard mutex or completion: but we cannot take i_rwsem in fault,
|
| 2670 | * and bloating every shmem inode for this unlikely case would be sad.
|
| 2671 |
|
| 2672 | static vm_fault_t shmem_falloc_wait(struct vm_fault *vmf, struct inode *inode)
|
| 2673 | {
|
| 2674 | struct shmem_falloc *shmem_falloc;
|
| 2675 | struct file *fpin = NULL;
|
| 2676 | vm_fault_t ret = 0;
|
| 2677 |
|
| 2678 | spin_lock(&inode->i_lock);
|
| 2679 | shmem_falloc = inode->i_private;
|
| 2680 | if (shmem_falloc &&
|
| 2681 | shmem_falloc->waitq &&
|
| 2682 | vmf->pgoff >= shmem_falloc->start &&
|
| 2683 | vmf->pgoff < shmem_falloc->next) {
|
| 2684 | wait_queue_head_t *shmem_falloc_waitq;
|
| 2685 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
|
| 2686 |
|
| 2687 | ret = VM_FAULT_NOPAGE;
|
| 2688 | fpin = maybe_unlock_mmap_for_io(vmf, NULL);
|
| 2689 | shmem_falloc_waitq = shmem_falloc->waitq;
|
| 2690 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
|
| 2691 | TASK_UNINTERRUPTIBLE);
|
| 2692 | spin_unlock(&inode->i_lock);
|
| 2693 | schedule();
|
| 2694 |
|
| 2695 | |
| 2696 | * shmem_falloc_waitq points into the shmem_fallocate()
|
| 2697 | * stack of the hole-punching task: shmem_falloc_waitq
|
| 2698 | * is usually invalid by the time we reach here, but
|
| 2699 | * finish_wait() does not dereference it in that case;
|
| 2700 | * though i_lock needed lest racing with wake_up_all().
|
| 2701 |
|
| 2702 | spin_lock(&inode->i_lock);
|
| 2703 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
|
| 2704 | }
|
| 2705 | spin_unlock(&inode->i_lock);
|
| 2706 | if (fpin) {
|
| 2707 | fput(fpin);
|
| 2708 | ret = VM_FAULT_RETRY;
|
| 2709 | }
|
| 2710 | return ret;
|
| 2711 | }
|
| 2712 |
|
| 2713 | static vm_fault_t shmem_fault(struct vm_fault *vmf)
|
| 2714 | {
|
| 2715 | struct inode *inode = file_inode(vmf->vma->vm_file);
|
| 2716 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
|
| 2717 | struct folio *folio = NULL;
|
| 2718 | vm_fault_t ret = 0;
|
| 2719 | int err;
|
| 2720 |
|
| 2721 | |
| 2722 | * Trinity finds that probing a hole which tmpfs is punching can
|
| 2723 | * prevent the hole-punch from ever completing: noted in i_private.
|
| 2724 |
|
| 2725 | if (unlikely(READ_ONCE(inode->i_private))) {
|
| 2726 | ret = shmem_falloc_wait(vmf, inode);
|
| 2727 | if (ret)
|
| 2728 | return ret;
|
| 2729 | }
|
| 2730 |
|
| 2731 | WARN_ON_ONCE(vmf->page != NULL);
|
| 2732 | err = shmem_get_folio_gfp(inode, vmf->pgoff, 0, &folio, SGP_CACHE,
|
| 2733 | gfp, vmf, &ret);
|
| 2734 | if (err)
|
| 2735 | return vmf_error(err);
|
| 2736 | if (folio) {
|
| 2737 | vmf->page = folio_file_page(folio, vmf->pgoff);
|
| 2738 | ret |= VM_FAULT_LOCKED;
|
| 2739 | }
|
| 2740 | return ret;
|
| 2741 | }
|
| 2742 |
|
| 2743 | unsigned long shmem_get_unmapped_area(struct file *file,
|
| 2744 | unsigned long uaddr, unsigned long len,
|
| 2745 | unsigned long pgoff, unsigned long flags)
|
| 2746 | {
|
| 2747 | unsigned long addr;
|
| 2748 | unsigned long offset;
|
| 2749 | unsigned long inflated_len;
|
| 2750 | unsigned long inflated_addr;
|
| 2751 | unsigned long inflated_offset;
|
| 2752 | unsigned long hpage_size;
|
| 2753 |
|
| 2754 | if (len > TASK_SIZE)
|
| 2755 | return -ENOMEM;
|
| 2756 |
|
| 2757 | addr = mm_get_unmapped_area(file, uaddr, len, pgoff, flags);
|
| 2758 |
|
| 2759 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
|
| 2760 | return addr;
|
| 2761 | if (IS_ERR_VALUE(addr))
|
| 2762 | return addr;
|
| 2763 | if (addr & ~PAGE_MASK)
|
| 2764 | return addr;
|
| 2765 | if (addr > TASK_SIZE - len)
|
| 2766 | return addr;
|
| 2767 |
|
| 2768 | if (shmem_huge == SHMEM_HUGE_DENY)
|
| 2769 | return addr;
|
| 2770 | if (flags & MAP_FIXED)
|
| 2771 | return addr;
|
| 2772 | |
| 2773 | * Our priority is to support MAP_SHARED mapped hugely;
|
| 2774 | * and support MAP_PRIVATE mapped hugely too, until it is COWed.
|
| 2775 | * But if caller specified an address hint and we allocated area there
|
| 2776 | * successfully, respect that as before.
|
| 2777 |
|
| 2778 | if (uaddr == addr)
|
| 2779 | return addr;
|
| 2780 |
|
| 2781 | hpage_size = HPAGE_PMD_SIZE;
|
| 2782 | if (shmem_huge != SHMEM_HUGE_FORCE) {
|
| 2783 | struct super_block *sb;
|
| 2784 | unsigned long __maybe_unused hpage_orders;
|
| 2785 | int order = 0;
|
| 2786 |
|
| 2787 | if (file) {
|
| 2788 | VM_BUG_ON(file->f_op != &shmem_file_operations);
|
| 2789 | sb = file_inode(file)->i_sb;
|
| 2790 | } else {
|
| 2791 | |
| 2792 | * Called directly from mm/mmap.c, or drivers/char/mem.c
|
| 2793 | * for "/dev/zero", to create a shared anonymous object.
|
| 2794 |
|
| 2795 | if (IS_ERR(shm_mnt))
|
| 2796 | return addr;
|
| 2797 | sb = shm_mnt->mnt_sb;
|
| 2798 |
|
| 2799 | |
| 2800 | * Find the highest mTHP order used for anonymous shmem to
|
| 2801 | * provide a suitable alignment address.
|
| 2802 |
|
| 2803 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 2804 | hpage_orders = READ_ONCE(huge_shmem_orders_always);
|
| 2805 | hpage_orders |= READ_ONCE(huge_shmem_orders_within_size);
|
| 2806 | hpage_orders |= READ_ONCE(huge_shmem_orders_madvise);
|
| 2807 | if (SHMEM_SB(sb)->huge != SHMEM_HUGE_NEVER)
|
| 2808 | hpage_orders |= READ_ONCE(huge_shmem_orders_inherit);
|
| 2809 |
|
| 2810 | if (hpage_orders > 0) {
|
| 2811 | order = highest_order(hpage_orders);
|
| 2812 | hpage_size = PAGE_SIZE << order;
|
| 2813 | }
|
| 2814 | #endif
|
| 2815 | }
|
| 2816 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER && !order)
|
| 2817 | return addr;
|
| 2818 | }
|
| 2819 |
|
| 2820 | if (len < hpage_size)
|
| 2821 | return addr;
|
| 2822 |
|
| 2823 | offset = (pgoff << PAGE_SHIFT) & (hpage_size - 1);
|
| 2824 | if (offset && offset + len < 2 * hpage_size)
|
| 2825 | return addr;
|
| 2826 | if ((addr & (hpage_size - 1)) == offset)
|
| 2827 | return addr;
|
| 2828 |
|
| 2829 | inflated_len = len + hpage_size - PAGE_SIZE;
|
| 2830 | if (inflated_len > TASK_SIZE)
|
| 2831 | return addr;
|
| 2832 | if (inflated_len < len)
|
| 2833 | return addr;
|
| 2834 |
|
| 2835 | inflated_addr = mm_get_unmapped_area(NULL, uaddr, inflated_len, 0, flags);
|
| 2836 | if (IS_ERR_VALUE(inflated_addr))
|
| 2837 | return addr;
|
| 2838 | if (inflated_addr & ~PAGE_MASK)
|
| 2839 | return addr;
|
| 2840 |
|
| 2841 | inflated_offset = inflated_addr & (hpage_size - 1);
|
| 2842 | inflated_addr += offset - inflated_offset;
|
| 2843 | if (inflated_offset > offset)
|
| 2844 | inflated_addr += hpage_size;
|
| 2845 |
|
| 2846 | if (inflated_addr > TASK_SIZE - len)
|
| 2847 | return addr;
|
| 2848 | return inflated_addr;
|
| 2849 | }
|
| 2850 |
|
| 2851 | #ifdef CONFIG_NUMA
|
| 2852 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
|
| 2853 | {
|
| 2854 | struct inode *inode = file_inode(vma->vm_file);
|
| 2855 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
|
| 2856 | }
|
| 2857 |
|
| 2858 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
|
| 2859 | unsigned long addr, pgoff_t *ilx)
|
| 2860 | {
|
| 2861 | struct inode *inode = file_inode(vma->vm_file);
|
| 2862 | pgoff_t index;
|
| 2863 |
|
| 2864 | |
| 2865 | * Bias interleave by inode number to distribute better across nodes;
|
| 2866 | * but this interface is independent of which page order is used, so
|
| 2867 | * supplies only that bias, letting caller apply the offset (adjusted
|
| 2868 | * by page order, as in shmem_get_pgoff_policy() and get_vma_policy()).
|
| 2869 |
|
| 2870 | *ilx = inode->i_ino;
|
| 2871 | index = linear_page_index(vma, addr);
|
| 2872 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
|
| 2873 | }
|
| 2874 |
|
| 2875 | static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
|
| 2876 | pgoff_t index, unsigned int order, pgoff_t *ilx)
|
| 2877 | {
|
| 2878 | struct mempolicy *mpol;
|
| 2879 |
|
| 2880 |
|
| 2881 | *ilx = info->vfs_inode.i_ino + (index >> order);
|
| 2882 |
|
| 2883 | mpol = mpol_shared_policy_lookup(&info->policy, index);
|
| 2884 | return mpol ? mpol : get_task_policy(current);
|
| 2885 | }
|
| 2886 | #else
|
| 2887 | static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
|
| 2888 | pgoff_t index, unsigned int order, pgoff_t *ilx)
|
| 2889 | {
|
| 2890 | *ilx = 0;
|
| 2891 | return NULL;
|
| 2892 | }
|
| 2893 | #endif
|
| 2894 |
|
| 2895 | int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
|
| 2896 | {
|
| 2897 | struct inode *inode = file_inode(file);
|
| 2898 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 2899 | int retval = -ENOMEM;
|
| 2900 |
|
| 2901 | |
| 2902 | * What serializes the accesses to info->flags?
|
| 2903 | * ipc_lock_object() when called from shmctl_do_lock(),
|
| 2904 | * no serialization needed when called from shm_destroy().
|
| 2905 |
|
| 2906 | if (lock && !(info->flags & SHMEM_F_LOCKED)) {
|
| 2907 | if (!user_shm_lock(inode->i_size, ucounts))
|
| 2908 | goto out_nomem;
|
| 2909 | info->flags |= SHMEM_F_LOCKED;
|
| 2910 | mapping_set_unevictable(file->f_mapping);
|
| 2911 | }
|
| 2912 | if (!lock && (info->flags & SHMEM_F_LOCKED) && ucounts) {
|
| 2913 | user_shm_unlock(inode->i_size, ucounts);
|
| 2914 | info->flags &= ~SHMEM_F_LOCKED;
|
| 2915 | mapping_clear_unevictable(file->f_mapping);
|
| 2916 | }
|
| 2917 | retval = 0;
|
| 2918 |
|
| 2919 | out_nomem:
|
| 2920 | return retval;
|
| 2921 | }
|
| 2922 |
|
| 2923 | static int shmem_mmap_prepare(struct vm_area_desc *desc)
|
| 2924 | {
|
| 2925 | struct file *file = desc->file;
|
| 2926 | struct inode *inode = file_inode(file);
|
| 2927 |
|
| 2928 | file_accessed(file);
|
| 2929 |
|
| 2930 | if (inode->i_nlink)
|
| 2931 | desc->vm_ops = &shmem_vm_ops;
|
| 2932 | else
|
| 2933 | desc->vm_ops = &shmem_anon_vm_ops;
|
| 2934 | return 0;
|
| 2935 | }
|
| 2936 |
|
| 2937 | static int shmem_file_open(struct inode *inode, struct file *file)
|
| 2938 | {
|
| 2939 | file->f_mode |= FMODE_CAN_ODIRECT;
|
| 2940 | return generic_file_open(inode, file);
|
| 2941 | }
|
| 2942 |
|
| 2943 | #ifdef CONFIG_TMPFS_XATTR
|
| 2944 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
|
| 2945 |
|
| 2946 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 2947 | |
| 2948 | * shmem_inode_casefold_flags - Deal with casefold file attribute flag
|
| 2949 | *
|
| 2950 | * The casefold file attribute needs some special checks. I can just be added to
|
| 2951 | * an empty dir, and can't be removed from a non-empty dir.
|
| 2952 |
|
| 2953 | static int shmem_inode_casefold_flags(struct inode *inode, unsigned int fsflags,
|
| 2954 | struct dentry *dentry, unsigned int *i_flags)
|
| 2955 | {
|
| 2956 | unsigned int old = inode->i_flags;
|
| 2957 | struct super_block *sb = inode->i_sb;
|
| 2958 |
|
| 2959 | if (fsflags & FS_CASEFOLD_FL) {
|
| 2960 | if (!(old & S_CASEFOLD)) {
|
| 2961 | if (!sb->s_encoding)
|
| 2962 | return -EOPNOTSUPP;
|
| 2963 |
|
| 2964 | if (!S_ISDIR(inode->i_mode))
|
| 2965 | return -ENOTDIR;
|
| 2966 |
|
| 2967 | if (dentry && !simple_empty(dentry))
|
| 2968 | return -ENOTEMPTY;
|
| 2969 | }
|
| 2970 |
|
| 2971 | *i_flags = *i_flags | S_CASEFOLD;
|
| 2972 | } else if (old & S_CASEFOLD) {
|
| 2973 | if (dentry && !simple_empty(dentry))
|
| 2974 | return -ENOTEMPTY;
|
| 2975 | }
|
| 2976 |
|
| 2977 | return 0;
|
| 2978 | }
|
| 2979 | #else
|
| 2980 | static int shmem_inode_casefold_flags(struct inode *inode, unsigned int fsflags,
|
| 2981 | struct dentry *dentry, unsigned int *i_flags)
|
| 2982 | {
|
| 2983 | if (fsflags & FS_CASEFOLD_FL)
|
| 2984 | return -EOPNOTSUPP;
|
| 2985 |
|
| 2986 | return 0;
|
| 2987 | }
|
| 2988 | #endif
|
| 2989 |
|
| 2990 | |
| 2991 | * chattr's fsflags are unrelated to extended attributes,
|
| 2992 | * but tmpfs has chosen to enable them under the same config option.
|
| 2993 |
|
| 2994 | static int shmem_set_inode_flags(struct inode *inode, unsigned int fsflags, struct dentry *dentry)
|
| 2995 | {
|
| 2996 | unsigned int i_flags = 0;
|
| 2997 | int ret;
|
| 2998 |
|
| 2999 | ret = shmem_inode_casefold_flags(inode, fsflags, dentry, &i_flags);
|
| 3000 | if (ret)
|
| 3001 | return ret;
|
| 3002 |
|
| 3003 | if (fsflags & FS_NOATIME_FL)
|
| 3004 | i_flags |= S_NOATIME;
|
| 3005 | if (fsflags & FS_APPEND_FL)
|
| 3006 | i_flags |= S_APPEND;
|
| 3007 | if (fsflags & FS_IMMUTABLE_FL)
|
| 3008 | i_flags |= S_IMMUTABLE;
|
| 3009 | |
| 3010 | * But FS_NODUMP_FL does not require any action in i_flags.
|
| 3011 |
|
| 3012 | inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE | S_CASEFOLD);
|
| 3013 |
|
| 3014 | return 0;
|
| 3015 | }
|
| 3016 | #else
|
| 3017 | static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags, struct dentry *dentry)
|
| 3018 | {
|
| 3019 | }
|
| 3020 | #define shmem_initxattrs NULL
|
| 3021 | #endif
|
| 3022 |
|
| 3023 | static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode)
|
| 3024 | {
|
| 3025 | return &SHMEM_I(inode)->dir_offsets;
|
| 3026 | }
|
| 3027 |
|
| 3028 | static struct inode *__shmem_get_inode(const struct mnt_idmap *idmap,
|
| 3029 | struct super_block *sb,
|
| 3030 | struct inode *dir, umode_t mode,
|
| 3031 | dev_t dev, vma_flags_t flags)
|
| 3032 | {
|
| 3033 | struct inode *inode;
|
| 3034 | struct shmem_inode_info *info;
|
| 3035 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
|
| 3036 | ino_t ino;
|
| 3037 | int err;
|
| 3038 |
|
| 3039 | err = shmem_reserve_inode(sb, &ino);
|
| 3040 | if (err)
|
| 3041 | return ERR_PTR(err);
|
| 3042 |
|
| 3043 | inode = new_inode(sb);
|
| 3044 | if (!inode) {
|
| 3045 | shmem_free_inode(sb, 0);
|
| 3046 | return ERR_PTR(-ENOSPC);
|
| 3047 | }
|
| 3048 |
|
| 3049 | inode->i_ino = ino;
|
| 3050 | inode_init_owner(idmap, inode, dir, mode);
|
| 3051 | inode->i_blocks = 0;
|
| 3052 | simple_inode_init_ts(inode);
|
| 3053 | inode->i_generation = get_random_u32();
|
| 3054 | info = SHMEM_I(inode);
|
| 3055 | memset(info, 0, (char *)inode - (char *)info);
|
| 3056 | INIT_LIST_HEAD_RCU(&info->xattrs);
|
| 3057 | spin_lock_init(&info->lock);
|
| 3058 | atomic_set(&info->stop_eviction, 0);
|
| 3059 | info->seals = F_SEAL_SEAL;
|
| 3060 | info->flags = vma_flags_test(&flags, VMA_NORESERVE_BIT)
|
| 3061 | ? SHMEM_F_NORESERVE : 0;
|
| 3062 | info->i_crtime = inode_get_mtime(inode);
|
| 3063 | info->fsflags = (dir == NULL) ? 0 :
|
| 3064 | SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
|
| 3065 | if (info->fsflags)
|
| 3066 | shmem_set_inode_flags(inode, info->fsflags, NULL);
|
| 3067 | INIT_LIST_HEAD(&info->shrinklist);
|
| 3068 | INIT_LIST_HEAD(&info->swaplist);
|
| 3069 | cache_no_acl(inode);
|
| 3070 | if (sbinfo->noswap)
|
| 3071 | mapping_set_unevictable(inode->i_mapping);
|
| 3072 | mapping_set_large_folios(inode->i_mapping);
|
| 3073 |
|
| 3074 | switch (mode & S_IFMT) {
|
| 3075 | default:
|
| 3076 | inode->i_op = &shmem_special_inode_operations;
|
| 3077 | init_special_inode(inode, mode, dev);
|
| 3078 | break;
|
| 3079 | case S_IFREG:
|
| 3080 | inode->i_mapping->a_ops = &shmem_aops;
|
| 3081 | inode->i_op = &shmem_inode_operations;
|
| 3082 | inode->i_fop = &shmem_file_operations;
|
| 3083 | mpol_shared_policy_init(&info->policy,
|
| 3084 | shmem_get_sbmpol(sbinfo));
|
| 3085 | break;
|
| 3086 | case S_IFDIR:
|
| 3087 | inc_nlink(inode);
|
| 3088 |
|
| 3089 | inode->i_size = 2 * BOGO_DIRENT_SIZE;
|
| 3090 | inode->i_op = &shmem_dir_inode_operations;
|
| 3091 | inode->i_fop = &simple_offset_dir_operations;
|
| 3092 | simple_offset_init(shmem_get_offset_ctx(inode));
|
| 3093 | break;
|
| 3094 | case S_IFLNK:
|
| 3095 | |
| 3096 | * Must not load anything in the rbtree,
|
| 3097 | * mpol_free_shared_policy will not be called.
|
| 3098 |
|
| 3099 | mpol_shared_policy_init(&info->policy, NULL);
|
| 3100 | break;
|
| 3101 | }
|
| 3102 |
|
| 3103 | lockdep_annotate_inode_mutex_key(inode);
|
| 3104 | return inode;
|
| 3105 | }
|
| 3106 |
|
| 3107 | #ifdef CONFIG_TMPFS_QUOTA
|
| 3108 | static struct inode *shmem_get_inode(const struct mnt_idmap *idmap,
|
| 3109 | struct super_block *sb, struct inode *dir,
|
| 3110 | umode_t mode, dev_t dev, vma_flags_t flags)
|
| 3111 | {
|
| 3112 | int err;
|
| 3113 | struct inode *inode;
|
| 3114 |
|
| 3115 | inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
|
| 3116 | if (IS_ERR(inode))
|
| 3117 | return inode;
|
| 3118 |
|
| 3119 | err = dquot_initialize(inode);
|
| 3120 | if (err)
|
| 3121 | goto errout;
|
| 3122 |
|
| 3123 | err = dquot_alloc_inode(inode);
|
| 3124 | if (err) {
|
| 3125 | dquot_drop(inode);
|
| 3126 | goto errout;
|
| 3127 | }
|
| 3128 | return inode;
|
| 3129 |
|
| 3130 | errout:
|
| 3131 | inode->i_flags |= S_NOQUOTA;
|
| 3132 | iput(inode);
|
| 3133 | return ERR_PTR(err);
|
| 3134 | }
|
| 3135 | #else
|
| 3136 | static struct inode *shmem_get_inode(const struct mnt_idmap *idmap,
|
| 3137 | struct super_block *sb, struct inode *dir,
|
| 3138 | umode_t mode, dev_t dev, vma_flags_t flags)
|
| 3139 | {
|
| 3140 | return __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
|
| 3141 | }
|
| 3142 | #endif
|
| 3143 |
|
| 3144 | #ifdef CONFIG_USERFAULTFD
|
| 3145 | static struct folio *shmem_mfill_folio_alloc(struct vm_area_struct *vma,
|
| 3146 | unsigned long addr)
|
| 3147 | {
|
| 3148 | struct inode *inode = file_inode(vma->vm_file);
|
| 3149 | struct address_space *mapping = inode->i_mapping;
|
| 3150 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 3151 | pgoff_t pgoff = linear_page_index(vma, addr);
|
| 3152 | gfp_t gfp = mapping_gfp_mask(mapping);
|
| 3153 | struct folio *folio;
|
| 3154 |
|
| 3155 | if (unlikely(pgoff >= DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE)))
|
| 3156 | return NULL;
|
| 3157 |
|
| 3158 | folio = shmem_alloc_folio(gfp, 0, info, pgoff);
|
| 3159 | if (!folio)
|
| 3160 | return NULL;
|
| 3161 |
|
| 3162 | if (mem_cgroup_charge(folio, vma->vm_mm, GFP_KERNEL)) {
|
| 3163 | folio_put(folio);
|
| 3164 | return NULL;
|
| 3165 | }
|
| 3166 |
|
| 3167 | return folio;
|
| 3168 | }
|
| 3169 |
|
| 3170 | static int shmem_mfill_filemap_add(struct folio *folio,
|
| 3171 | struct vm_area_struct *vma,
|
| 3172 | unsigned long addr)
|
| 3173 | {
|
| 3174 | struct inode *inode = file_inode(vma->vm_file);
|
| 3175 | struct address_space *mapping = inode->i_mapping;
|
| 3176 | pgoff_t pgoff = linear_page_index(vma, addr);
|
| 3177 | gfp_t gfp = mapping_gfp_mask(mapping);
|
| 3178 | int err;
|
| 3179 |
|
| 3180 | __folio_set_locked(folio);
|
| 3181 | __folio_set_swapbacked(folio);
|
| 3182 |
|
| 3183 | err = shmem_add_to_page_cache(folio, mapping, pgoff, NULL, gfp);
|
| 3184 | if (err)
|
| 3185 | goto err_unlock;
|
| 3186 |
|
| 3187 | if (shmem_inode_acct_blocks(inode, 1)) {
|
| 3188 | err = -ENOMEM;
|
| 3189 | goto err_delete_from_cache;
|
| 3190 | }
|
| 3191 |
|
| 3192 | folio_add_lru(folio);
|
| 3193 | shmem_recalc_inode(inode, 1, 0);
|
| 3194 |
|
| 3195 | return 0;
|
| 3196 |
|
| 3197 | err_delete_from_cache:
|
| 3198 | filemap_remove_folio(folio);
|
| 3199 | err_unlock:
|
| 3200 | folio_unlock(folio);
|
| 3201 | return err;
|
| 3202 | }
|
| 3203 |
|
| 3204 | static void shmem_mfill_filemap_remove(struct folio *folio,
|
| 3205 | struct vm_area_struct *vma)
|
| 3206 | {
|
| 3207 | struct inode *inode = file_inode(vma->vm_file);
|
| 3208 |
|
| 3209 | filemap_remove_folio(folio);
|
| 3210 | shmem_recalc_inode(inode, 0, 0);
|
| 3211 | folio_unlock(folio);
|
| 3212 | }
|
| 3213 |
|
| 3214 | static struct folio *shmem_get_folio_noalloc(struct inode *inode, pgoff_t pgoff)
|
| 3215 | {
|
| 3216 | struct folio *folio;
|
| 3217 | int err;
|
| 3218 |
|
| 3219 | err = shmem_get_folio(inode, pgoff, 0, &folio, SGP_NOALLOC);
|
| 3220 | if (err)
|
| 3221 | return ERR_PTR(err);
|
| 3222 |
|
| 3223 | return folio;
|
| 3224 | }
|
| 3225 |
|
| 3226 | static bool shmem_can_userfault(struct vm_area_struct *vma, vm_flags_t vm_flags)
|
| 3227 | {
|
| 3228 | return true;
|
| 3229 | }
|
| 3230 |
|
| 3231 | static const struct vm_uffd_ops shmem_uffd_ops = {
|
| 3232 | .can_userfault = shmem_can_userfault,
|
| 3233 | .get_folio_noalloc = shmem_get_folio_noalloc,
|
| 3234 | .alloc_folio = shmem_mfill_folio_alloc,
|
| 3235 | .filemap_add = shmem_mfill_filemap_add,
|
| 3236 | .filemap_remove = shmem_mfill_filemap_remove,
|
| 3237 | };
|
| 3238 | #endif
|
| 3239 |
|
| 3240 | #ifdef CONFIG_TMPFS
|
| 3241 | static const struct inode_operations shmem_symlink_inode_operations;
|
| 3242 | static const struct inode_operations shmem_short_symlink_operations;
|
| 3243 |
|
| 3244 | static int
|
| 3245 | shmem_write_begin(const struct kiocb *iocb, struct address_space *mapping,
|
| 3246 | loff_t pos, unsigned len,
|
| 3247 | struct folio **foliop, void **fsdata)
|
| 3248 | {
|
| 3249 | struct inode *inode = mapping->host;
|
| 3250 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 3251 | pgoff_t index = pos >> PAGE_SHIFT;
|
| 3252 | struct folio *folio;
|
| 3253 | int ret = 0;
|
| 3254 |
|
| 3255 |
|
| 3256 | if (unlikely(info->seals & (F_SEAL_GROW |
|
| 3257 | F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
|
| 3258 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
|
| 3259 | return -EPERM;
|
| 3260 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
|
| 3261 | return -EPERM;
|
| 3262 | }
|
| 3263 |
|
| 3264 | if (unlikely((info->flags & SHMEM_F_MAPPING_FROZEN) &&
|
| 3265 | pos + len > inode->i_size))
|
| 3266 | return -EPERM;
|
| 3267 |
|
| 3268 | ret = shmem_get_folio(inode, index, pos + len, &folio, SGP_WRITE);
|
| 3269 | if (ret)
|
| 3270 | return ret;
|
| 3271 |
|
| 3272 | if (folio_contain_hwpoisoned_page(folio)) {
|
| 3273 | folio_unlock(folio);
|
| 3274 | folio_put(folio);
|
| 3275 | return -EIO;
|
| 3276 | }
|
| 3277 |
|
| 3278 | *foliop = folio;
|
| 3279 | return 0;
|
| 3280 | }
|
| 3281 |
|
| 3282 | static int
|
| 3283 | shmem_write_end(const struct kiocb *iocb, struct address_space *mapping,
|
| 3284 | loff_t pos, unsigned len, unsigned copied,
|
| 3285 | struct folio *folio, void *fsdata)
|
| 3286 | {
|
| 3287 | struct inode *inode = mapping->host;
|
| 3288 |
|
| 3289 | if (pos + copied > inode->i_size)
|
| 3290 | i_size_write(inode, pos + copied);
|
| 3291 |
|
| 3292 | if (!folio_test_uptodate(folio)) {
|
| 3293 | if (copied < folio_size(folio)) {
|
| 3294 | size_t from = offset_in_folio(folio, pos);
|
| 3295 | folio_zero_segments(folio, 0, from,
|
| 3296 | from + copied, folio_size(folio));
|
| 3297 | }
|
| 3298 | folio_mark_uptodate(folio);
|
| 3299 | }
|
| 3300 | folio_mark_dirty(folio);
|
| 3301 | folio_unlock(folio);
|
| 3302 | folio_put(folio);
|
| 3303 |
|
| 3304 | return copied;
|
| 3305 | }
|
| 3306 |
|
| 3307 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
|
| 3308 | {
|
| 3309 | struct file *file = iocb->ki_filp;
|
| 3310 | struct inode *inode = file_inode(file);
|
| 3311 | struct address_space *mapping = inode->i_mapping;
|
| 3312 | pgoff_t index;
|
| 3313 | unsigned long offset;
|
| 3314 | int error = 0;
|
| 3315 | ssize_t retval = 0;
|
| 3316 |
|
| 3317 | for (;;) {
|
| 3318 | struct folio *folio = NULL;
|
| 3319 | struct page *page = NULL;
|
| 3320 | unsigned long nr, ret;
|
| 3321 | loff_t end_offset, i_size = i_size_read(inode);
|
| 3322 | bool fallback_page_copy = false;
|
| 3323 | size_t fsize;
|
| 3324 |
|
| 3325 | if (unlikely(iocb->ki_pos >= i_size))
|
| 3326 | break;
|
| 3327 |
|
| 3328 | index = iocb->ki_pos >> PAGE_SHIFT;
|
| 3329 | error = shmem_get_folio(inode, index, 0, &folio, SGP_READ);
|
| 3330 | if (error) {
|
| 3331 | if (error == -EINVAL)
|
| 3332 | error = 0;
|
| 3333 | break;
|
| 3334 | }
|
| 3335 | if (folio) {
|
| 3336 | folio_unlock(folio);
|
| 3337 |
|
| 3338 | page = folio_file_page(folio, index);
|
| 3339 | if (PageHWPoison(page)) {
|
| 3340 | folio_put(folio);
|
| 3341 | error = -EIO;
|
| 3342 | break;
|
| 3343 | }
|
| 3344 |
|
| 3345 | if (folio_test_large(folio) &&
|
| 3346 | folio_test_has_hwpoisoned(folio))
|
| 3347 | fallback_page_copy = true;
|
| 3348 | }
|
| 3349 |
|
| 3350 | |
| 3351 | * We must evaluate after, since reads (unlike writes)
|
| 3352 | * are called without i_rwsem protection against truncate
|
| 3353 |
|
| 3354 | i_size = i_size_read(inode);
|
| 3355 | if (unlikely(iocb->ki_pos >= i_size)) {
|
| 3356 | if (folio)
|
| 3357 | folio_put(folio);
|
| 3358 | break;
|
| 3359 | }
|
| 3360 | end_offset = min_t(loff_t, i_size, iocb->ki_pos + to->count);
|
| 3361 | if (folio && likely(!fallback_page_copy))
|
| 3362 | fsize = folio_size(folio);
|
| 3363 | else
|
| 3364 | fsize = PAGE_SIZE;
|
| 3365 | offset = iocb->ki_pos & (fsize - 1);
|
| 3366 | nr = min_t(loff_t, end_offset - iocb->ki_pos, fsize - offset);
|
| 3367 |
|
| 3368 | if (folio) {
|
| 3369 | |
| 3370 | * If users can be writing to this page using arbitrary
|
| 3371 | * virtual addresses, take care about potential aliasing
|
| 3372 | * before reading the page on the kernel side.
|
| 3373 |
|
| 3374 | if (mapping_writably_mapped(mapping)) {
|
| 3375 | if (likely(!fallback_page_copy))
|
| 3376 | flush_dcache_folio(folio);
|
| 3377 | else
|
| 3378 | flush_dcache_page(page);
|
| 3379 | }
|
| 3380 |
|
| 3381 | |
| 3382 | * Mark the folio accessed if we read the beginning.
|
| 3383 |
|
| 3384 | if (!offset)
|
| 3385 | folio_mark_accessed(folio);
|
| 3386 | |
| 3387 | * Ok, we have the page, and it's up-to-date, so
|
| 3388 | * now we can copy it to user space...
|
| 3389 |
|
| 3390 | if (likely(!fallback_page_copy))
|
| 3391 | ret = copy_folio_to_iter(folio, offset, nr, to);
|
| 3392 | else
|
| 3393 | ret = copy_page_to_iter(page, offset, nr, to);
|
| 3394 | folio_put(folio);
|
| 3395 | } else if (user_backed_iter(to)) {
|
| 3396 | |
| 3397 | * Copy to user tends to be so well optimized, but
|
| 3398 | * clear_user() not so much, that it is noticeably
|
| 3399 | * faster to copy the zero page instead of clearing.
|
| 3400 |
|
| 3401 | ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
|
| 3402 | } else {
|
| 3403 | |
| 3404 | * But submitting the same page twice in a row to
|
| 3405 | * splice() - or others? - can result in confusion:
|
| 3406 | * so don't attempt that optimization on pipes etc.
|
| 3407 |
|
| 3408 | ret = iov_iter_zero(nr, to);
|
| 3409 | }
|
| 3410 |
|
| 3411 | retval += ret;
|
| 3412 | iocb->ki_pos += ret;
|
| 3413 |
|
| 3414 | if (!iov_iter_count(to))
|
| 3415 | break;
|
| 3416 | if (ret < nr) {
|
| 3417 | error = -EFAULT;
|
| 3418 | break;
|
| 3419 | }
|
| 3420 | cond_resched();
|
| 3421 | }
|
| 3422 |
|
| 3423 | file_accessed(file);
|
| 3424 | return retval ? retval : error;
|
| 3425 | }
|
| 3426 |
|
| 3427 | static ssize_t shmem_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
|
| 3428 | {
|
| 3429 | struct file *file = iocb->ki_filp;
|
| 3430 | struct inode *inode = file->f_mapping->host;
|
| 3431 | ssize_t ret;
|
| 3432 |
|
| 3433 | inode_lock(inode);
|
| 3434 | ret = generic_write_checks(iocb, from);
|
| 3435 | if (ret <= 0)
|
| 3436 | goto unlock;
|
| 3437 | ret = file_remove_privs(file);
|
| 3438 | if (ret)
|
| 3439 | goto unlock;
|
| 3440 | ret = file_update_time(file);
|
| 3441 | if (ret)
|
| 3442 | goto unlock;
|
| 3443 | ret = generic_perform_write(iocb, from);
|
| 3444 | unlock:
|
| 3445 | inode_unlock(inode);
|
| 3446 | return ret;
|
| 3447 | }
|
| 3448 |
|
| 3449 | static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
|
| 3450 | struct pipe_buffer *buf)
|
| 3451 | {
|
| 3452 | return true;
|
| 3453 | }
|
| 3454 |
|
| 3455 | static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
|
| 3456 | struct pipe_buffer *buf)
|
| 3457 | {
|
| 3458 | }
|
| 3459 |
|
| 3460 | static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
|
| 3461 | struct pipe_buffer *buf)
|
| 3462 | {
|
| 3463 | return false;
|
| 3464 | }
|
| 3465 |
|
| 3466 | static const struct pipe_buf_operations zero_pipe_buf_ops = {
|
| 3467 | .release = zero_pipe_buf_release,
|
| 3468 | .try_steal = zero_pipe_buf_try_steal,
|
| 3469 | .get = zero_pipe_buf_get,
|
| 3470 | };
|
| 3471 |
|
| 3472 | static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
|
| 3473 | loff_t fpos, size_t size)
|
| 3474 | {
|
| 3475 | size_t offset = fpos & ~PAGE_MASK;
|
| 3476 |
|
| 3477 | size = min_t(size_t, size, PAGE_SIZE - offset);
|
| 3478 |
|
| 3479 | if (!pipe_is_full(pipe)) {
|
| 3480 | struct pipe_buffer *buf = pipe_head_buf(pipe);
|
| 3481 |
|
| 3482 | *buf = (struct pipe_buffer) {
|
| 3483 | .ops = &zero_pipe_buf_ops,
|
| 3484 | .page = ZERO_PAGE(0),
|
| 3485 | .offset = offset,
|
| 3486 | .len = size,
|
| 3487 | };
|
| 3488 | pipe->head++;
|
| 3489 | }
|
| 3490 |
|
| 3491 | return size;
|
| 3492 | }
|
| 3493 |
|
| 3494 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
|
| 3495 | struct pipe_inode_info *pipe,
|
| 3496 | size_t len, unsigned int flags)
|
| 3497 | {
|
| 3498 | struct inode *inode = file_inode(in);
|
| 3499 | struct address_space *mapping = inode->i_mapping;
|
| 3500 | struct folio *folio = NULL;
|
| 3501 | size_t total_spliced = 0, used, npages, n, part;
|
| 3502 | loff_t isize;
|
| 3503 | int error = 0;
|
| 3504 |
|
| 3505 |
|
| 3506 | used = pipe_buf_usage(pipe);
|
| 3507 | npages = max_t(ssize_t, pipe->max_usage - used, 0);
|
| 3508 | len = min_t(size_t, len, npages * PAGE_SIZE);
|
| 3509 |
|
| 3510 | do {
|
| 3511 | bool fallback_page_splice = false;
|
| 3512 | struct page *page = NULL;
|
| 3513 | pgoff_t index;
|
| 3514 | size_t size;
|
| 3515 |
|
| 3516 | if (*ppos >= i_size_read(inode))
|
| 3517 | break;
|
| 3518 |
|
| 3519 | index = *ppos >> PAGE_SHIFT;
|
| 3520 | error = shmem_get_folio(inode, index, 0, &folio, SGP_READ);
|
| 3521 | if (error) {
|
| 3522 | if (error == -EINVAL)
|
| 3523 | error = 0;
|
| 3524 | break;
|
| 3525 | }
|
| 3526 | if (folio) {
|
| 3527 | folio_unlock(folio);
|
| 3528 |
|
| 3529 | page = folio_file_page(folio, index);
|
| 3530 | if (PageHWPoison(page)) {
|
| 3531 | error = -EIO;
|
| 3532 | break;
|
| 3533 | }
|
| 3534 |
|
| 3535 | if (folio_test_large(folio) &&
|
| 3536 | folio_test_has_hwpoisoned(folio))
|
| 3537 | fallback_page_splice = true;
|
| 3538 | }
|
| 3539 |
|
| 3540 | |
| 3541 | * i_size must be checked after we know the pages are Uptodate.
|
| 3542 | *
|
| 3543 | * Checking i_size after the check allows us to calculate
|
| 3544 | * the correct value for "nr", which means the zero-filled
|
| 3545 | * part of the page is not copied back to userspace (unless
|
| 3546 | * another truncate extends the file - this is desired though).
|
| 3547 |
|
| 3548 | isize = i_size_read(inode);
|
| 3549 | if (unlikely(*ppos >= isize))
|
| 3550 | break;
|
| 3551 | |
| 3552 | * Fallback to PAGE_SIZE splice if the large folio has hwpoisoned
|
| 3553 | * pages.
|
| 3554 |
|
| 3555 | size = len;
|
| 3556 | if (unlikely(fallback_page_splice)) {
|
| 3557 | size_t offset = *ppos & ~PAGE_MASK;
|
| 3558 |
|
| 3559 | size = umin(size, PAGE_SIZE - offset);
|
| 3560 | }
|
| 3561 | part = min_t(loff_t, isize - *ppos, size);
|
| 3562 |
|
| 3563 | if (folio) {
|
| 3564 | |
| 3565 | * If users can be writing to this page using arbitrary
|
| 3566 | * virtual addresses, take care about potential aliasing
|
| 3567 | * before reading the page on the kernel side.
|
| 3568 |
|
| 3569 | if (mapping_writably_mapped(mapping)) {
|
| 3570 | if (likely(!fallback_page_splice))
|
| 3571 | flush_dcache_folio(folio);
|
| 3572 | else
|
| 3573 | flush_dcache_page(page);
|
| 3574 | }
|
| 3575 | folio_mark_accessed(folio);
|
| 3576 | |
| 3577 | * Ok, we have the page, and it's up-to-date, so we can
|
| 3578 | * now splice it into the pipe.
|
| 3579 |
|
| 3580 | n = splice_folio_into_pipe(pipe, folio, *ppos, part);
|
| 3581 | folio_put(folio);
|
| 3582 | folio = NULL;
|
| 3583 | } else {
|
| 3584 | n = splice_zeropage_into_pipe(pipe, *ppos, part);
|
| 3585 | }
|
| 3586 |
|
| 3587 | if (!n)
|
| 3588 | break;
|
| 3589 | len -= n;
|
| 3590 | total_spliced += n;
|
| 3591 | *ppos += n;
|
| 3592 | in->f_ra.prev_pos = *ppos;
|
| 3593 | if (pipe_is_full(pipe))
|
| 3594 | break;
|
| 3595 |
|
| 3596 | cond_resched();
|
| 3597 | } while (len);
|
| 3598 |
|
| 3599 | if (folio)
|
| 3600 | folio_put(folio);
|
| 3601 |
|
| 3602 | file_accessed(in);
|
| 3603 | return total_spliced ? total_spliced : error;
|
| 3604 | }
|
| 3605 |
|
| 3606 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
|
| 3607 | {
|
| 3608 | struct address_space *mapping = file->f_mapping;
|
| 3609 | struct inode *inode = mapping->host;
|
| 3610 |
|
| 3611 | if (whence != SEEK_DATA && whence != SEEK_HOLE)
|
| 3612 | return generic_file_llseek_size(file, offset, whence,
|
| 3613 | MAX_LFS_FILESIZE, i_size_read(inode));
|
| 3614 | if (offset < 0)
|
| 3615 | return -ENXIO;
|
| 3616 |
|
| 3617 | inode_lock(inode);
|
| 3618 |
|
| 3619 | offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
|
| 3620 | if (offset >= 0)
|
| 3621 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
|
| 3622 | inode_unlock(inode);
|
| 3623 | return offset;
|
| 3624 | }
|
| 3625 |
|
| 3626 | static long shmem_fallocate(struct file *file, int mode, loff_t offset,
|
| 3627 | loff_t len)
|
| 3628 | {
|
| 3629 | struct inode *inode = file_inode(file);
|
| 3630 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 3631 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 3632 | struct shmem_falloc shmem_falloc;
|
| 3633 | pgoff_t start, index, end, undo_fallocend;
|
| 3634 | loff_t aligned_end;
|
| 3635 | int error;
|
| 3636 |
|
| 3637 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
|
| 3638 | return -EOPNOTSUPP;
|
| 3639 |
|
| 3640 | inode_lock(inode);
|
| 3641 |
|
| 3642 | if (info->flags & SHMEM_F_MAPPING_FROZEN) {
|
| 3643 | error = -EPERM;
|
| 3644 | goto out;
|
| 3645 | }
|
| 3646 |
|
| 3647 | if (mode & FALLOC_FL_PUNCH_HOLE) {
|
| 3648 | struct address_space *mapping = file->f_mapping;
|
| 3649 | loff_t unmap_start = round_up(offset, PAGE_SIZE);
|
| 3650 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
|
| 3651 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
|
| 3652 |
|
| 3653 |
|
| 3654 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
|
| 3655 | error = -EPERM;
|
| 3656 | goto out;
|
| 3657 | }
|
| 3658 |
|
| 3659 | shmem_falloc.waitq = &shmem_falloc_waitq;
|
| 3660 | shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
|
| 3661 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
|
| 3662 | spin_lock(&inode->i_lock);
|
| 3663 | WRITE_ONCE(inode->i_private, &shmem_falloc);
|
| 3664 | spin_unlock(&inode->i_lock);
|
| 3665 |
|
| 3666 | if ((u64)unmap_end > (u64)unmap_start)
|
| 3667 | unmap_mapping_range(mapping, unmap_start,
|
| 3668 | 1 + unmap_end - unmap_start, 0);
|
| 3669 | shmem_truncate_range(inode, offset, offset + len - 1);
|
| 3670 |
|
| 3671 |
|
| 3672 | spin_lock(&inode->i_lock);
|
| 3673 | WRITE_ONCE(inode->i_private, NULL);
|
| 3674 | wake_up_all(&shmem_falloc_waitq);
|
| 3675 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
|
| 3676 | spin_unlock(&inode->i_lock);
|
| 3677 | error = 0;
|
| 3678 | goto out;
|
| 3679 | }
|
| 3680 |
|
| 3681 |
|
| 3682 | error = inode_newsize_ok(inode, offset + len);
|
| 3683 | if (error)
|
| 3684 | goto out;
|
| 3685 |
|
| 3686 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
|
| 3687 | error = -EPERM;
|
| 3688 | goto out;
|
| 3689 | }
|
| 3690 |
|
| 3691 |
|
| 3692 | if (check_add_overflow(offset + len, (loff_t)PAGE_SIZE - 1,
|
| 3693 | &aligned_end)) {
|
| 3694 | error = -EFBIG;
|
| 3695 | goto out;
|
| 3696 | }
|
| 3697 |
|
| 3698 | start = offset >> PAGE_SHIFT;
|
| 3699 | end = aligned_end >> PAGE_SHIFT;
|
| 3700 |
|
| 3701 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
|
| 3702 | error = -ENOSPC;
|
| 3703 | goto out;
|
| 3704 | }
|
| 3705 |
|
| 3706 | shmem_falloc.waitq = NULL;
|
| 3707 | shmem_falloc.start = start;
|
| 3708 | shmem_falloc.next = start;
|
| 3709 | shmem_falloc.nr_falloced = 0;
|
| 3710 | shmem_falloc.nr_unswapped = 0;
|
| 3711 | spin_lock(&inode->i_lock);
|
| 3712 | WRITE_ONCE(inode->i_private, &shmem_falloc);
|
| 3713 | spin_unlock(&inode->i_lock);
|
| 3714 |
|
| 3715 | |
| 3716 | * info->fallocend is only relevant when huge pages might be
|
| 3717 | * involved: to prevent split_huge_page() freeing fallocated
|
| 3718 | * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
|
| 3719 |
|
| 3720 | undo_fallocend = info->fallocend;
|
| 3721 | if (info->fallocend < end)
|
| 3722 | info->fallocend = end;
|
| 3723 |
|
| 3724 | for (index = start; index < end; ) {
|
| 3725 | struct folio *folio;
|
| 3726 |
|
| 3727 | |
| 3728 | * Check for fatal signal so that we abort early in OOM
|
| 3729 | * situations. We don't want to abort in case of non-fatal
|
| 3730 | * signals as large fallocate can take noticeable time and
|
| 3731 | * e.g. periodic timers may result in fallocate constantly
|
| 3732 | * restarting.
|
| 3733 |
|
| 3734 | if (fatal_signal_pending(current))
|
| 3735 | error = -EINTR;
|
| 3736 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
|
| 3737 | error = -ENOMEM;
|
| 3738 | else
|
| 3739 | error = shmem_get_folio(inode, index, offset + len,
|
| 3740 | &folio, SGP_FALLOC);
|
| 3741 | if (error) {
|
| 3742 | info->fallocend = undo_fallocend;
|
| 3743 |
|
| 3744 | if (index > start) {
|
| 3745 | shmem_undo_range(inode,
|
| 3746 | (loff_t)start << PAGE_SHIFT,
|
| 3747 | ((loff_t)index << PAGE_SHIFT) - 1, true);
|
| 3748 | }
|
| 3749 | goto undone;
|
| 3750 | }
|
| 3751 |
|
| 3752 | |
| 3753 | * Here is a more important optimization than it appears:
|
| 3754 | * a second SGP_FALLOC on the same large folio will clear it,
|
| 3755 | * making it uptodate and un-undoable if we fail later.
|
| 3756 |
|
| 3757 | index = folio_next_index(folio);
|
| 3758 |
|
| 3759 | if (!index)
|
| 3760 | index--;
|
| 3761 |
|
| 3762 | |
| 3763 | * Inform shmem_writeout() how far we have reached.
|
| 3764 | * No need for lock or barrier: we have the page lock.
|
| 3765 |
|
| 3766 | if (!folio_test_uptodate(folio))
|
| 3767 | shmem_falloc.nr_falloced += index - shmem_falloc.next;
|
| 3768 | shmem_falloc.next = index;
|
| 3769 |
|
| 3770 | |
| 3771 | * If !uptodate, leave it that way so that freeable folios
|
| 3772 | * can be recognized if we need to rollback on error later.
|
| 3773 | * But mark it dirty so that memory pressure will swap rather
|
| 3774 | * than free the folios we are allocating (and SGP_CACHE folios
|
| 3775 | * might still be clean: we now need to mark those dirty too).
|
| 3776 |
|
| 3777 | folio_mark_dirty(folio);
|
| 3778 | folio_unlock(folio);
|
| 3779 | folio_put(folio);
|
| 3780 | cond_resched();
|
| 3781 | }
|
| 3782 |
|
| 3783 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
|
| 3784 | i_size_write(inode, offset + len);
|
| 3785 | undone:
|
| 3786 | spin_lock(&inode->i_lock);
|
| 3787 | WRITE_ONCE(inode->i_private, NULL);
|
| 3788 | spin_unlock(&inode->i_lock);
|
| 3789 | out:
|
| 3790 | if (!error)
|
| 3791 | file_modified(file);
|
| 3792 | inode_unlock(inode);
|
| 3793 | return error;
|
| 3794 | }
|
| 3795 |
|
| 3796 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
|
| 3797 | {
|
| 3798 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
|
| 3799 |
|
| 3800 | buf->f_type = TMPFS_MAGIC;
|
| 3801 | buf->f_bsize = PAGE_SIZE;
|
| 3802 | buf->f_namelen = NAME_MAX;
|
| 3803 | if (sbinfo->max_blocks) {
|
| 3804 | buf->f_blocks = sbinfo->max_blocks;
|
| 3805 | buf->f_bavail =
|
| 3806 | buf->f_bfree = sbinfo->max_blocks -
|
| 3807 | percpu_counter_sum(&sbinfo->used_blocks);
|
| 3808 | }
|
| 3809 | if (sbinfo->max_inodes) {
|
| 3810 | buf->f_files = sbinfo->max_inodes;
|
| 3811 | buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE;
|
| 3812 | }
|
| 3813 |
|
| 3814 |
|
| 3815 | buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
|
| 3816 |
|
| 3817 | return 0;
|
| 3818 | }
|
| 3819 |
|
| 3820 | |
| 3821 | * File creation. Allocate an inode, and we're done..
|
| 3822 |
|
| 3823 | static int
|
| 3824 | shmem_mknod(const struct mnt_idmap *idmap, struct inode *dir,
|
| 3825 | struct dentry *dentry, umode_t mode, dev_t dev)
|
| 3826 | {
|
| 3827 | struct inode *inode;
|
| 3828 | int error;
|
| 3829 |
|
| 3830 | if (!generic_ci_validate_strict_name(dir, &dentry->d_name))
|
| 3831 | return -EINVAL;
|
| 3832 |
|
| 3833 | inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev,
|
| 3834 | mk_vma_flags(VMA_NORESERVE_BIT));
|
| 3835 | if (IS_ERR(inode))
|
| 3836 | return PTR_ERR(inode);
|
| 3837 |
|
| 3838 | error = simple_acl_create(dir, inode);
|
| 3839 | if (error)
|
| 3840 | goto out_iput;
|
| 3841 | error = security_inode_init_security(inode, dir, &dentry->d_name,
|
| 3842 | shmem_initxattrs, NULL);
|
| 3843 | if (error && error != -EOPNOTSUPP)
|
| 3844 | goto out_iput;
|
| 3845 |
|
| 3846 | error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
|
| 3847 | if (error)
|
| 3848 | goto out_iput;
|
| 3849 |
|
| 3850 | dir->i_size += BOGO_DIRENT_SIZE;
|
| 3851 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
|
| 3852 | inode_inc_iversion(dir);
|
| 3853 |
|
| 3854 | d_make_persistent(dentry, inode);
|
| 3855 | return error;
|
| 3856 |
|
| 3857 | out_iput:
|
| 3858 | iput(inode);
|
| 3859 | return error;
|
| 3860 | }
|
| 3861 |
|
| 3862 | static int
|
| 3863 | shmem_tmpfile(const struct mnt_idmap *idmap, struct inode *dir,
|
| 3864 | struct file *file, umode_t mode)
|
| 3865 | {
|
| 3866 | struct inode *inode;
|
| 3867 | int error;
|
| 3868 |
|
| 3869 | inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0,
|
| 3870 | mk_vma_flags(VMA_NORESERVE_BIT));
|
| 3871 | if (IS_ERR(inode)) {
|
| 3872 | error = PTR_ERR(inode);
|
| 3873 | goto err_out;
|
| 3874 | }
|
| 3875 | error = security_inode_init_security(inode, dir, NULL,
|
| 3876 | shmem_initxattrs, NULL);
|
| 3877 | if (error && error != -EOPNOTSUPP)
|
| 3878 | goto out_iput;
|
| 3879 | error = simple_acl_create(dir, inode);
|
| 3880 | if (error)
|
| 3881 | goto out_iput;
|
| 3882 | d_tmpfile(file, inode);
|
| 3883 |
|
| 3884 | err_out:
|
| 3885 | return finish_open_simple(file, error);
|
| 3886 | out_iput:
|
| 3887 | iput(inode);
|
| 3888 | return error;
|
| 3889 | }
|
| 3890 |
|
| 3891 | static struct dentry *shmem_mkdir(const struct mnt_idmap *idmap, struct inode *dir,
|
| 3892 | struct dentry *dentry, umode_t mode)
|
| 3893 | {
|
| 3894 | int error;
|
| 3895 |
|
| 3896 | error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
|
| 3897 | if (error)
|
| 3898 | return ERR_PTR(error);
|
| 3899 | inc_nlink(dir);
|
| 3900 | return NULL;
|
| 3901 | }
|
| 3902 |
|
| 3903 | static int shmem_create(const struct mnt_idmap *idmap, struct inode *dir,
|
| 3904 | struct dentry *dentry, umode_t mode)
|
| 3905 | {
|
| 3906 | return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
|
| 3907 | }
|
| 3908 |
|
| 3909 | |
| 3910 | * Link a file..
|
| 3911 |
|
| 3912 | static int shmem_link(struct dentry *old_dentry, struct inode *dir,
|
| 3913 | struct dentry *dentry)
|
| 3914 | {
|
| 3915 | struct inode *inode = d_inode(old_dentry);
|
| 3916 | int ret;
|
| 3917 |
|
| 3918 | |
| 3919 | * No ordinary (disk based) filesystem counts links as inodes;
|
| 3920 | * but each new link needs a new dentry, pinning lowmem, and
|
| 3921 | * tmpfs dentries cannot be pruned until they are unlinked.
|
| 3922 | * But if an O_TMPFILE file is linked into the tmpfs, the
|
| 3923 | * first link must skip that, to get the accounting right.
|
| 3924 |
|
| 3925 | if (inode->i_nlink) {
|
| 3926 | ret = shmem_reserve_inode(inode->i_sb, NULL);
|
| 3927 | if (ret)
|
| 3928 | return ret;
|
| 3929 | }
|
| 3930 |
|
| 3931 | ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
|
| 3932 | if (ret) {
|
| 3933 | if (inode->i_nlink)
|
| 3934 | shmem_free_inode(inode->i_sb, 0);
|
| 3935 | return ret;
|
| 3936 | }
|
| 3937 |
|
| 3938 | dir->i_size += BOGO_DIRENT_SIZE;
|
| 3939 | inode_inc_iversion(dir);
|
| 3940 | return simple_link(old_dentry, dir, dentry);
|
| 3941 | }
|
| 3942 |
|
| 3943 | static int shmem_unlink(struct inode *dir, struct dentry *dentry)
|
| 3944 | {
|
| 3945 | struct inode *inode = d_inode(dentry);
|
| 3946 |
|
| 3947 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
|
| 3948 | shmem_free_inode(inode->i_sb, 0);
|
| 3949 |
|
| 3950 | simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
|
| 3951 |
|
| 3952 | dir->i_size -= BOGO_DIRENT_SIZE;
|
| 3953 | inode_inc_iversion(dir);
|
| 3954 | simple_unlink(dir, dentry);
|
| 3955 |
|
| 3956 | |
| 3957 | * For now, VFS can't deal with case-insensitive negative dentries, so
|
| 3958 | * we invalidate them
|
| 3959 |
|
| 3960 | if (IS_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
|
| 3961 | d_invalidate(dentry);
|
| 3962 |
|
| 3963 | return 0;
|
| 3964 | }
|
| 3965 |
|
| 3966 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
|
| 3967 | {
|
| 3968 | if (!simple_empty(dentry))
|
| 3969 | return -ENOTEMPTY;
|
| 3970 |
|
| 3971 | drop_nlink(d_inode(dentry));
|
| 3972 | drop_nlink(dir);
|
| 3973 | return shmem_unlink(dir, dentry);
|
| 3974 | }
|
| 3975 |
|
| 3976 | static int shmem_whiteout(const struct mnt_idmap *idmap,
|
| 3977 | struct inode *old_dir, struct dentry *old_dentry)
|
| 3978 | {
|
| 3979 | struct dentry *whiteout;
|
| 3980 | int error;
|
| 3981 |
|
| 3982 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
|
| 3983 | if (!whiteout)
|
| 3984 | return -ENOMEM;
|
| 3985 | error = shmem_mknod(idmap, old_dir, whiteout,
|
| 3986 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
|
| 3987 | dput(whiteout);
|
| 3988 | return error;
|
| 3989 | }
|
| 3990 |
|
| 3991 | |
| 3992 | * The VFS layer already does all the dentry stuff for rename,
|
| 3993 | * we just have to decrement the usage count for the target if
|
| 3994 | * it exists so that the VFS layer correctly free's it when it
|
| 3995 | * gets overwritten.
|
| 3996 |
|
| 3997 | static int shmem_rename2(const struct mnt_idmap *idmap,
|
| 3998 | struct inode *old_dir, struct dentry *old_dentry,
|
| 3999 | struct inode *new_dir, struct dentry *new_dentry,
|
| 4000 | unsigned int flags)
|
| 4001 | {
|
| 4002 | struct inode *inode = d_inode(old_dentry);
|
| 4003 | int they_are_dirs = S_ISDIR(inode->i_mode);
|
| 4004 | bool had_offset = false;
|
| 4005 | int error;
|
| 4006 |
|
| 4007 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
|
| 4008 | return -EINVAL;
|
| 4009 |
|
| 4010 | if (flags & RENAME_EXCHANGE)
|
| 4011 | return simple_offset_rename_exchange(old_dir, old_dentry,
|
| 4012 | new_dir, new_dentry);
|
| 4013 |
|
| 4014 | if (!simple_empty(new_dentry))
|
| 4015 | return -ENOTEMPTY;
|
| 4016 |
|
| 4017 | error = simple_offset_add(shmem_get_offset_ctx(new_dir), new_dentry);
|
| 4018 | if (error == -EBUSY)
|
| 4019 | had_offset = true;
|
| 4020 | else if (unlikely(error))
|
| 4021 | return error;
|
| 4022 |
|
| 4023 | if (flags & RENAME_WHITEOUT) {
|
| 4024 | error = shmem_whiteout(idmap, old_dir, old_dentry);
|
| 4025 | if (error) {
|
| 4026 | if (!had_offset)
|
| 4027 | simple_offset_remove(shmem_get_offset_ctx(new_dir),
|
| 4028 | new_dentry);
|
| 4029 | return error;
|
| 4030 | }
|
| 4031 | }
|
| 4032 |
|
| 4033 | simple_offset_rename(old_dir, old_dentry, new_dir, new_dentry);
|
| 4034 | if (d_really_is_positive(new_dentry)) {
|
| 4035 | (void) shmem_unlink(new_dir, new_dentry);
|
| 4036 | if (they_are_dirs) {
|
| 4037 | drop_nlink(d_inode(new_dentry));
|
| 4038 | drop_nlink(old_dir);
|
| 4039 | }
|
| 4040 | } else if (they_are_dirs) {
|
| 4041 | drop_nlink(old_dir);
|
| 4042 | inc_nlink(new_dir);
|
| 4043 | }
|
| 4044 |
|
| 4045 | old_dir->i_size -= BOGO_DIRENT_SIZE;
|
| 4046 | new_dir->i_size += BOGO_DIRENT_SIZE;
|
| 4047 | simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
|
| 4048 | inode_inc_iversion(old_dir);
|
| 4049 | inode_inc_iversion(new_dir);
|
| 4050 | return 0;
|
| 4051 | }
|
| 4052 |
|
| 4053 | static int shmem_symlink(const struct mnt_idmap *idmap, struct inode *dir,
|
| 4054 | struct dentry *dentry, const char *symname)
|
| 4055 | {
|
| 4056 | int error;
|
| 4057 | int len;
|
| 4058 | struct inode *inode;
|
| 4059 | struct folio *folio;
|
| 4060 | char *link;
|
| 4061 |
|
| 4062 | len = strlen(symname) + 1;
|
| 4063 | if (len > PAGE_SIZE)
|
| 4064 | return -ENAMETOOLONG;
|
| 4065 |
|
| 4066 | inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
|
| 4067 | mk_vma_flags(VMA_NORESERVE_BIT));
|
| 4068 | if (IS_ERR(inode))
|
| 4069 | return PTR_ERR(inode);
|
| 4070 |
|
| 4071 | error = security_inode_init_security(inode, dir, &dentry->d_name,
|
| 4072 | shmem_initxattrs, NULL);
|
| 4073 | if (error && error != -EOPNOTSUPP)
|
| 4074 | goto out_iput;
|
| 4075 |
|
| 4076 | error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
|
| 4077 | if (error)
|
| 4078 | goto out_iput;
|
| 4079 |
|
| 4080 | inode->i_size = len-1;
|
| 4081 | if (len <= SHORT_SYMLINK_LEN) {
|
| 4082 | link = kmemdup(symname, len, GFP_KERNEL);
|
| 4083 | if (!link) {
|
| 4084 | error = -ENOMEM;
|
| 4085 | goto out_remove_offset;
|
| 4086 | }
|
| 4087 | inode->i_op = &shmem_short_symlink_operations;
|
| 4088 | inode_set_cached_link(inode, link, len - 1);
|
| 4089 | } else {
|
| 4090 | inode_nohighmem(inode);
|
| 4091 | inode->i_mapping->a_ops = &shmem_aops;
|
| 4092 | error = shmem_get_folio(inode, 0, 0, &folio, SGP_WRITE);
|
| 4093 | if (error)
|
| 4094 | goto out_remove_offset;
|
| 4095 | inode->i_op = &shmem_symlink_inode_operations;
|
| 4096 | memcpy(folio_address(folio), symname, len);
|
| 4097 | folio_zero_range(folio, len, folio_size(folio) - len);
|
| 4098 | folio_mark_uptodate(folio);
|
| 4099 | folio_mark_dirty(folio);
|
| 4100 | folio_unlock(folio);
|
| 4101 | folio_put(folio);
|
| 4102 | }
|
| 4103 | dir->i_size += BOGO_DIRENT_SIZE;
|
| 4104 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
|
| 4105 | inode_inc_iversion(dir);
|
| 4106 | d_make_persistent(dentry, inode);
|
| 4107 | return 0;
|
| 4108 |
|
| 4109 | out_remove_offset:
|
| 4110 | simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
|
| 4111 | out_iput:
|
| 4112 | iput(inode);
|
| 4113 | return error;
|
| 4114 | }
|
| 4115 |
|
| 4116 | static void shmem_put_link(void *arg)
|
| 4117 | {
|
| 4118 | folio_mark_accessed(arg);
|
| 4119 | folio_put(arg);
|
| 4120 | }
|
| 4121 |
|
| 4122 | static const char *shmem_get_link(struct dentry *dentry, struct inode *inode,
|
| 4123 | struct delayed_call *done)
|
| 4124 | {
|
| 4125 | struct folio *folio = NULL;
|
| 4126 | int error;
|
| 4127 |
|
| 4128 | if (!dentry) {
|
| 4129 | folio = filemap_get_folio(inode->i_mapping, 0);
|
| 4130 | if (IS_ERR(folio))
|
| 4131 | return ERR_PTR(-ECHILD);
|
| 4132 | if (PageHWPoison(folio_page(folio, 0)) ||
|
| 4133 | !folio_test_uptodate(folio)) {
|
| 4134 | folio_put(folio);
|
| 4135 | return ERR_PTR(-ECHILD);
|
| 4136 | }
|
| 4137 | } else {
|
| 4138 | error = shmem_get_folio(inode, 0, 0, &folio, SGP_READ);
|
| 4139 | if (error)
|
| 4140 | return ERR_PTR(error);
|
| 4141 | if (!folio)
|
| 4142 | return ERR_PTR(-ECHILD);
|
| 4143 | if (PageHWPoison(folio_page(folio, 0))) {
|
| 4144 | folio_unlock(folio);
|
| 4145 | folio_put(folio);
|
| 4146 | return ERR_PTR(-ECHILD);
|
| 4147 | }
|
| 4148 | folio_unlock(folio);
|
| 4149 | }
|
| 4150 | set_delayed_call(done, shmem_put_link, folio);
|
| 4151 | return folio_address(folio);
|
| 4152 | }
|
| 4153 |
|
| 4154 | #ifdef CONFIG_TMPFS_XATTR
|
| 4155 |
|
| 4156 | static int shmem_fileattr_get(struct dentry *dentry, struct file_kattr *fa)
|
| 4157 | {
|
| 4158 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
|
| 4159 |
|
| 4160 | fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
|
| 4161 |
|
| 4162 | return 0;
|
| 4163 | }
|
| 4164 |
|
| 4165 | static int shmem_fileattr_set(const struct mnt_idmap *idmap,
|
| 4166 | struct dentry *dentry, struct file_kattr *fa)
|
| 4167 | {
|
| 4168 | struct inode *inode = d_inode(dentry);
|
| 4169 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 4170 | int ret, flags;
|
| 4171 |
|
| 4172 | if (fileattr_has_fsx(fa))
|
| 4173 | return -EOPNOTSUPP;
|
| 4174 | if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
|
| 4175 | return -EOPNOTSUPP;
|
| 4176 |
|
| 4177 | flags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
|
| 4178 | (fa->flags & SHMEM_FL_USER_MODIFIABLE);
|
| 4179 |
|
| 4180 | ret = shmem_set_inode_flags(inode, flags, dentry);
|
| 4181 |
|
| 4182 | if (ret)
|
| 4183 | return ret;
|
| 4184 |
|
| 4185 | info->fsflags = flags;
|
| 4186 |
|
| 4187 | inode_set_ctime_current(inode);
|
| 4188 | inode_inc_iversion(inode);
|
| 4189 | return 0;
|
| 4190 | }
|
| 4191 |
|
| 4192 | |
| 4193 | * Superblocks without xattr inode operations may get some security.* xattr
|
| 4194 | * support from the LSM "for free". As soon as we have any other xattrs
|
| 4195 | * like ACLs, we also need to implement the security.* handlers at
|
| 4196 | * filesystem level, though.
|
| 4197 |
|
| 4198 |
|
| 4199 | |
| 4200 | * Callback for security_inode_init_security() for acquiring xattrs.
|
| 4201 |
|
| 4202 | static int shmem_initxattrs(struct inode *inode,
|
| 4203 | const struct xattr *xattr_array, void *fs_info)
|
| 4204 | {
|
| 4205 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 4206 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 4207 | const struct xattr *xattr;
|
| 4208 | size_t ispace = 0;
|
| 4209 |
|
| 4210 | if (sbinfo->max_inodes) {
|
| 4211 | for (xattr = xattr_array; xattr->name != NULL; xattr++) {
|
| 4212 | ispace += simple_xattr_space(xattr->name,
|
| 4213 | xattr->value_len + XATTR_SECURITY_PREFIX_LEN);
|
| 4214 | }
|
| 4215 | if (ispace) {
|
| 4216 | raw_spin_lock(&sbinfo->stat_lock);
|
| 4217 | if (sbinfo->free_ispace < ispace)
|
| 4218 | ispace = 0;
|
| 4219 | else
|
| 4220 | sbinfo->free_ispace -= ispace;
|
| 4221 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 4222 | if (!ispace)
|
| 4223 | return -ENOSPC;
|
| 4224 | }
|
| 4225 | }
|
| 4226 |
|
| 4227 | for (xattr = xattr_array; xattr->name != NULL; xattr++) {
|
| 4228 | CLASS(simple_xattr, new_xattr)(xattr->value, xattr->value_len);
|
| 4229 | if (IS_ERR(new_xattr))
|
| 4230 | break;
|
| 4231 |
|
| 4232 | new_xattr->name = kasprintf(GFP_KERNEL_ACCOUNT,
|
| 4233 | XATTR_SECURITY_PREFIX "%s", xattr->name);
|
| 4234 | if (!new_xattr->name)
|
| 4235 | break;
|
| 4236 |
|
| 4237 | if (simple_xattr_add(&sbinfo->xa_cache, &info->xattrs, new_xattr))
|
| 4238 | break;
|
| 4239 |
|
| 4240 | if (sbinfo->max_inodes)
|
| 4241 | ispace -= simple_xattr_space(new_xattr->name, new_xattr->size);
|
| 4242 | retain_and_null_ptr(new_xattr);
|
| 4243 | }
|
| 4244 |
|
| 4245 | if (xattr->name != NULL) {
|
| 4246 | if (ispace) {
|
| 4247 | raw_spin_lock(&sbinfo->stat_lock);
|
| 4248 | sbinfo->free_ispace += ispace;
|
| 4249 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 4250 | }
|
| 4251 | return -ENOMEM;
|
| 4252 | }
|
| 4253 | WARN_ON(ispace);
|
| 4254 |
|
| 4255 | return 0;
|
| 4256 | }
|
| 4257 |
|
| 4258 | static int shmem_xattr_handler_get(const struct xattr_handler *handler,
|
| 4259 | struct dentry *unused, struct inode *inode,
|
| 4260 | const char *name, void *buffer, size_t size)
|
| 4261 | {
|
| 4262 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 4263 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 4264 |
|
| 4265 | name = xattr_full_name(handler, name);
|
| 4266 | return simple_xattr_get(&sbinfo->xa_cache, &info->xattrs, name, buffer, size);
|
| 4267 | }
|
| 4268 |
|
| 4269 | static int shmem_xattr_handler_set(const struct xattr_handler *handler,
|
| 4270 | const struct mnt_idmap *idmap,
|
| 4271 | struct dentry *unused, struct inode *inode,
|
| 4272 | const char *name, const void *value,
|
| 4273 | size_t size, int flags)
|
| 4274 | {
|
| 4275 | struct shmem_inode_info *info = SHMEM_I(inode);
|
| 4276 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
|
| 4277 | struct simple_xattr *old_xattr;
|
| 4278 | size_t ispace = 0;
|
| 4279 |
|
| 4280 | name = xattr_full_name(handler, name);
|
| 4281 |
|
| 4282 | if (value && sbinfo->max_inodes) {
|
| 4283 | ispace = simple_xattr_space(name, size);
|
| 4284 | raw_spin_lock(&sbinfo->stat_lock);
|
| 4285 | if (sbinfo->free_ispace < ispace)
|
| 4286 | ispace = 0;
|
| 4287 | else
|
| 4288 | sbinfo->free_ispace -= ispace;
|
| 4289 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 4290 | if (!ispace)
|
| 4291 | return -ENOSPC;
|
| 4292 | }
|
| 4293 |
|
| 4294 | old_xattr = simple_xattr_set(&sbinfo->xa_cache, &info->xattrs, name, value, size, flags);
|
| 4295 | if (!IS_ERR(old_xattr)) {
|
| 4296 | ispace = 0;
|
| 4297 | if (old_xattr && sbinfo->max_inodes)
|
| 4298 | ispace = simple_xattr_space(old_xattr->name,
|
| 4299 | old_xattr->size);
|
| 4300 | simple_xattr_free_rcu(old_xattr);
|
| 4301 | old_xattr = NULL;
|
| 4302 | inode_set_ctime_current(inode);
|
| 4303 | inode_inc_iversion(inode);
|
| 4304 | }
|
| 4305 | if (ispace) {
|
| 4306 | raw_spin_lock(&sbinfo->stat_lock);
|
| 4307 | sbinfo->free_ispace += ispace;
|
| 4308 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 4309 | }
|
| 4310 | return PTR_ERR(old_xattr);
|
| 4311 | }
|
| 4312 |
|
| 4313 | static const struct xattr_handler shmem_security_xattr_handler = {
|
| 4314 | .prefix = XATTR_SECURITY_PREFIX,
|
| 4315 | .get = shmem_xattr_handler_get,
|
| 4316 | .set = shmem_xattr_handler_set,
|
| 4317 | };
|
| 4318 |
|
| 4319 | static const struct xattr_handler shmem_trusted_xattr_handler = {
|
| 4320 | .prefix = XATTR_TRUSTED_PREFIX,
|
| 4321 | .get = shmem_xattr_handler_get,
|
| 4322 | .set = shmem_xattr_handler_set,
|
| 4323 | };
|
| 4324 |
|
| 4325 | static const struct xattr_handler shmem_user_xattr_handler = {
|
| 4326 | .prefix = XATTR_USER_PREFIX,
|
| 4327 | .get = shmem_xattr_handler_get,
|
| 4328 | .set = shmem_xattr_handler_set,
|
| 4329 | };
|
| 4330 |
|
| 4331 | static const struct xattr_handler * const shmem_xattr_handlers[] = {
|
| 4332 | &shmem_security_xattr_handler,
|
| 4333 | &shmem_trusted_xattr_handler,
|
| 4334 | &shmem_user_xattr_handler,
|
| 4335 | NULL
|
| 4336 | };
|
| 4337 |
|
| 4338 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
|
| 4339 | {
|
| 4340 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
|
| 4341 |
|
| 4342 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
|
| 4343 | }
|
| 4344 | #endif
|
| 4345 |
|
| 4346 | static const struct inode_operations shmem_short_symlink_operations = {
|
| 4347 | .getattr = shmem_getattr,
|
| 4348 | .setattr = shmem_setattr,
|
| 4349 | .get_link = simple_get_link,
|
| 4350 | #ifdef CONFIG_TMPFS_XATTR
|
| 4351 | .listxattr = shmem_listxattr,
|
| 4352 | #endif
|
| 4353 | };
|
| 4354 |
|
| 4355 | static const struct inode_operations shmem_symlink_inode_operations = {
|
| 4356 | .getattr = shmem_getattr,
|
| 4357 | .setattr = shmem_setattr,
|
| 4358 | .get_link = shmem_get_link,
|
| 4359 | #ifdef CONFIG_TMPFS_XATTR
|
| 4360 | .listxattr = shmem_listxattr,
|
| 4361 | #endif
|
| 4362 | };
|
| 4363 |
|
| 4364 | static struct dentry *shmem_get_parent(struct dentry *child)
|
| 4365 | {
|
| 4366 | return ERR_PTR(-ESTALE);
|
| 4367 | }
|
| 4368 |
|
| 4369 | static int shmem_match(struct inode *ino, void *vfh)
|
| 4370 | {
|
| 4371 | __u32 *fh = vfh;
|
| 4372 | __u64 inum = fh[2];
|
| 4373 | inum = (inum << 32) | fh[1];
|
| 4374 | return ino->i_ino == inum && fh[0] == ino->i_generation;
|
| 4375 | }
|
| 4376 |
|
| 4377 |
|
| 4378 | static struct dentry *shmem_find_alias(struct inode *inode)
|
| 4379 | {
|
| 4380 | struct dentry *alias = d_find_alias(inode);
|
| 4381 |
|
| 4382 | return alias ?: d_find_any_alias(inode);
|
| 4383 | }
|
| 4384 |
|
| 4385 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
|
| 4386 | struct fid *fid, int fh_len, int fh_type)
|
| 4387 | {
|
| 4388 | struct inode *inode;
|
| 4389 | struct dentry *dentry = NULL;
|
| 4390 | u64 inum;
|
| 4391 |
|
| 4392 | if (fh_len < 3)
|
| 4393 | return NULL;
|
| 4394 |
|
| 4395 | inum = fid->raw[2];
|
| 4396 | inum = (inum << 32) | fid->raw[1];
|
| 4397 |
|
| 4398 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
|
| 4399 | shmem_match, fid->raw);
|
| 4400 | if (inode) {
|
| 4401 | dentry = shmem_find_alias(inode);
|
| 4402 | iput(inode);
|
| 4403 | }
|
| 4404 |
|
| 4405 | return dentry;
|
| 4406 | }
|
| 4407 |
|
| 4408 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
|
| 4409 | struct inode *parent)
|
| 4410 | {
|
| 4411 | if (*len < 3) {
|
| 4412 | *len = 3;
|
| 4413 | return FILEID_INVALID;
|
| 4414 | }
|
| 4415 |
|
| 4416 | if (inode_unhashed(inode)) {
|
| 4417 | |
| 4418 | * so as we hash inodes here rather than at creation
|
| 4419 | * time, we need a lock to ensure we only try
|
| 4420 | * to do it once
|
| 4421 |
|
| 4422 | static DEFINE_SPINLOCK(lock);
|
| 4423 | spin_lock(&lock);
|
| 4424 | if (inode_unhashed(inode))
|
| 4425 | __insert_inode_hash(inode,
|
| 4426 | inode->i_ino + inode->i_generation);
|
| 4427 | spin_unlock(&lock);
|
| 4428 | }
|
| 4429 |
|
| 4430 | fh[0] = inode->i_generation;
|
| 4431 | fh[1] = inode->i_ino;
|
| 4432 | fh[2] = ((__u64)inode->i_ino) >> 32;
|
| 4433 |
|
| 4434 | *len = 3;
|
| 4435 | return 1;
|
| 4436 | }
|
| 4437 |
|
| 4438 | static const struct export_operations shmem_export_ops = {
|
| 4439 | .get_parent = shmem_get_parent,
|
| 4440 | .encode_fh = shmem_encode_fh,
|
| 4441 | .fh_to_dentry = shmem_fh_to_dentry,
|
| 4442 | };
|
| 4443 |
|
| 4444 | enum shmem_param {
|
| 4445 | Opt_gid,
|
| 4446 | Opt_huge,
|
| 4447 | Opt_mode,
|
| 4448 | Opt_mpol,
|
| 4449 | Opt_nr_blocks,
|
| 4450 | Opt_nr_inodes,
|
| 4451 | Opt_size,
|
| 4452 | Opt_uid,
|
| 4453 | Opt_inode32,
|
| 4454 | Opt_inode64,
|
| 4455 | Opt_noswap,
|
| 4456 | Opt_quota,
|
| 4457 | Opt_usrquota,
|
| 4458 | Opt_grpquota,
|
| 4459 | Opt_usrquota_block_hardlimit,
|
| 4460 | Opt_usrquota_inode_hardlimit,
|
| 4461 | Opt_grpquota_block_hardlimit,
|
| 4462 | Opt_grpquota_inode_hardlimit,
|
| 4463 | Opt_casefold_version,
|
| 4464 | Opt_casefold,
|
| 4465 | Opt_strict_encoding,
|
| 4466 | };
|
| 4467 |
|
| 4468 | static const struct constant_table shmem_param_enums_huge[] = {
|
| 4469 | {"never", SHMEM_HUGE_NEVER },
|
| 4470 | {"always", SHMEM_HUGE_ALWAYS },
|
| 4471 | {"within_size", SHMEM_HUGE_WITHIN_SIZE },
|
| 4472 | {"advise", SHMEM_HUGE_ADVISE },
|
| 4473 | {}
|
| 4474 | };
|
| 4475 |
|
| 4476 | const struct fs_parameter_spec shmem_fs_parameters[] = {
|
| 4477 | fsparam_gid ("gid", Opt_gid),
|
| 4478 | fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
|
| 4479 | fsparam_u32oct("mode", Opt_mode),
|
| 4480 | fsparam_string("mpol", Opt_mpol),
|
| 4481 | fsparam_string("nr_blocks", Opt_nr_blocks),
|
| 4482 | fsparam_string("nr_inodes", Opt_nr_inodes),
|
| 4483 | fsparam_string("size", Opt_size),
|
| 4484 | fsparam_uid ("uid", Opt_uid),
|
| 4485 | fsparam_flag ("inode32", Opt_inode32),
|
| 4486 | fsparam_flag ("inode64", Opt_inode64),
|
| 4487 | fsparam_flag ("noswap", Opt_noswap),
|
| 4488 | #ifdef CONFIG_TMPFS_QUOTA
|
| 4489 | fsparam_flag ("quota", Opt_quota),
|
| 4490 | fsparam_flag ("usrquota", Opt_usrquota),
|
| 4491 | fsparam_flag ("grpquota", Opt_grpquota),
|
| 4492 | fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit),
|
| 4493 | fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit),
|
| 4494 | fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit),
|
| 4495 | fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit),
|
| 4496 | #endif
|
| 4497 | fsparam_string("casefold", Opt_casefold_version),
|
| 4498 | fsparam_flag ("casefold", Opt_casefold),
|
| 4499 | fsparam_flag ("strict_encoding", Opt_strict_encoding),
|
| 4500 | {}
|
| 4501 | };
|
| 4502 |
|
| 4503 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 4504 | static int shmem_parse_opt_casefold(struct fs_context *fc, struct fs_parameter *param,
|
| 4505 | bool latest_version)
|
| 4506 | {
|
| 4507 | struct shmem_options *ctx = fc->fs_private;
|
| 4508 | int version = UTF8_LATEST;
|
| 4509 | struct unicode_map *encoding;
|
| 4510 | char *version_str = param->string + 5;
|
| 4511 |
|
| 4512 | if (!latest_version) {
|
| 4513 | if (strncmp(param->string, "utf8-", 5))
|
| 4514 | return invalfc(fc, "Only UTF-8 encodings are supported "
|
| 4515 | "in the format: utf8-<version number>");
|
| 4516 |
|
| 4517 | version = utf8_parse_version(version_str);
|
| 4518 | if (version < 0)
|
| 4519 | return invalfc(fc, "Invalid UTF-8 version: %s", version_str);
|
| 4520 | }
|
| 4521 |
|
| 4522 | encoding = utf8_load(version);
|
| 4523 |
|
| 4524 | if (IS_ERR(encoding)) {
|
| 4525 | return invalfc(fc, "Failed loading UTF-8 version: utf8-%u.%u.%u\n",
|
| 4526 | unicode_major(version), unicode_minor(version),
|
| 4527 | unicode_rev(version));
|
| 4528 | }
|
| 4529 |
|
| 4530 | pr_info("tmpfs: Using encoding : utf8-%u.%u.%u\n",
|
| 4531 | unicode_major(version), unicode_minor(version), unicode_rev(version));
|
| 4532 |
|
| 4533 | ctx->encoding = encoding;
|
| 4534 |
|
| 4535 | return 0;
|
| 4536 | }
|
| 4537 | #else
|
| 4538 | static int shmem_parse_opt_casefold(struct fs_context *fc, struct fs_parameter *param,
|
| 4539 | bool latest_version)
|
| 4540 | {
|
| 4541 | return invalfc(fc, "tmpfs: Kernel not built with CONFIG_UNICODE\n");
|
| 4542 | }
|
| 4543 | #endif
|
| 4544 |
|
| 4545 | static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
|
| 4546 | {
|
| 4547 | struct shmem_options *ctx = fc->fs_private;
|
| 4548 | struct fs_parse_result result;
|
| 4549 | unsigned long long size;
|
| 4550 | char *rest;
|
| 4551 | int opt;
|
| 4552 | kuid_t kuid;
|
| 4553 | kgid_t kgid;
|
| 4554 |
|
| 4555 | opt = fs_parse(fc, shmem_fs_parameters, param, &result);
|
| 4556 | if (opt < 0)
|
| 4557 | return opt;
|
| 4558 |
|
| 4559 | switch (opt) {
|
| 4560 | case Opt_size:
|
| 4561 | size = memparse(param->string, &rest);
|
| 4562 | if (*rest == '%') {
|
| 4563 | size <<= PAGE_SHIFT;
|
| 4564 | size *= totalram_pages();
|
| 4565 | do_div(size, 100);
|
| 4566 | rest++;
|
| 4567 | }
|
| 4568 | if (*rest)
|
| 4569 | goto bad_value;
|
| 4570 | ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
|
| 4571 | ctx->seen |= SHMEM_SEEN_BLOCKS;
|
| 4572 | break;
|
| 4573 | case Opt_nr_blocks:
|
| 4574 | ctx->blocks = memparse(param->string, &rest);
|
| 4575 | if (*rest || ctx->blocks > LONG_MAX)
|
| 4576 | goto bad_value;
|
| 4577 | ctx->seen |= SHMEM_SEEN_BLOCKS;
|
| 4578 | break;
|
| 4579 | case Opt_nr_inodes:
|
| 4580 | ctx->inodes = memparse(param->string, &rest);
|
| 4581 | if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE)
|
| 4582 | goto bad_value;
|
| 4583 | ctx->seen |= SHMEM_SEEN_INODES;
|
| 4584 | break;
|
| 4585 | case Opt_mode:
|
| 4586 | ctx->mode = result.uint_32 & 07777;
|
| 4587 | break;
|
| 4588 | case Opt_uid:
|
| 4589 | kuid = result.uid;
|
| 4590 |
|
| 4591 | |
| 4592 | * The requested uid must be representable in the
|
| 4593 | * filesystem's idmapping.
|
| 4594 |
|
| 4595 | if (!kuid_has_mapping(fc->user_ns, kuid))
|
| 4596 | goto bad_value;
|
| 4597 |
|
| 4598 | ctx->uid = kuid;
|
| 4599 | break;
|
| 4600 | case Opt_gid:
|
| 4601 | kgid = result.gid;
|
| 4602 |
|
| 4603 | |
| 4604 | * The requested gid must be representable in the
|
| 4605 | * filesystem's idmapping.
|
| 4606 |
|
| 4607 | if (!kgid_has_mapping(fc->user_ns, kgid))
|
| 4608 | goto bad_value;
|
| 4609 |
|
| 4610 | ctx->gid = kgid;
|
| 4611 | break;
|
| 4612 | case Opt_huge:
|
| 4613 | ctx->huge = result.uint_32;
|
| 4614 | if (ctx->huge != SHMEM_HUGE_NEVER &&
|
| 4615 | !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
|
| 4616 | has_transparent_hugepage()))
|
| 4617 | goto unsupported_parameter;
|
| 4618 | ctx->seen |= SHMEM_SEEN_HUGE;
|
| 4619 | break;
|
| 4620 | case Opt_mpol:
|
| 4621 | if (IS_ENABLED(CONFIG_NUMA)) {
|
| 4622 | mpol_put(ctx->mpol);
|
| 4623 | ctx->mpol = NULL;
|
| 4624 | if (mpol_parse_str(param->string, &ctx->mpol))
|
| 4625 | goto bad_value;
|
| 4626 | break;
|
| 4627 | }
|
| 4628 | goto unsupported_parameter;
|
| 4629 | case Opt_inode32:
|
| 4630 | ctx->full_inums = false;
|
| 4631 | ctx->seen |= SHMEM_SEEN_INUMS;
|
| 4632 | break;
|
| 4633 | case Opt_inode64:
|
| 4634 | if (sizeof(ino_t) < 8) {
|
| 4635 | return invalfc(fc,
|
| 4636 | "Cannot use inode64 with <64bit inums in kernel\n");
|
| 4637 | }
|
| 4638 | ctx->full_inums = true;
|
| 4639 | ctx->seen |= SHMEM_SEEN_INUMS;
|
| 4640 | break;
|
| 4641 | case Opt_noswap:
|
| 4642 | if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
|
| 4643 | return invalfc(fc,
|
| 4644 | "Turning off swap in unprivileged tmpfs mounts unsupported");
|
| 4645 | }
|
| 4646 | ctx->noswap = true;
|
| 4647 | break;
|
| 4648 | case Opt_quota:
|
| 4649 | if (fc->user_ns != &init_user_ns)
|
| 4650 | return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
|
| 4651 | ctx->seen |= SHMEM_SEEN_QUOTA;
|
| 4652 | ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP);
|
| 4653 | break;
|
| 4654 | case Opt_usrquota:
|
| 4655 | if (fc->user_ns != &init_user_ns)
|
| 4656 | return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
|
| 4657 | ctx->seen |= SHMEM_SEEN_QUOTA;
|
| 4658 | ctx->quota_types |= QTYPE_MASK_USR;
|
| 4659 | break;
|
| 4660 | case Opt_grpquota:
|
| 4661 | if (fc->user_ns != &init_user_ns)
|
| 4662 | return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
|
| 4663 | ctx->seen |= SHMEM_SEEN_QUOTA;
|
| 4664 | ctx->quota_types |= QTYPE_MASK_GRP;
|
| 4665 | break;
|
| 4666 | case Opt_usrquota_block_hardlimit:
|
| 4667 | size = memparse(param->string, &rest);
|
| 4668 | if (*rest || !size)
|
| 4669 | goto bad_value;
|
| 4670 | if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
|
| 4671 | return invalfc(fc,
|
| 4672 | "User quota block hardlimit too large.");
|
| 4673 | ctx->qlimits.usrquota_bhardlimit = size;
|
| 4674 | break;
|
| 4675 | case Opt_grpquota_block_hardlimit:
|
| 4676 | size = memparse(param->string, &rest);
|
| 4677 | if (*rest || !size)
|
| 4678 | goto bad_value;
|
| 4679 | if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
|
| 4680 | return invalfc(fc,
|
| 4681 | "Group quota block hardlimit too large.");
|
| 4682 | ctx->qlimits.grpquota_bhardlimit = size;
|
| 4683 | break;
|
| 4684 | case Opt_usrquota_inode_hardlimit:
|
| 4685 | size = memparse(param->string, &rest);
|
| 4686 | if (*rest || !size)
|
| 4687 | goto bad_value;
|
| 4688 | if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
|
| 4689 | return invalfc(fc,
|
| 4690 | "User quota inode hardlimit too large.");
|
| 4691 | ctx->qlimits.usrquota_ihardlimit = size;
|
| 4692 | break;
|
| 4693 | case Opt_grpquota_inode_hardlimit:
|
| 4694 | size = memparse(param->string, &rest);
|
| 4695 | if (*rest || !size)
|
| 4696 | goto bad_value;
|
| 4697 | if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
|
| 4698 | return invalfc(fc,
|
| 4699 | "Group quota inode hardlimit too large.");
|
| 4700 | ctx->qlimits.grpquota_ihardlimit = size;
|
| 4701 | break;
|
| 4702 | case Opt_casefold_version:
|
| 4703 | return shmem_parse_opt_casefold(fc, param, false);
|
| 4704 | case Opt_casefold:
|
| 4705 | return shmem_parse_opt_casefold(fc, param, true);
|
| 4706 | case Opt_strict_encoding:
|
| 4707 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 4708 | ctx->strict_encoding = true;
|
| 4709 | break;
|
| 4710 | #else
|
| 4711 | return invalfc(fc, "tmpfs: Kernel not built with CONFIG_UNICODE\n");
|
| 4712 | #endif
|
| 4713 | }
|
| 4714 | return 0;
|
| 4715 |
|
| 4716 | unsupported_parameter:
|
| 4717 | return invalfc(fc, "Unsupported parameter '%s'", param->key);
|
| 4718 | bad_value:
|
| 4719 | return invalfc(fc, "Bad value for '%s'", param->key);
|
| 4720 | }
|
| 4721 |
|
| 4722 | static char *shmem_next_opt(char **s)
|
| 4723 | {
|
| 4724 | char *sbegin = *s;
|
| 4725 | char *p;
|
| 4726 |
|
| 4727 | if (sbegin == NULL)
|
| 4728 | return NULL;
|
| 4729 |
|
| 4730 | |
| 4731 | * NUL-terminate this option: unfortunately,
|
| 4732 | * mount options form a comma-separated list,
|
| 4733 | * but mpol's nodelist may also contain commas.
|
| 4734 |
|
| 4735 | for (;;) {
|
| 4736 | p = strchr(*s, ',');
|
| 4737 | if (p == NULL)
|
| 4738 | break;
|
| 4739 | *s = p + 1;
|
| 4740 | if (!isdigit(*(p+1))) {
|
| 4741 | *p = '\0';
|
| 4742 | return sbegin;
|
| 4743 | }
|
| 4744 | }
|
| 4745 |
|
| 4746 | *s = NULL;
|
| 4747 | return sbegin;
|
| 4748 | }
|
| 4749 |
|
| 4750 | static int shmem_parse_monolithic(struct fs_context *fc, void *data)
|
| 4751 | {
|
| 4752 | return vfs_parse_monolithic_sep(fc, data, shmem_next_opt);
|
| 4753 | }
|
| 4754 |
|
| 4755 | |
| 4756 | * Reconfigure a shmem filesystem.
|
| 4757 |
|
| 4758 | static int shmem_reconfigure(struct fs_context *fc)
|
| 4759 | {
|
| 4760 | struct shmem_options *ctx = fc->fs_private;
|
| 4761 | struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
|
| 4762 | unsigned long used_isp;
|
| 4763 | struct mempolicy *mpol = NULL;
|
| 4764 | const char *err;
|
| 4765 |
|
| 4766 | raw_spin_lock(&sbinfo->stat_lock);
|
| 4767 | used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace;
|
| 4768 |
|
| 4769 | if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
|
| 4770 | if (!sbinfo->max_blocks) {
|
| 4771 | err = "Cannot retroactively limit size";
|
| 4772 | goto out;
|
| 4773 | }
|
| 4774 | if (percpu_counter_compare(&sbinfo->used_blocks,
|
| 4775 | ctx->blocks) > 0) {
|
| 4776 | err = "Too small a size for current use";
|
| 4777 | goto out;
|
| 4778 | }
|
| 4779 | }
|
| 4780 | if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
|
| 4781 | if (!sbinfo->max_inodes) {
|
| 4782 | err = "Cannot retroactively limit inodes";
|
| 4783 | goto out;
|
| 4784 | }
|
| 4785 | if (ctx->inodes * BOGO_INODE_SIZE < used_isp) {
|
| 4786 | err = "Too few inodes for current use";
|
| 4787 | goto out;
|
| 4788 | }
|
| 4789 | }
|
| 4790 |
|
| 4791 | if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
|
| 4792 | sbinfo->next_ino > UINT_MAX) {
|
| 4793 | err = "Current inum too high to switch to 32-bit inums";
|
| 4794 | goto out;
|
| 4795 | }
|
| 4796 |
|
| 4797 | |
| 4798 | * "noswap" doesn't use fsparam_flag_no, i.e. there's no "swap"
|
| 4799 | * counterpart for (re-)enabling swap.
|
| 4800 |
|
| 4801 | if (ctx->noswap && !sbinfo->noswap) {
|
| 4802 | err = "Cannot disable swap on remount";
|
| 4803 | goto out;
|
| 4804 | }
|
| 4805 |
|
| 4806 | if (ctx->seen & SHMEM_SEEN_QUOTA &&
|
| 4807 | !sb_any_quota_loaded(fc->root->d_sb)) {
|
| 4808 | err = "Cannot enable quota on remount";
|
| 4809 | goto out;
|
| 4810 | }
|
| 4811 |
|
| 4812 | #ifdef CONFIG_TMPFS_QUOTA
|
| 4813 | #define CHANGED_LIMIT(name) \
|
| 4814 | (ctx->qlimits.name## hardlimit && \
|
| 4815 | (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
|
| 4816 |
|
| 4817 | if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) ||
|
| 4818 | CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) {
|
| 4819 | err = "Cannot change global quota limit on remount";
|
| 4820 | goto out;
|
| 4821 | }
|
| 4822 | #endif
|
| 4823 |
|
| 4824 | if (ctx->seen & SHMEM_SEEN_HUGE)
|
| 4825 | sbinfo->huge = ctx->huge;
|
| 4826 | if (ctx->seen & SHMEM_SEEN_INUMS)
|
| 4827 | sbinfo->full_inums = ctx->full_inums;
|
| 4828 | if (ctx->seen & SHMEM_SEEN_BLOCKS)
|
| 4829 | sbinfo->max_blocks = ctx->blocks;
|
| 4830 | if (ctx->seen & SHMEM_SEEN_INODES) {
|
| 4831 | sbinfo->max_inodes = ctx->inodes;
|
| 4832 | sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp;
|
| 4833 | }
|
| 4834 |
|
| 4835 | |
| 4836 | * Preserve previous mempolicy unless mpol remount option was specified.
|
| 4837 |
|
| 4838 | if (ctx->mpol) {
|
| 4839 | mpol = sbinfo->mpol;
|
| 4840 | sbinfo->mpol = ctx->mpol;
|
| 4841 | ctx->mpol = NULL;
|
| 4842 | }
|
| 4843 |
|
| 4844 | if (ctx->noswap)
|
| 4845 | sbinfo->noswap = true;
|
| 4846 |
|
| 4847 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 4848 | mpol_put(mpol);
|
| 4849 | return 0;
|
| 4850 | out:
|
| 4851 | raw_spin_unlock(&sbinfo->stat_lock);
|
| 4852 | return invalfc(fc, "%s", err);
|
| 4853 | }
|
| 4854 |
|
| 4855 | static int shmem_show_options(struct seq_file *seq, struct dentry *root)
|
| 4856 | {
|
| 4857 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
|
| 4858 | struct mempolicy *mpol;
|
| 4859 |
|
| 4860 | if (sbinfo->max_blocks != shmem_default_max_blocks())
|
| 4861 | seq_printf(seq, ",size=%luk", K(sbinfo->max_blocks));
|
| 4862 | if (sbinfo->max_inodes != shmem_default_max_inodes())
|
| 4863 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
|
| 4864 | if (sbinfo->mode != (0777 | S_ISVTX))
|
| 4865 | seq_printf(seq, ",mode=%03ho", sbinfo->mode);
|
| 4866 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
|
| 4867 | seq_printf(seq, ",uid=%u",
|
| 4868 | from_kuid_munged(&init_user_ns, sbinfo->uid));
|
| 4869 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
|
| 4870 | seq_printf(seq, ",gid=%u",
|
| 4871 | from_kgid_munged(&init_user_ns, sbinfo->gid));
|
| 4872 |
|
| 4873 | |
| 4874 | * Showing inode{64,32} might be useful even if it's the system default,
|
| 4875 | * since then people don't have to resort to checking both here and
|
| 4876 | * /proc/config.gz to confirm 64-bit inums were successfully applied
|
| 4877 | * (which may not even exist if IKCONFIG_PROC isn't enabled).
|
| 4878 | *
|
| 4879 | * We hide it when inode64 isn't the default and we are using 32-bit
|
| 4880 | * inodes, since that probably just means the feature isn't even under
|
| 4881 | * consideration.
|
| 4882 | *
|
| 4883 | * As such:
|
| 4884 | *
|
| 4885 | * +-----------------+-----------------+
|
| 4886 | * | TMPFS_INODE64=y | TMPFS_INODE64=n |
|
| 4887 | * +------------------+-----------------+-----------------+
|
| 4888 | * | full_inums=true | show | show |
|
| 4889 | * | full_inums=false | show | hide |
|
| 4890 | * +------------------+-----------------+-----------------+
|
| 4891 | *
|
| 4892 |
|
| 4893 | if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
|
| 4894 | seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
|
| 4895 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 4896 |
|
| 4897 | if (sbinfo->huge)
|
| 4898 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
|
| 4899 | #endif
|
| 4900 | mpol = shmem_get_sbmpol(sbinfo);
|
| 4901 | shmem_show_mpol(seq, mpol);
|
| 4902 | mpol_put(mpol);
|
| 4903 | if (sbinfo->noswap)
|
| 4904 | seq_printf(seq, ",noswap");
|
| 4905 | #ifdef CONFIG_TMPFS_QUOTA
|
| 4906 | if (sb_has_quota_active(root->d_sb, USRQUOTA))
|
| 4907 | seq_printf(seq, ",usrquota");
|
| 4908 | if (sb_has_quota_active(root->d_sb, GRPQUOTA))
|
| 4909 | seq_printf(seq, ",grpquota");
|
| 4910 | if (sbinfo->qlimits.usrquota_bhardlimit)
|
| 4911 | seq_printf(seq, ",usrquota_block_hardlimit=%lld",
|
| 4912 | sbinfo->qlimits.usrquota_bhardlimit);
|
| 4913 | if (sbinfo->qlimits.grpquota_bhardlimit)
|
| 4914 | seq_printf(seq, ",grpquota_block_hardlimit=%lld",
|
| 4915 | sbinfo->qlimits.grpquota_bhardlimit);
|
| 4916 | if (sbinfo->qlimits.usrquota_ihardlimit)
|
| 4917 | seq_printf(seq, ",usrquota_inode_hardlimit=%lld",
|
| 4918 | sbinfo->qlimits.usrquota_ihardlimit);
|
| 4919 | if (sbinfo->qlimits.grpquota_ihardlimit)
|
| 4920 | seq_printf(seq, ",grpquota_inode_hardlimit=%lld",
|
| 4921 | sbinfo->qlimits.grpquota_ihardlimit);
|
| 4922 | #endif
|
| 4923 | return 0;
|
| 4924 | }
|
| 4925 |
|
| 4926 | #endif
|
| 4927 |
|
| 4928 | static void shmem_put_super(struct super_block *sb)
|
| 4929 | {
|
| 4930 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
|
| 4931 |
|
| 4932 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 4933 | if (sb->s_encoding)
|
| 4934 | utf8_unload(sb->s_encoding);
|
| 4935 | #endif
|
| 4936 |
|
| 4937 | #ifdef CONFIG_TMPFS_QUOTA
|
| 4938 | shmem_disable_quotas(sb);
|
| 4939 | #endif
|
| 4940 | free_percpu(sbinfo->ino_batch);
|
| 4941 | percpu_counter_destroy(&sbinfo->used_blocks);
|
| 4942 | mpol_put(sbinfo->mpol);
|
| 4943 | #ifdef CONFIG_TMPFS_XATTR
|
| 4944 | simple_xattr_cache_cleanup(&sbinfo->xa_cache);
|
| 4945 | #endif
|
| 4946 | kfree(sbinfo);
|
| 4947 | sb->s_fs_info = NULL;
|
| 4948 | }
|
| 4949 |
|
| 4950 | #if IS_ENABLED(CONFIG_UNICODE) && defined(CONFIG_TMPFS)
|
| 4951 | static const struct dentry_operations shmem_ci_dentry_ops = {
|
| 4952 | .d_hash = generic_ci_d_hash,
|
| 4953 | .d_compare = generic_ci_d_compare,
|
| 4954 | };
|
| 4955 | #endif
|
| 4956 |
|
| 4957 | static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
|
| 4958 | {
|
| 4959 | struct shmem_options *ctx = fc->fs_private;
|
| 4960 | struct inode *inode;
|
| 4961 | struct shmem_sb_info *sbinfo;
|
| 4962 | int error = -ENOMEM;
|
| 4963 |
|
| 4964 |
|
| 4965 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
|
| 4966 | L1_CACHE_BYTES), GFP_KERNEL);
|
| 4967 | if (!sbinfo)
|
| 4968 | return error;
|
| 4969 |
|
| 4970 | sb->s_fs_info = sbinfo;
|
| 4971 |
|
| 4972 | #ifdef CONFIG_TMPFS
|
| 4973 | |
| 4974 | * Per default we only allow half of the physical ram per
|
| 4975 | * tmpfs instance, limiting inodes to one per page of lowmem;
|
| 4976 | * but the internal instance is left unlimited.
|
| 4977 |
|
| 4978 | if (!(sb->s_flags & SB_KERNMOUNT)) {
|
| 4979 | if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
|
| 4980 | ctx->blocks = shmem_default_max_blocks();
|
| 4981 | if (!(ctx->seen & SHMEM_SEEN_INODES))
|
| 4982 | ctx->inodes = shmem_default_max_inodes();
|
| 4983 | if (!(ctx->seen & SHMEM_SEEN_INUMS))
|
| 4984 | ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
|
| 4985 | sbinfo->noswap = ctx->noswap;
|
| 4986 | } else {
|
| 4987 | sb->s_flags |= SB_NOUSER;
|
| 4988 | }
|
| 4989 | sb->s_export_op = &shmem_export_ops;
|
| 4990 | sb->s_flags |= SB_NOSEC;
|
| 4991 |
|
| 4992 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 4993 | if (!ctx->encoding && ctx->strict_encoding) {
|
| 4994 | pr_err("tmpfs: strict_encoding option without encoding is forbidden\n");
|
| 4995 | error = -EINVAL;
|
| 4996 | goto failed;
|
| 4997 | }
|
| 4998 |
|
| 4999 | if (ctx->encoding) {
|
| 5000 | sb->s_encoding = ctx->encoding;
|
| 5001 | set_default_d_op(sb, &shmem_ci_dentry_ops);
|
| 5002 | if (ctx->strict_encoding)
|
| 5003 | sb->s_encoding_flags = SB_ENC_STRICT_MODE_FL;
|
| 5004 | }
|
| 5005 | #endif
|
| 5006 |
|
| 5007 | #else
|
| 5008 | sb->s_flags |= SB_NOUSER;
|
| 5009 | #endif
|
| 5010 | sb->s_d_flags |= DCACHE_DONTCACHE;
|
| 5011 | sbinfo->max_blocks = ctx->blocks;
|
| 5012 | sbinfo->max_inodes = ctx->inodes;
|
| 5013 | sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE;
|
| 5014 | if (sb->s_flags & SB_KERNMOUNT) {
|
| 5015 | sbinfo->ino_batch = alloc_percpu(ino_t);
|
| 5016 | if (!sbinfo->ino_batch)
|
| 5017 | goto failed;
|
| 5018 | }
|
| 5019 | sbinfo->uid = ctx->uid;
|
| 5020 | sbinfo->gid = ctx->gid;
|
| 5021 | sbinfo->full_inums = ctx->full_inums;
|
| 5022 | sbinfo->mode = ctx->mode;
|
| 5023 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 5024 | if (ctx->seen & SHMEM_SEEN_HUGE)
|
| 5025 | sbinfo->huge = ctx->huge;
|
| 5026 | else
|
| 5027 | sbinfo->huge = tmpfs_huge;
|
| 5028 | #endif
|
| 5029 | sbinfo->mpol = ctx->mpol;
|
| 5030 | ctx->mpol = NULL;
|
| 5031 |
|
| 5032 | raw_spin_lock_init(&sbinfo->stat_lock);
|
| 5033 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
|
| 5034 | goto failed;
|
| 5035 | spin_lock_init(&sbinfo->shrinklist_lock);
|
| 5036 | INIT_LIST_HEAD(&sbinfo->shrinklist);
|
| 5037 |
|
| 5038 | sb->s_maxbytes = MAX_LFS_FILESIZE;
|
| 5039 | sb->s_blocksize = PAGE_SIZE;
|
| 5040 | sb->s_blocksize_bits = PAGE_SHIFT;
|
| 5041 | sb->s_magic = TMPFS_MAGIC;
|
| 5042 | sb->s_op = &shmem_ops;
|
| 5043 | sb->s_time_gran = 1;
|
| 5044 | #ifdef CONFIG_TMPFS_XATTR
|
| 5045 | sb->s_xattr = shmem_xattr_handlers;
|
| 5046 | #endif
|
| 5047 | #ifdef CONFIG_TMPFS_POSIX_ACL
|
| 5048 | sb->s_flags |= SB_POSIXACL;
|
| 5049 | #endif
|
| 5050 | uuid_t uuid;
|
| 5051 | uuid_gen(&uuid);
|
| 5052 | super_set_uuid(sb, uuid.b, sizeof(uuid));
|
| 5053 |
|
| 5054 | #ifdef CONFIG_TMPFS_QUOTA
|
| 5055 | if (ctx->seen & SHMEM_SEEN_QUOTA) {
|
| 5056 | sb->dq_op = &shmem_quota_operations;
|
| 5057 | sb->s_qcop = &dquot_quotactl_sysfile_ops;
|
| 5058 | sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
|
| 5059 |
|
| 5060 |
|
| 5061 | memcpy(&sbinfo->qlimits, &ctx->qlimits,
|
| 5062 | sizeof(struct shmem_quota_limits));
|
| 5063 |
|
| 5064 | if (shmem_enable_quotas(sb, ctx->quota_types))
|
| 5065 | goto failed;
|
| 5066 | }
|
| 5067 | #endif
|
| 5068 |
|
| 5069 | inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL,
|
| 5070 | S_IFDIR | sbinfo->mode, 0,
|
| 5071 | mk_vma_flags(VMA_NORESERVE_BIT));
|
| 5072 | if (IS_ERR(inode)) {
|
| 5073 | error = PTR_ERR(inode);
|
| 5074 | goto failed;
|
| 5075 | }
|
| 5076 | inode->i_uid = sbinfo->uid;
|
| 5077 | inode->i_gid = sbinfo->gid;
|
| 5078 | sb->s_root = d_make_root(inode);
|
| 5079 | if (!sb->s_root)
|
| 5080 | goto failed;
|
| 5081 | return 0;
|
| 5082 |
|
| 5083 | failed:
|
| 5084 | shmem_put_super(sb);
|
| 5085 | return error;
|
| 5086 | }
|
| 5087 |
|
| 5088 | static int shmem_get_tree(struct fs_context *fc)
|
| 5089 | {
|
| 5090 | return get_tree_nodev(fc, shmem_fill_super);
|
| 5091 | }
|
| 5092 |
|
| 5093 | static void shmem_free_fc(struct fs_context *fc)
|
| 5094 | {
|
| 5095 | struct shmem_options *ctx = fc->fs_private;
|
| 5096 |
|
| 5097 | if (ctx) {
|
| 5098 | mpol_put(ctx->mpol);
|
| 5099 | kfree(ctx);
|
| 5100 | }
|
| 5101 | }
|
| 5102 |
|
| 5103 | static const struct fs_context_operations shmem_fs_context_ops = {
|
| 5104 | .free = shmem_free_fc,
|
| 5105 | .get_tree = shmem_get_tree,
|
| 5106 | #ifdef CONFIG_TMPFS
|
| 5107 | .parse_monolithic = shmem_parse_monolithic,
|
| 5108 | .parse_param = shmem_parse_one,
|
| 5109 | .reconfigure = shmem_reconfigure,
|
| 5110 | #endif
|
| 5111 | };
|
| 5112 |
|
| 5113 | static struct kmem_cache *shmem_inode_cachep __ro_after_init;
|
| 5114 |
|
| 5115 | static struct inode *shmem_alloc_inode(struct super_block *sb)
|
| 5116 | {
|
| 5117 | struct shmem_inode_info *info;
|
| 5118 | info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
|
| 5119 | if (!info)
|
| 5120 | return NULL;
|
| 5121 | return &info->vfs_inode;
|
| 5122 | }
|
| 5123 |
|
| 5124 | static void shmem_free_in_core_inode(struct inode *inode)
|
| 5125 | {
|
| 5126 | if (S_ISLNK(inode->i_mode))
|
| 5127 | kfree(inode->i_link);
|
| 5128 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
|
| 5129 | }
|
| 5130 |
|
| 5131 | static void shmem_destroy_inode(struct inode *inode)
|
| 5132 | {
|
| 5133 | if (S_ISREG(inode->i_mode))
|
| 5134 | mpol_free_shared_policy(&SHMEM_I(inode)->policy);
|
| 5135 | if (S_ISDIR(inode->i_mode))
|
| 5136 | simple_offset_destroy(shmem_get_offset_ctx(inode));
|
| 5137 | }
|
| 5138 |
|
| 5139 | static void shmem_init_inode(void *foo)
|
| 5140 | {
|
| 5141 | struct shmem_inode_info *info = foo;
|
| 5142 | inode_init_once(&info->vfs_inode);
|
| 5143 | }
|
| 5144 |
|
| 5145 | static void __init shmem_init_inodecache(void)
|
| 5146 | {
|
| 5147 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
|
| 5148 | sizeof(struct shmem_inode_info),
|
| 5149 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
|
| 5150 | }
|
| 5151 |
|
| 5152 | static void __init shmem_destroy_inodecache(void)
|
| 5153 | {
|
| 5154 | kmem_cache_destroy(shmem_inode_cachep);
|
| 5155 | }
|
| 5156 |
|
| 5157 |
|
| 5158 | static int shmem_error_remove_folio(struct address_space *mapping,
|
| 5159 | struct folio *folio)
|
| 5160 | {
|
| 5161 | return 0;
|
| 5162 | }
|
| 5163 |
|
| 5164 | static const struct address_space_operations shmem_aops = {
|
| 5165 | .dirty_folio = noop_dirty_folio,
|
| 5166 | #ifdef CONFIG_TMPFS
|
| 5167 | .write_begin = shmem_write_begin,
|
| 5168 | .write_end = shmem_write_end,
|
| 5169 | #endif
|
| 5170 | #ifdef CONFIG_MIGRATION
|
| 5171 | .migrate_folio = migrate_folio,
|
| 5172 | #endif
|
| 5173 | .error_remove_folio = shmem_error_remove_folio,
|
| 5174 | };
|
| 5175 |
|
| 5176 | static const struct file_operations shmem_file_operations = {
|
| 5177 | .mmap_prepare = shmem_mmap_prepare,
|
| 5178 | .open = shmem_file_open,
|
| 5179 | .get_unmapped_area = shmem_get_unmapped_area,
|
| 5180 | #ifdef CONFIG_TMPFS
|
| 5181 | .llseek = shmem_file_llseek,
|
| 5182 | .read_iter = shmem_file_read_iter,
|
| 5183 | .write_iter = shmem_file_write_iter,
|
| 5184 | .fsync = noop_fsync,
|
| 5185 | .splice_read = shmem_file_splice_read,
|
| 5186 | .splice_write = iter_file_splice_write,
|
| 5187 | .fallocate = shmem_fallocate,
|
| 5188 | .setlease = generic_setlease,
|
| 5189 | #endif
|
| 5190 | };
|
| 5191 |
|
| 5192 | static const struct inode_operations shmem_inode_operations = {
|
| 5193 | .getattr = shmem_getattr,
|
| 5194 | .setattr = shmem_setattr,
|
| 5195 | #ifdef CONFIG_TMPFS_XATTR
|
| 5196 | .listxattr = shmem_listxattr,
|
| 5197 | .set_acl = simple_set_acl,
|
| 5198 | .fileattr_get = shmem_fileattr_get,
|
| 5199 | .fileattr_set = shmem_fileattr_set,
|
| 5200 | #endif
|
| 5201 | };
|
| 5202 |
|
| 5203 | static const struct inode_operations shmem_dir_inode_operations = {
|
| 5204 | #ifdef CONFIG_TMPFS
|
| 5205 | .getattr = shmem_getattr,
|
| 5206 | .create = shmem_create,
|
| 5207 | .lookup = simple_lookup,
|
| 5208 | .link = shmem_link,
|
| 5209 | .unlink = shmem_unlink,
|
| 5210 | .symlink = shmem_symlink,
|
| 5211 | .mkdir = shmem_mkdir,
|
| 5212 | .rmdir = shmem_rmdir,
|
| 5213 | .mknod = shmem_mknod,
|
| 5214 | .rename = shmem_rename2,
|
| 5215 | .tmpfile = shmem_tmpfile,
|
| 5216 | .get_offset_ctx = shmem_get_offset_ctx,
|
| 5217 | #endif
|
| 5218 | #ifdef CONFIG_TMPFS_XATTR
|
| 5219 | .listxattr = shmem_listxattr,
|
| 5220 | .fileattr_get = shmem_fileattr_get,
|
| 5221 | .fileattr_set = shmem_fileattr_set,
|
| 5222 | #endif
|
| 5223 | #ifdef CONFIG_TMPFS_POSIX_ACL
|
| 5224 | .setattr = shmem_setattr,
|
| 5225 | .set_acl = simple_set_acl,
|
| 5226 | #endif
|
| 5227 | };
|
| 5228 |
|
| 5229 | static const struct inode_operations shmem_special_inode_operations = {
|
| 5230 | .getattr = shmem_getattr,
|
| 5231 | #ifdef CONFIG_TMPFS_XATTR
|
| 5232 | .listxattr = shmem_listxattr,
|
| 5233 | #endif
|
| 5234 | #ifdef CONFIG_TMPFS_POSIX_ACL
|
| 5235 | .setattr = shmem_setattr,
|
| 5236 | .set_acl = simple_set_acl,
|
| 5237 | #endif
|
| 5238 | };
|
| 5239 |
|
| 5240 | static const struct super_operations shmem_ops = {
|
| 5241 | .alloc_inode = shmem_alloc_inode,
|
| 5242 | .free_inode = shmem_free_in_core_inode,
|
| 5243 | .destroy_inode = shmem_destroy_inode,
|
| 5244 | #ifdef CONFIG_TMPFS
|
| 5245 | .statfs = shmem_statfs,
|
| 5246 | .show_options = shmem_show_options,
|
| 5247 | #endif
|
| 5248 | #ifdef CONFIG_TMPFS_QUOTA
|
| 5249 | .get_dquots = shmem_get_dquots,
|
| 5250 | #endif
|
| 5251 | .evict_inode = shmem_evict_inode,
|
| 5252 | .drop_inode = inode_just_drop,
|
| 5253 | .put_super = shmem_put_super,
|
| 5254 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 5255 | .nr_cached_objects = shmem_unused_huge_count,
|
| 5256 | .free_cached_objects = shmem_unused_huge_scan,
|
| 5257 | #endif
|
| 5258 | };
|
| 5259 |
|
| 5260 | static const struct vm_operations_struct shmem_vm_ops = {
|
| 5261 | .fault = shmem_fault,
|
| 5262 | .map_pages = filemap_map_pages,
|
| 5263 | #ifdef CONFIG_NUMA
|
| 5264 | .set_policy = shmem_set_policy,
|
| 5265 | .get_policy = shmem_get_policy,
|
| 5266 | #endif
|
| 5267 | #ifdef CONFIG_USERFAULTFD
|
| 5268 | .uffd_ops = &shmem_uffd_ops,
|
| 5269 | #endif
|
| 5270 | };
|
| 5271 |
|
| 5272 | static const struct vm_operations_struct shmem_anon_vm_ops = {
|
| 5273 | .fault = shmem_fault,
|
| 5274 | .map_pages = filemap_map_pages,
|
| 5275 | #ifdef CONFIG_NUMA
|
| 5276 | .set_policy = shmem_set_policy,
|
| 5277 | .get_policy = shmem_get_policy,
|
| 5278 | #endif
|
| 5279 | #ifdef CONFIG_USERFAULTFD
|
| 5280 | .uffd_ops = &shmem_uffd_ops,
|
| 5281 | #endif
|
| 5282 | };
|
| 5283 |
|
| 5284 | int shmem_init_fs_context(struct fs_context *fc)
|
| 5285 | {
|
| 5286 | struct shmem_options *ctx;
|
| 5287 |
|
| 5288 | ctx = kzalloc_obj(struct shmem_options);
|
| 5289 | if (!ctx)
|
| 5290 | return -ENOMEM;
|
| 5291 |
|
| 5292 | ctx->mode = 0777 | S_ISVTX;
|
| 5293 | ctx->uid = current_fsuid();
|
| 5294 | ctx->gid = current_fsgid();
|
| 5295 |
|
| 5296 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 5297 | ctx->encoding = NULL;
|
| 5298 | #endif
|
| 5299 |
|
| 5300 | fc->fs_private = ctx;
|
| 5301 | fc->ops = &shmem_fs_context_ops;
|
| 5302 | #ifdef CONFIG_TMPFS
|
| 5303 | fc->sb_flags |= SB_I_VERSION;
|
| 5304 | #endif
|
| 5305 | return 0;
|
| 5306 | }
|
| 5307 |
|
| 5308 | static struct file_system_type shmem_fs_type = {
|
| 5309 | .owner = THIS_MODULE,
|
| 5310 | .name = "tmpfs",
|
| 5311 | .init_fs_context = shmem_init_fs_context,
|
| 5312 | #ifdef CONFIG_TMPFS
|
| 5313 | .parameters = shmem_fs_parameters,
|
| 5314 | #endif
|
| 5315 | .kill_sb = kill_anon_super,
|
| 5316 | .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP | FS_MGTIME,
|
| 5317 | };
|
| 5318 |
|
| 5319 | #if defined(CONFIG_SYSFS) && defined(CONFIG_TMPFS)
|
| 5320 |
|
| 5321 | #define __INIT_KOBJ_ATTR(_name, _mode, _show, _store) \
|
| 5322 | { \
|
| 5323 | .attr = { .name = __stringify(_name), .mode = _mode }, \
|
| 5324 | .show = _show, \
|
| 5325 | .store = _store, \
|
| 5326 | }
|
| 5327 |
|
| 5328 | #define TMPFS_ATTR_W(_name, _store) \
|
| 5329 | static struct kobj_attribute tmpfs_attr_##_name = \
|
| 5330 | __INIT_KOBJ_ATTR(_name, 0200, NULL, _store)
|
| 5331 |
|
| 5332 | #define TMPFS_ATTR_RW(_name, _show, _store) \
|
| 5333 | static struct kobj_attribute tmpfs_attr_##_name = \
|
| 5334 | __INIT_KOBJ_ATTR(_name, 0644, _show, _store)
|
| 5335 |
|
| 5336 | #define TMPFS_ATTR_RO(_name, _show) \
|
| 5337 | static struct kobj_attribute tmpfs_attr_##_name = \
|
| 5338 | __INIT_KOBJ_ATTR(_name, 0444, _show, NULL)
|
| 5339 |
|
| 5340 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 5341 | static ssize_t casefold_show(struct kobject *kobj, struct kobj_attribute *a,
|
| 5342 | char *buf)
|
| 5343 | {
|
| 5344 | return sysfs_emit(buf, "supported\n");
|
| 5345 | }
|
| 5346 | TMPFS_ATTR_RO(casefold, casefold_show);
|
| 5347 | #endif
|
| 5348 |
|
| 5349 | static struct attribute *tmpfs_attributes[] = {
|
| 5350 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 5351 | &tmpfs_attr_casefold.attr,
|
| 5352 | #endif
|
| 5353 | NULL
|
| 5354 | };
|
| 5355 |
|
| 5356 | static const struct attribute_group tmpfs_attribute_group = {
|
| 5357 | .attrs = tmpfs_attributes,
|
| 5358 | .name = "features"
|
| 5359 | };
|
| 5360 |
|
| 5361 | static struct kobject *tmpfs_kobj;
|
| 5362 |
|
| 5363 | static int __init tmpfs_sysfs_init(void)
|
| 5364 | {
|
| 5365 | int ret;
|
| 5366 |
|
| 5367 | tmpfs_kobj = kobject_create_and_add("tmpfs", fs_kobj);
|
| 5368 | if (!tmpfs_kobj)
|
| 5369 | return -ENOMEM;
|
| 5370 |
|
| 5371 | ret = sysfs_create_group(tmpfs_kobj, &tmpfs_attribute_group);
|
| 5372 | if (ret)
|
| 5373 | kobject_put(tmpfs_kobj);
|
| 5374 |
|
| 5375 | return ret;
|
| 5376 | }
|
| 5377 | #endif
|
| 5378 |
|
| 5379 | void __init shmem_init(void)
|
| 5380 | {
|
| 5381 | int error;
|
| 5382 |
|
| 5383 | shmem_init_inodecache();
|
| 5384 |
|
| 5385 | #ifdef CONFIG_TMPFS_QUOTA
|
| 5386 | register_quota_format(&shmem_quota_format);
|
| 5387 | #endif
|
| 5388 |
|
| 5389 | error = register_filesystem(&shmem_fs_type);
|
| 5390 | if (error) {
|
| 5391 | pr_err("Could not register tmpfs\n");
|
| 5392 | goto out2;
|
| 5393 | }
|
| 5394 |
|
| 5395 | shm_mnt = kern_mount(&shmem_fs_type);
|
| 5396 | if (IS_ERR(shm_mnt)) {
|
| 5397 | error = PTR_ERR(shm_mnt);
|
| 5398 | pr_err("Could not kern_mount tmpfs\n");
|
| 5399 | goto out1;
|
| 5400 | }
|
| 5401 |
|
| 5402 | #if defined(CONFIG_SYSFS) && defined(CONFIG_TMPFS)
|
| 5403 | error = tmpfs_sysfs_init();
|
| 5404 | if (error) {
|
| 5405 | pr_err("Could not init tmpfs sysfs\n");
|
| 5406 | goto out1;
|
| 5407 | }
|
| 5408 | #endif
|
| 5409 |
|
| 5410 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
| 5411 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
|
| 5412 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
|
| 5413 | else
|
| 5414 | shmem_huge = SHMEM_HUGE_NEVER;
|
| 5415 |
|
| 5416 | |
| 5417 | * Default to setting PMD-sized THP to inherit the global setting and
|
| 5418 | * disable all other multi-size THPs.
|
| 5419 |
|
| 5420 | if (!shmem_orders_configured)
|
| 5421 | huge_shmem_orders_inherit = BIT(HPAGE_PMD_ORDER);
|
| 5422 | #endif
|
| 5423 | return;
|
| 5424 |
|
| 5425 | out1:
|
| 5426 | unregister_filesystem(&shmem_fs_type);
|
| 5427 | out2:
|
| 5428 | #ifdef CONFIG_TMPFS_QUOTA
|
| 5429 | unregister_quota_format(&shmem_quota_format);
|
| 5430 | #endif
|
| 5431 | shmem_destroy_inodecache();
|
| 5432 | shm_mnt = ERR_PTR(error);
|
| 5433 | }
|
| 5434 |
|
| 5435 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
|
| 5436 | static ssize_t shmem_enabled_show(struct kobject *kobj,
|
| 5437 | struct kobj_attribute *attr, char *buf)
|
| 5438 | {
|
| 5439 | static const int values[] = {
|
| 5440 | SHMEM_HUGE_ALWAYS,
|
| 5441 | SHMEM_HUGE_WITHIN_SIZE,
|
| 5442 | SHMEM_HUGE_ADVISE,
|
| 5443 | SHMEM_HUGE_NEVER,
|
| 5444 | SHMEM_HUGE_DENY,
|
| 5445 | SHMEM_HUGE_FORCE,
|
| 5446 | };
|
| 5447 | int len = 0;
|
| 5448 | int i;
|
| 5449 |
|
| 5450 | for (i = 0; i < ARRAY_SIZE(values); i++) {
|
| 5451 | len += sysfs_emit_at(buf, len,
|
| 5452 | shmem_huge == values[i] ? "%s[%s]" : "%s%s",
|
| 5453 | i ? " " : "", shmem_format_huge(values[i]));
|
| 5454 | }
|
| 5455 | len += sysfs_emit_at(buf, len, "\n");
|
| 5456 |
|
| 5457 | return len;
|
| 5458 | }
|
| 5459 |
|
| 5460 | static ssize_t shmem_enabled_store(struct kobject *kobj,
|
| 5461 | struct kobj_attribute *attr, const char *buf, size_t count)
|
| 5462 | {
|
| 5463 | char tmp[16];
|
| 5464 | int huge, err;
|
| 5465 |
|
| 5466 | if (count + 1 > sizeof(tmp))
|
| 5467 | return -EINVAL;
|
| 5468 | memcpy(tmp, buf, count);
|
| 5469 | tmp[count] = '\0';
|
| 5470 | if (count && tmp[count - 1] == '\n')
|
| 5471 | tmp[count - 1] = '\0';
|
| 5472 |
|
| 5473 | huge = shmem_parse_huge(tmp);
|
| 5474 | if (huge == -EINVAL)
|
| 5475 | return huge;
|
| 5476 |
|
| 5477 | shmem_huge = huge;
|
| 5478 | if (shmem_huge > SHMEM_HUGE_DENY)
|
| 5479 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
|
| 5480 |
|
| 5481 | err = start_stop_khugepaged();
|
| 5482 | return err ? err : count;
|
| 5483 | }
|
| 5484 |
|
| 5485 | struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
|
| 5486 | static DEFINE_SPINLOCK(huge_shmem_orders_lock);
|
| 5487 |
|
| 5488 | enum huge_mode {
|
| 5489 | HUGE_SHMEM_ENABLED_ALWAYS = 0,
|
| 5490 | HUGE_SHMEM_ENABLED_INHERIT,
|
| 5491 | HUGE_SHMEM_ENABLED_WITHIN_SIZE,
|
| 5492 | HUGE_SHMEM_ENABLED_ADVISE,
|
| 5493 | HUGE_SHMEM_ENABLED_NEVER,
|
| 5494 | };
|
| 5495 |
|
| 5496 | static const char * const huge_mode_strings[] = {
|
| 5497 | [HUGE_SHMEM_ENABLED_ALWAYS] = "always",
|
| 5498 | [HUGE_SHMEM_ENABLED_INHERIT] = "inherit",
|
| 5499 | [HUGE_SHMEM_ENABLED_WITHIN_SIZE] = "within_size",
|
| 5500 | [HUGE_SHMEM_ENABLED_ADVISE] = "advise",
|
| 5501 | [HUGE_SHMEM_ENABLED_NEVER] = "never",
|
| 5502 | };
|
| 5503 |
|
| 5504 | static unsigned long * const huge_mode_orders[] = {
|
| 5505 | [HUGE_SHMEM_ENABLED_ALWAYS] = &huge_shmem_orders_always,
|
| 5506 | [HUGE_SHMEM_ENABLED_INHERIT] = &huge_shmem_orders_inherit,
|
| 5507 | [HUGE_SHMEM_ENABLED_WITHIN_SIZE] = &huge_shmem_orders_within_size,
|
| 5508 | [HUGE_SHMEM_ENABLED_ADVISE] = &huge_shmem_orders_madvise,
|
| 5509 | };
|
| 5510 |
|
| 5511 | static ssize_t thpsize_shmem_enabled_show(struct kobject *kobj,
|
| 5512 | struct kobj_attribute *attr, char *buf)
|
| 5513 | {
|
| 5514 | int order = to_thpsize(kobj)->order;
|
| 5515 | int active = HUGE_SHMEM_ENABLED_NEVER;
|
| 5516 | int len = 0;
|
| 5517 | int i;
|
| 5518 |
|
| 5519 | for (i = 0; i < ARRAY_SIZE(huge_mode_orders); i++) {
|
| 5520 | if (test_bit(order, huge_mode_orders[i])) {
|
| 5521 | active = i;
|
| 5522 | break;
|
| 5523 | }
|
| 5524 | }
|
| 5525 |
|
| 5526 | for (i = 0; i < ARRAY_SIZE(huge_mode_strings); i++) {
|
| 5527 | if (i == active)
|
| 5528 | len += sysfs_emit_at(buf, len, "[%s] ",
|
| 5529 | huge_mode_strings[i]);
|
| 5530 | else
|
| 5531 | len += sysfs_emit_at(buf, len, "%s ",
|
| 5532 | huge_mode_strings[i]);
|
| 5533 | }
|
| 5534 |
|
| 5535 |
|
| 5536 | buf[len - 1] = '\n';
|
| 5537 |
|
| 5538 | return len;
|
| 5539 | }
|
| 5540 |
|
| 5541 | static bool set_shmem_enabled_mode(int order, enum huge_mode mode)
|
| 5542 | {
|
| 5543 | bool changed = false;
|
| 5544 | enum huge_mode idx;
|
| 5545 |
|
| 5546 | spin_lock(&huge_shmem_orders_lock);
|
| 5547 | for (idx = 0; idx < ARRAY_SIZE(huge_mode_orders); idx++) {
|
| 5548 | if (idx == mode)
|
| 5549 | changed |= !__test_and_set_bit(order, huge_mode_orders[idx]);
|
| 5550 | else
|
| 5551 | changed |= __test_and_clear_bit(order, huge_mode_orders[idx]);
|
| 5552 | }
|
| 5553 | spin_unlock(&huge_shmem_orders_lock);
|
| 5554 |
|
| 5555 | return changed;
|
| 5556 | }
|
| 5557 |
|
| 5558 | static ssize_t thpsize_shmem_enabled_store(struct kobject *kobj,
|
| 5559 | struct kobj_attribute *attr,
|
| 5560 | const char *buf, size_t count)
|
| 5561 | {
|
| 5562 | int order = to_thpsize(kobj)->order;
|
| 5563 | int mode;
|
| 5564 |
|
| 5565 | mode = sysfs_match_string(huge_mode_strings, buf);
|
| 5566 | if (mode < 0)
|
| 5567 | return mode;
|
| 5568 |
|
| 5569 |
|
| 5570 | if (mode == HUGE_SHMEM_ENABLED_INHERIT &&
|
| 5571 | shmem_huge == SHMEM_HUGE_FORCE && !is_pmd_order(order))
|
| 5572 | return -EINVAL;
|
| 5573 |
|
| 5574 | if (set_shmem_enabled_mode(order, mode)) {
|
| 5575 | int err = start_stop_khugepaged();
|
| 5576 | if (err)
|
| 5577 | return err;
|
| 5578 | } else {
|
| 5579 | |
| 5580 | * Recalculate watermarks even when the mode hasn't changed
|
| 5581 | * to preserve the legacy behavior, as this is always called
|
| 5582 | * inside start_stop_khugepaged().
|
| 5583 |
|
| 5584 | set_recommended_min_free_kbytes();
|
| 5585 | }
|
| 5586 |
|
| 5587 | return count;
|
| 5588 | }
|
| 5589 |
|
| 5590 | struct kobj_attribute thpsize_shmem_enabled_attr =
|
| 5591 | __ATTR(shmem_enabled, 0644, thpsize_shmem_enabled_show, thpsize_shmem_enabled_store);
|
| 5592 | #endif
|
| 5593 |
|
| 5594 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE)
|
| 5595 |
|
| 5596 | static int __init setup_transparent_hugepage_shmem(char *str)
|
| 5597 | {
|
| 5598 | int huge;
|
| 5599 |
|
| 5600 | huge = shmem_parse_huge(str);
|
| 5601 | if (huge == -EINVAL) {
|
| 5602 | pr_warn("transparent_hugepage_shmem= cannot parse, ignored\n");
|
| 5603 | return huge;
|
| 5604 | }
|
| 5605 |
|
| 5606 | shmem_huge = huge;
|
| 5607 | return 1;
|
| 5608 | }
|
| 5609 | __setup("transparent_hugepage_shmem=", setup_transparent_hugepage_shmem);
|
| 5610 |
|
| 5611 | static int __init setup_transparent_hugepage_tmpfs(char *str)
|
| 5612 | {
|
| 5613 | int huge;
|
| 5614 |
|
| 5615 | huge = shmem_parse_huge(str);
|
| 5616 | if (huge < 0) {
|
| 5617 | pr_warn("transparent_hugepage_tmpfs= cannot parse, ignored\n");
|
| 5618 | return huge;
|
| 5619 | }
|
| 5620 |
|
| 5621 | tmpfs_huge = huge;
|
| 5622 | return 1;
|
| 5623 | }
|
| 5624 | __setup("transparent_hugepage_tmpfs=", setup_transparent_hugepage_tmpfs);
|
| 5625 |
|
| 5626 | static char str_dup[PAGE_SIZE] __initdata;
|
| 5627 | static int __init setup_thp_shmem(char *str)
|
| 5628 | {
|
| 5629 | char *token, *range, *policy, *subtoken;
|
| 5630 | unsigned long always, inherit, madvise, within_size;
|
| 5631 | char *start_size, *end_size;
|
| 5632 | int start, end, nr;
|
| 5633 | char *p;
|
| 5634 |
|
| 5635 | if (!str || strlen(str) + 1 > PAGE_SIZE)
|
| 5636 | goto err;
|
| 5637 | strscpy(str_dup, str);
|
| 5638 |
|
| 5639 | always = huge_shmem_orders_always;
|
| 5640 | inherit = huge_shmem_orders_inherit;
|
| 5641 | madvise = huge_shmem_orders_madvise;
|
| 5642 | within_size = huge_shmem_orders_within_size;
|
| 5643 | p = str_dup;
|
| 5644 | while ((token = strsep(&p, ";")) != NULL) {
|
| 5645 | range = strsep(&token, ":");
|
| 5646 | policy = token;
|
| 5647 |
|
| 5648 | if (!policy)
|
| 5649 | goto err;
|
| 5650 |
|
| 5651 | while ((subtoken = strsep(&range, ",")) != NULL) {
|
| 5652 | if (strchr(subtoken, '-')) {
|
| 5653 | start_size = strsep(&subtoken, "-");
|
| 5654 | end_size = subtoken;
|
| 5655 |
|
| 5656 | start = get_order_from_str(start_size,
|
| 5657 | THP_ORDERS_ALL_FILE_DEFAULT);
|
| 5658 | end = get_order_from_str(end_size,
|
| 5659 | THP_ORDERS_ALL_FILE_DEFAULT);
|
| 5660 | } else {
|
| 5661 | start_size = end_size = subtoken;
|
| 5662 | start = end = get_order_from_str(subtoken,
|
| 5663 | THP_ORDERS_ALL_FILE_DEFAULT);
|
| 5664 | }
|
| 5665 |
|
| 5666 | if (start < 0) {
|
| 5667 | pr_err("invalid size %s in thp_shmem boot parameter\n",
|
| 5668 | start_size);
|
| 5669 | goto err;
|
| 5670 | }
|
| 5671 |
|
| 5672 | if (end < 0) {
|
| 5673 | pr_err("invalid size %s in thp_shmem boot parameter\n",
|
| 5674 | end_size);
|
| 5675 | goto err;
|
| 5676 | }
|
| 5677 |
|
| 5678 | if (start > end)
|
| 5679 | goto err;
|
| 5680 |
|
| 5681 | nr = end - start + 1;
|
| 5682 | if (!strcmp(policy, "always")) {
|
| 5683 | bitmap_set(&always, start, nr);
|
| 5684 | bitmap_clear(&inherit, start, nr);
|
| 5685 | bitmap_clear(&madvise, start, nr);
|
| 5686 | bitmap_clear(&within_size, start, nr);
|
| 5687 | } else if (!strcmp(policy, "advise")) {
|
| 5688 | bitmap_set(&madvise, start, nr);
|
| 5689 | bitmap_clear(&inherit, start, nr);
|
| 5690 | bitmap_clear(&always, start, nr);
|
| 5691 | bitmap_clear(&within_size, start, nr);
|
| 5692 | } else if (!strcmp(policy, "inherit")) {
|
| 5693 | bitmap_set(&inherit, start, nr);
|
| 5694 | bitmap_clear(&madvise, start, nr);
|
| 5695 | bitmap_clear(&always, start, nr);
|
| 5696 | bitmap_clear(&within_size, start, nr);
|
| 5697 | } else if (!strcmp(policy, "within_size")) {
|
| 5698 | bitmap_set(&within_size, start, nr);
|
| 5699 | bitmap_clear(&inherit, start, nr);
|
| 5700 | bitmap_clear(&madvise, start, nr);
|
| 5701 | bitmap_clear(&always, start, nr);
|
| 5702 | } else if (!strcmp(policy, "never")) {
|
| 5703 | bitmap_clear(&inherit, start, nr);
|
| 5704 | bitmap_clear(&madvise, start, nr);
|
| 5705 | bitmap_clear(&always, start, nr);
|
| 5706 | bitmap_clear(&within_size, start, nr);
|
| 5707 | } else {
|
| 5708 | pr_err("invalid policy %s in thp_shmem boot parameter\n", policy);
|
| 5709 | goto err;
|
| 5710 | }
|
| 5711 | }
|
| 5712 | }
|
| 5713 |
|
| 5714 | huge_shmem_orders_always = always;
|
| 5715 | huge_shmem_orders_madvise = madvise;
|
| 5716 | huge_shmem_orders_inherit = inherit;
|
| 5717 | huge_shmem_orders_within_size = within_size;
|
| 5718 | shmem_orders_configured = true;
|
| 5719 | return 1;
|
| 5720 |
|
| 5721 | err:
|
| 5722 | pr_warn("thp_shmem=%s: error parsing string, ignoring setting\n", str);
|
| 5723 | return 0;
|
| 5724 | }
|
| 5725 | __setup("thp_shmem=", setup_thp_shmem);
|
| 5726 |
|
| 5727 | #endif
|
| 5728 |
|
| 5729 | #else
|
| 5730 |
|
| 5731 | |
| 5732 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code
|
| 5733 | *
|
| 5734 | * This is intended for small system where the benefits of the full
|
| 5735 | * shmem code (swap-backed and resource-limited) are outweighed by
|
| 5736 | * their complexity. On systems without swap this code should be
|
| 5737 | * effectively equivalent, but much lighter weight.
|
| 5738 |
|
| 5739 |
|
| 5740 | static struct file_system_type shmem_fs_type = {
|
| 5741 | .name = "tmpfs",
|
| 5742 | .init_fs_context = ramfs_init_fs_context,
|
| 5743 | .parameters = ramfs_fs_parameters,
|
| 5744 | .kill_sb = ramfs_kill_sb,
|
| 5745 | .fs_flags = FS_USERNS_MOUNT,
|
| 5746 | };
|
| 5747 |
|
| 5748 | void __init shmem_init(void)
|
| 5749 | {
|
| 5750 | BUG_ON(register_filesystem(&shmem_fs_type) != 0);
|
| 5751 |
|
| 5752 | shm_mnt = kern_mount(&shmem_fs_type);
|
| 5753 | BUG_ON(IS_ERR(shm_mnt));
|
| 5754 | }
|
| 5755 |
|
| 5756 | int shmem_unuse(unsigned int type)
|
| 5757 | {
|
| 5758 | return 0;
|
| 5759 | }
|
| 5760 |
|
| 5761 | int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
|
| 5762 | {
|
| 5763 | return 0;
|
| 5764 | }
|
| 5765 |
|
| 5766 | void shmem_unlock_mapping(struct address_space *mapping)
|
| 5767 | {
|
| 5768 | }
|
| 5769 |
|
| 5770 | #ifdef CONFIG_MMU
|
| 5771 | unsigned long shmem_get_unmapped_area(struct file *file,
|
| 5772 | unsigned long addr, unsigned long len,
|
| 5773 | unsigned long pgoff, unsigned long flags)
|
| 5774 | {
|
| 5775 | return mm_get_unmapped_area(file, addr, len, pgoff, flags);
|
| 5776 | }
|
| 5777 | #endif
|
| 5778 |
|
| 5779 | void shmem_truncate_range(struct inode *inode, loff_t lstart, uoff_t lend)
|
| 5780 | {
|
| 5781 | truncate_inode_pages_range(inode->i_mapping, lstart, lend);
|
| 5782 | }
|
| 5783 | EXPORT_SYMBOL_GPL(shmem_truncate_range);
|
| 5784 |
|
| 5785 | #define shmem_vm_ops generic_file_vm_ops
|
| 5786 | #define shmem_anon_vm_ops generic_file_vm_ops
|
| 5787 | #define shmem_file_operations ramfs_file_operations
|
| 5788 |
|
| 5789 | static inline int shmem_acct_size(unsigned long flags, loff_t size)
|
| 5790 | {
|
| 5791 | return 0;
|
| 5792 | }
|
| 5793 |
|
| 5794 | static inline void shmem_unacct_size(unsigned long flags, loff_t size)
|
| 5795 | {
|
| 5796 | }
|
| 5797 |
|
| 5798 | static inline struct inode *shmem_get_inode(const struct mnt_idmap *idmap,
|
| 5799 | struct super_block *sb, struct inode *dir,
|
| 5800 | umode_t mode, dev_t dev, vma_flags_t flags)
|
| 5801 | {
|
| 5802 | struct inode *inode = ramfs_get_inode(sb, dir, mode, dev);
|
| 5803 | return inode ? inode : ERR_PTR(-ENOSPC);
|
| 5804 | }
|
| 5805 |
|
| 5806 | #endif
|
| 5807 |
|
| 5808 |
|
| 5809 |
|
| 5810 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name,
|
| 5811 | loff_t size, vma_flags_t flags,
|
| 5812 | unsigned int i_flags)
|
| 5813 | {
|
| 5814 | const unsigned long shmem_flags =
|
| 5815 | vma_flags_test(&flags, VMA_NORESERVE_BIT) ? SHMEM_F_NORESERVE : 0;
|
| 5816 | struct inode *inode;
|
| 5817 | struct file *res;
|
| 5818 |
|
| 5819 | if (IS_ERR(mnt))
|
| 5820 | return ERR_CAST(mnt);
|
| 5821 |
|
| 5822 | if (size < 0 || size > MAX_LFS_FILESIZE)
|
| 5823 | return ERR_PTR(-EINVAL);
|
| 5824 |
|
| 5825 | if (is_idmapped_mnt(mnt))
|
| 5826 | return ERR_PTR(-EINVAL);
|
| 5827 |
|
| 5828 | if (shmem_acct_size(shmem_flags, size))
|
| 5829 | return ERR_PTR(-ENOMEM);
|
| 5830 |
|
| 5831 | inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
|
| 5832 | S_IFREG | S_IRWXUGO, 0, flags);
|
| 5833 | if (IS_ERR(inode)) {
|
| 5834 | shmem_unacct_size(shmem_flags, size);
|
| 5835 | return ERR_CAST(inode);
|
| 5836 | }
|
| 5837 | inode->i_flags |= i_flags;
|
| 5838 | inode->i_size = size;
|
| 5839 | clear_nlink(inode);
|
| 5840 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
|
| 5841 | if (!IS_ERR(res))
|
| 5842 | res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
|
| 5843 | &shmem_file_operations);
|
| 5844 | if (IS_ERR(res))
|
| 5845 | iput(inode);
|
| 5846 | return res;
|
| 5847 | }
|
| 5848 |
|
| 5849 | |
| 5850 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
|
| 5851 | * kernel internal. There will be NO LSM permission checks against the
|
| 5852 | * underlying inode. So users of this interface must do LSM checks at a
|
| 5853 | * higher layer. The users are the big_key and shm implementations. LSM
|
| 5854 | * checks are provided at the key or shm level rather than the inode.
|
| 5855 | * @name: name for dentry (to be seen in /proc/<pid>/maps)
|
| 5856 | * @size: size to be set for the file
|
| 5857 | * @flags: VMA_NORESERVE_BIT suppresses pre-accounting of the entire object size
|
| 5858 |
|
| 5859 | struct file *shmem_kernel_file_setup(const char *name, loff_t size,
|
| 5860 | vma_flags_t flags)
|
| 5861 | {
|
| 5862 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
|
| 5863 | }
|
| 5864 | EXPORT_SYMBOL_GPL(shmem_kernel_file_setup);
|
| 5865 |
|
| 5866 | |
| 5867 | * shmem_file_setup - get an unlinked file living in tmpfs
|
| 5868 | * @name: name for dentry (to be seen in /proc/<pid>/maps)
|
| 5869 | * @size: size to be set for the file
|
| 5870 | * @flags: VMA_NORESERVE_BIT suppresses pre-accounting of the entire object size
|
| 5871 |
|
| 5872 | struct file *shmem_file_setup(const char *name, loff_t size, vma_flags_t flags)
|
| 5873 | {
|
| 5874 | return __shmem_file_setup(shm_mnt, name, size, flags, 0);
|
| 5875 | }
|
| 5876 | EXPORT_SYMBOL_GPL(shmem_file_setup);
|
| 5877 |
|
| 5878 | |
| 5879 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
|
| 5880 | * @mnt: the tmpfs mount where the file will be created
|
| 5881 | * @name: name for dentry (to be seen in /proc/<pid>/maps)
|
| 5882 | * @size: size to be set for the file
|
| 5883 | * @flags: VMA_NORESERVE_BIT suppresses pre-accounting of the entire object size
|
| 5884 |
|
| 5885 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
|
| 5886 | loff_t size, vma_flags_t flags)
|
| 5887 | {
|
| 5888 | return __shmem_file_setup(mnt, name, size, flags, 0);
|
| 5889 | }
|
| 5890 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
|
| 5891 |
|
| 5892 | static struct file *__shmem_zero_setup(unsigned long start, unsigned long end,
|
| 5893 | vma_flags_t flags)
|
| 5894 | {
|
| 5895 | loff_t size = end - start;
|
| 5896 |
|
| 5897 | |
| 5898 | * Cloning a new file under mmap_lock leads to a lock ordering conflict
|
| 5899 | * between XFS directory reading and selinux: since this file is only
|
| 5900 | * accessible to the user through its mapping, use S_PRIVATE flag to
|
| 5901 | * bypass file security, in the same way as shmem_kernel_file_setup().
|
| 5902 |
|
| 5903 | return shmem_kernel_file_setup("dev/zero", size, flags);
|
| 5904 | }
|
| 5905 |
|
| 5906 | |
| 5907 | * shmem_zero_setup - setup a shared anonymous mapping
|
| 5908 | * @vma: the vma to be mmapped is prepared by do_mmap
|
| 5909 | * Returns: 0 on success, or error
|
| 5910 |
|
| 5911 | int shmem_zero_setup(struct vm_area_struct *vma)
|
| 5912 | {
|
| 5913 | struct file *file = __shmem_zero_setup(vma->vm_start, vma->vm_end, vma->flags);
|
| 5914 |
|
| 5915 | if (IS_ERR(file))
|
| 5916 | return PTR_ERR(file);
|
| 5917 |
|
| 5918 | if (vma->vm_file)
|
| 5919 | fput(vma->vm_file);
|
| 5920 | vma->vm_file = file;
|
| 5921 | vma->vm_ops = &shmem_anon_vm_ops;
|
| 5922 |
|
| 5923 | return 0;
|
| 5924 | }
|
| 5925 |
|
| 5926 | |
| 5927 | * shmem_zero_setup_desc - same as shmem_zero_setup, but determined by VMA
|
| 5928 | * descriptor for convenience.
|
| 5929 | * @desc: Describes VMA
|
| 5930 | * Returns: 0 on success, or error
|
| 5931 |
|
| 5932 | int shmem_zero_setup_desc(struct vm_area_desc *desc)
|
| 5933 | {
|
| 5934 | struct file *file = __shmem_zero_setup(desc->start, desc->end, desc->vma_flags);
|
| 5935 |
|
| 5936 | if (IS_ERR(file))
|
| 5937 | return PTR_ERR(file);
|
| 5938 |
|
| 5939 | desc->vm_file = file;
|
| 5940 | desc->vm_ops = &shmem_anon_vm_ops;
|
| 5941 |
|
| 5942 | return 0;
|
| 5943 | }
|
| 5944 |
|
| 5945 | |
| 5946 | * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
|
| 5947 | * @mapping: the folio's address_space
|
| 5948 | * @index: the folio index
|
| 5949 | * @gfp: the page allocator flags to use if allocating
|
| 5950 | *
|
| 5951 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
|
| 5952 | * with any new page allocations done using the specified allocation flags.
|
| 5953 | * But read_cache_page_gfp() uses the ->read_folio() method: which does not
|
| 5954 | * suit tmpfs, since it may have pages in swapcache, and needs to find those
|
| 5955 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
|
| 5956 | *
|
| 5957 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
|
| 5958 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
|
| 5959 |
|
| 5960 | struct folio *shmem_read_folio_gfp(struct address_space *mapping,
|
| 5961 | pgoff_t index, gfp_t gfp)
|
| 5962 | {
|
| 5963 | #ifdef CONFIG_SHMEM
|
| 5964 | struct inode *inode = mapping->host;
|
| 5965 | struct folio *folio;
|
| 5966 | int error;
|
| 5967 |
|
| 5968 | error = shmem_get_folio_gfp(inode, index, i_size_read(inode),
|
| 5969 | &folio, SGP_CACHE, gfp, NULL, NULL);
|
| 5970 | if (error)
|
| 5971 | return ERR_PTR(error);
|
| 5972 |
|
| 5973 | folio_unlock(folio);
|
| 5974 | return folio;
|
| 5975 | #else
|
| 5976 | |
| 5977 | * The tiny !SHMEM case uses ramfs without swap
|
| 5978 |
|
| 5979 | return mapping_read_folio_gfp(mapping, index, gfp);
|
| 5980 | #endif
|
| 5981 | }
|
| 5982 | EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
|
| 5983 |
|
| 5984 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
|
| 5985 | pgoff_t index, gfp_t gfp)
|
| 5986 | {
|
| 5987 | struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
|
| 5988 | struct page *page;
|
| 5989 |
|
| 5990 | if (IS_ERR(folio))
|
| 5991 | return &folio->page;
|
| 5992 |
|
| 5993 | page = folio_file_page(folio, index);
|
| 5994 | if (PageHWPoison(page)) {
|
| 5995 | folio_put(folio);
|
| 5996 | return ERR_PTR(-EIO);
|
| 5997 | }
|
| 5998 |
|
| 5999 | return page;
|
| 6000 | }
|
| 6001 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
|