| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
/* |
| 3 |
* Simple file system for zoned block devices exposing zones as files. |
| 4 |
* |
| 5 |
* Copyright (C) 2019 Western Digital Corporation or its affiliates. |
| 6 |
*/ |
| 7 |
#include <linux/module.h> |
| 8 |
#include <linux/pagemap.h> |
| 9 |
#include <linux/magic.h> |
| 10 |
#include <linux/iomap.h> |
| 11 |
#include <linux/init.h> |
| 12 |
#include <linux/slab.h> |
| 13 |
#include <linux/blkdev.h> |
| 14 |
#include <linux/statfs.h> |
| 15 |
#include <linux/writeback.h> |
| 16 |
#include <linux/quotaops.h> |
| 17 |
#include <linux/seq_file.h> |
| 18 |
#include <linux/uio.h> |
| 19 |
#include <linux/mman.h> |
| 20 |
#include <linux/sched/mm.h> |
| 21 |
#include <linux/crc32.h> |
| 22 |
#include <linux/task_io_accounting_ops.h> |
| 23 |
#include <linux/fs_parser.h> |
| 24 |
#include <linux/fs_context.h> |
| 25 |
|
| 26 |
#include "zonefs.h" |
| 27 |
|
| 28 |
#define CREATE_TRACE_POINTS |
| 29 |
#include "trace.h" |
| 30 |
|
| 31 |
/* |
| 32 |
* Get the name of a zone group directory. |
| 33 |
*/ |
| 34 |
static const char *zonefs_zgroup_name(enum zonefs_ztype ztype) |
| 35 |
{ |
| 36 |
switch (ztype) { |
| 37 |
case ZONEFS_ZTYPE_CNV: |
| 38 |
return "cnv"; |
| 39 |
case ZONEFS_ZTYPE_SEQ: |
| 40 |
return "seq"; |
| 41 |
default: |
| 42 |
WARN_ON_ONCE(1); |
| 43 |
return "???"; |
| 44 |
} |
| 45 |
} |
| 46 |
|
| 47 |
/* |
| 48 |
* Manage the active zone count. |
| 49 |
*/ |
| 50 |
static void zonefs_account_active(struct super_block *sb, |
| 51 |
struct zonefs_zone *z) |
| 52 |
{ |
| 53 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 54 |
|
| 55 |
if (zonefs_zone_is_cnv(z)) |
| 56 |
return; |
| 57 |
|
| 58 |
/* |
| 59 |
* For zones that transitioned to the offline or readonly condition, |
| 60 |
* we only need to clear the active state. |
| 61 |
*/ |
| 62 |
if (z->z_flags & (ZONEFS_ZONE_OFFLINE | ZONEFS_ZONE_READONLY)) |
| 63 |
goto out; |
| 64 |
|
| 65 |
/* |
| 66 |
* If the zone is active, that is, if it is explicitly open or |
| 67 |
* partially written, check if it was already accounted as active. |
| 68 |
*/ |
| 69 |
if ((z->z_flags & ZONEFS_ZONE_OPEN) || |
| 70 |
(z->z_wpoffset > 0 && z->z_wpoffset < z->z_capacity)) { |
| 71 |
if (!(z->z_flags & ZONEFS_ZONE_ACTIVE)) { |
| 72 |
z->z_flags |= ZONEFS_ZONE_ACTIVE; |
| 73 |
atomic_inc(&sbi->s_active_seq_files); |
| 74 |
} |
| 75 |
return; |
| 76 |
} |
| 77 |
|
| 78 |
out: |
| 79 |
/* The zone is not active. If it was, update the active count */ |
| 80 |
if (z->z_flags & ZONEFS_ZONE_ACTIVE) { |
| 81 |
z->z_flags &= ~ZONEFS_ZONE_ACTIVE; |
| 82 |
atomic_dec(&sbi->s_active_seq_files); |
| 83 |
} |
| 84 |
} |
| 85 |
|
| 86 |
/* |
| 87 |
* Manage the active zone count. Called with zi->i_truncate_mutex held. |
| 88 |
*/ |
| 89 |
void zonefs_inode_account_active(struct inode *inode) |
| 90 |
{ |
| 91 |
lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex); |
| 92 |
|
| 93 |
return zonefs_account_active(inode->i_sb, zonefs_inode_zone(inode)); |
| 94 |
} |
| 95 |
|
| 96 |
/* |
| 97 |
* Execute a zone management operation. |
| 98 |
*/ |
| 99 |
static int zonefs_zone_mgmt(struct super_block *sb, |
| 100 |
struct zonefs_zone *z, enum req_op op) |
| 101 |
{ |
| 102 |
int ret; |
| 103 |
|
| 104 |
/* |
| 105 |
* With ZNS drives, closing an explicitly open zone that has not been |
| 106 |
* written will change the zone state to "closed", that is, the zone |
| 107 |
* will remain active. Since this can then cause failure of explicit |
| 108 |
* open operation on other zones if the drive active zone resources |
| 109 |
* are exceeded, make sure that the zone does not remain active by |
| 110 |
* resetting it. |
| 111 |
*/ |
| 112 |
if (op == REQ_OP_ZONE_CLOSE && !z->z_wpoffset) |
| 113 |
op = REQ_OP_ZONE_RESET; |
| 114 |
|
| 115 |
trace_zonefs_zone_mgmt(sb, z, op); |
| 116 |
ret = blkdev_zone_mgmt(sb->s_bdev, op, z->z_sector, |
| 117 |
z->z_size >> SECTOR_SHIFT); |
| 118 |
if (ret) { |
| 119 |
zonefs_err(sb, |
| 120 |
"Zone management operation %s at %llu failed %d\n", |
| 121 |
blk_op_str(op), z->z_sector, ret); |
| 122 |
return ret; |
| 123 |
} |
| 124 |
|
| 125 |
return 0; |
| 126 |
} |
| 127 |
|
| 128 |
int zonefs_inode_zone_mgmt(struct inode *inode, enum req_op op) |
| 129 |
{ |
| 130 |
lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex); |
| 131 |
|
| 132 |
return zonefs_zone_mgmt(inode->i_sb, zonefs_inode_zone(inode), op); |
| 133 |
} |
| 134 |
|
| 135 |
void zonefs_i_size_write(struct inode *inode, loff_t isize) |
| 136 |
{ |
| 137 |
struct zonefs_zone *z = zonefs_inode_zone(inode); |
| 138 |
|
| 139 |
i_size_write(inode, isize); |
| 140 |
|
| 141 |
/* |
| 142 |
* A full zone is no longer open/active and does not need |
| 143 |
* explicit closing. |
| 144 |
*/ |
| 145 |
if (isize >= z->z_capacity) { |
| 146 |
struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb); |
| 147 |
|
| 148 |
if (z->z_flags & ZONEFS_ZONE_ACTIVE) |
| 149 |
atomic_dec(&sbi->s_active_seq_files); |
| 150 |
z->z_flags &= ~(ZONEFS_ZONE_OPEN | ZONEFS_ZONE_ACTIVE); |
| 151 |
} |
| 152 |
} |
| 153 |
|
| 154 |
void zonefs_update_stats(struct inode *inode, loff_t new_isize) |
| 155 |
{ |
| 156 |
struct super_block *sb = inode->i_sb; |
| 157 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 158 |
loff_t old_isize = i_size_read(inode); |
| 159 |
loff_t nr_blocks; |
| 160 |
|
| 161 |
if (new_isize == old_isize) |
| 162 |
return; |
| 163 |
|
| 164 |
spin_lock(&sbi->s_lock); |
| 165 |
|
| 166 |
/* |
| 167 |
* This may be called for an update after an IO error. |
| 168 |
* So beware of the values seen. |
| 169 |
*/ |
| 170 |
if (new_isize < old_isize) { |
| 171 |
nr_blocks = (old_isize - new_isize) >> sb->s_blocksize_bits; |
| 172 |
if (sbi->s_used_blocks > nr_blocks) |
| 173 |
sbi->s_used_blocks -= nr_blocks; |
| 174 |
else |
| 175 |
sbi->s_used_blocks = 0; |
| 176 |
} else { |
| 177 |
sbi->s_used_blocks += |
| 178 |
(new_isize - old_isize) >> sb->s_blocksize_bits; |
| 179 |
if (sbi->s_used_blocks > sbi->s_blocks) |
| 180 |
sbi->s_used_blocks = sbi->s_blocks; |
| 181 |
} |
| 182 |
|
| 183 |
spin_unlock(&sbi->s_lock); |
| 184 |
} |
| 185 |
|
| 186 |
/* |
| 187 |
* Check a zone condition. Return the amount of written (and still readable) |
| 188 |
* data in the zone. |
| 189 |
*/ |
| 190 |
static loff_t zonefs_check_zone_condition(struct super_block *sb, |
| 191 |
struct zonefs_zone *z, |
| 192 |
struct blk_zone *zone) |
| 193 |
{ |
| 194 |
switch (zone->cond) { |
| 195 |
case BLK_ZONE_COND_OFFLINE: |
| 196 |
zonefs_warn(sb, "Zone %llu: offline zone\n", |
| 197 |
z->z_sector); |
| 198 |
z->z_flags |= ZONEFS_ZONE_OFFLINE; |
| 199 |
return 0; |
| 200 |
case BLK_ZONE_COND_READONLY: |
| 201 |
/* |
| 202 |
* The write pointer of read-only zones is invalid, so we cannot |
| 203 |
* determine the zone wpoffset (inode size). We thus keep the |
| 204 |
* zone wpoffset as is, which leads to an empty file |
| 205 |
* (wpoffset == 0) on mount. For a runtime error, this keeps |
| 206 |
* the inode size as it was when last updated so that the user |
| 207 |
* can recover data. |
| 208 |
*/ |
| 209 |
zonefs_warn(sb, "Zone %llu: read-only zone\n", |
| 210 |
z->z_sector); |
| 211 |
z->z_flags |= ZONEFS_ZONE_READONLY; |
| 212 |
if (zonefs_zone_is_cnv(z)) |
| 213 |
return z->z_capacity; |
| 214 |
return z->z_wpoffset; |
| 215 |
case BLK_ZONE_COND_FULL: |
| 216 |
/* The write pointer of full zones is invalid. */ |
| 217 |
return z->z_capacity; |
| 218 |
default: |
| 219 |
if (zonefs_zone_is_cnv(z)) |
| 220 |
return z->z_capacity; |
| 221 |
return (zone->wp - zone->start) << SECTOR_SHIFT; |
| 222 |
} |
| 223 |
} |
| 224 |
|
| 225 |
/* |
| 226 |
* Check a zone condition and adjust its inode access permissions for |
| 227 |
* offline and readonly zones. |
| 228 |
*/ |
| 229 |
static void zonefs_inode_update_mode(struct inode *inode) |
| 230 |
{ |
| 231 |
struct zonefs_zone *z = zonefs_inode_zone(inode); |
| 232 |
|
| 233 |
if (z->z_flags & ZONEFS_ZONE_OFFLINE) { |
| 234 |
/* Offline zones cannot be read nor written */ |
| 235 |
inode->i_flags |= S_IMMUTABLE; |
| 236 |
inode->i_mode &= ~0777; |
| 237 |
} else if (z->z_flags & ZONEFS_ZONE_READONLY) { |
| 238 |
/* Readonly zones cannot be written */ |
| 239 |
inode->i_flags |= S_IMMUTABLE; |
| 240 |
if (z->z_flags & ZONEFS_ZONE_INIT_MODE) |
| 241 |
inode->i_mode &= ~0777; |
| 242 |
else |
| 243 |
inode->i_mode &= ~0222; |
| 244 |
} |
| 245 |
|
| 246 |
z->z_flags &= ~ZONEFS_ZONE_INIT_MODE; |
| 247 |
z->z_mode = inode->i_mode; |
| 248 |
} |
| 249 |
|
| 250 |
static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx, |
| 251 |
void *data) |
| 252 |
{ |
| 253 |
struct blk_zone *z = data; |
| 254 |
|
| 255 |
*z = *zone; |
| 256 |
return 0; |
| 257 |
} |
| 258 |
|
| 259 |
static void zonefs_handle_io_error(struct inode *inode, struct blk_zone *zone, |
| 260 |
bool write) |
| 261 |
{ |
| 262 |
struct zonefs_zone *z = zonefs_inode_zone(inode); |
| 263 |
struct super_block *sb = inode->i_sb; |
| 264 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 265 |
loff_t isize, data_size; |
| 266 |
|
| 267 |
/* |
| 268 |
* Check the zone condition: if the zone is not "bad" (offline or |
| 269 |
* read-only), read errors are simply signaled to the IO issuer as long |
| 270 |
* as there is no inconsistency between the inode size and the amount of |
| 271 |
* data written in the zone (data_size). |
| 272 |
*/ |
| 273 |
data_size = zonefs_check_zone_condition(sb, z, zone); |
| 274 |
isize = i_size_read(inode); |
| 275 |
if (!(z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE)) && |
| 276 |
!write && isize == data_size) |
| 277 |
return; |
| 278 |
|
| 279 |
/* |
| 280 |
* At this point, we detected either a bad zone or an inconsistency |
| 281 |
* between the inode size and the amount of data written in the zone. |
| 282 |
* For the latter case, the cause may be a write IO error or an external |
| 283 |
* action on the device. Two error patterns exist: |
| 284 |
* 1) The inode size is lower than the amount of data in the zone: |
| 285 |
* a write operation partially failed and data was written at the end |
| 286 |
* of the file. This can happen in the case of a large direct IO |
| 287 |
* needing several BIOs and/or write requests to be processed. |
| 288 |
* 2) The inode size is larger than the amount of data in the zone: |
| 289 |
* this can happen with a deferred write error with the use of the |
| 290 |
* device side write cache after getting successful write IO |
| 291 |
* completions. Other possibilities are (a) an external corruption, |
| 292 |
* e.g. an application reset the zone directly, or (b) the device |
| 293 |
* has a serious problem (e.g. firmware bug). |
| 294 |
* |
| 295 |
* In all cases, warn about inode size inconsistency and handle the |
| 296 |
* IO error according to the zone condition and to the mount options. |
| 297 |
*/ |
| 298 |
if (isize != data_size) |
| 299 |
zonefs_warn(sb, |
| 300 |
"inode %lu: invalid size %lld (should be %lld)\n", |
| 301 |
inode->i_ino, isize, data_size); |
| 302 |
|
| 303 |
/* |
| 304 |
* First handle bad zones signaled by hardware. The mount options |
| 305 |
* errors=zone-ro and errors=zone-offline result in changing the |
| 306 |
* zone condition to read-only and offline respectively, as if the |
| 307 |
* condition was signaled by the hardware. |
| 308 |
*/ |
| 309 |
if ((z->z_flags & ZONEFS_ZONE_OFFLINE) || |
| 310 |
(sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)) { |
| 311 |
zonefs_warn(sb, "inode %lu: read/write access disabled\n", |
| 312 |
inode->i_ino); |
| 313 |
if (!(z->z_flags & ZONEFS_ZONE_OFFLINE)) |
| 314 |
z->z_flags |= ZONEFS_ZONE_OFFLINE; |
| 315 |
zonefs_inode_update_mode(inode); |
| 316 |
data_size = 0; |
| 317 |
} else if ((z->z_flags & ZONEFS_ZONE_READONLY) || |
| 318 |
(sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)) { |
| 319 |
zonefs_warn(sb, "inode %lu: write access disabled\n", |
| 320 |
inode->i_ino); |
| 321 |
if (!(z->z_flags & ZONEFS_ZONE_READONLY)) |
| 322 |
z->z_flags |= ZONEFS_ZONE_READONLY; |
| 323 |
zonefs_inode_update_mode(inode); |
| 324 |
data_size = isize; |
| 325 |
} else if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO && |
| 326 |
data_size > isize) { |
| 327 |
/* Do not expose garbage data */ |
| 328 |
data_size = isize; |
| 329 |
} |
| 330 |
|
| 331 |
/* |
| 332 |
* If the filesystem is mounted with the explicit-open mount option, we |
| 333 |
* need to clear the ZONEFS_ZONE_OPEN flag if the zone transitioned to |
| 334 |
* the read-only or offline condition, to avoid attempting an explicit |
| 335 |
* close of the zone when the inode file is closed. |
| 336 |
*/ |
| 337 |
if ((sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) && |
| 338 |
(z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE))) |
| 339 |
z->z_flags &= ~ZONEFS_ZONE_OPEN; |
| 340 |
|
| 341 |
/* |
| 342 |
* If error=remount-ro was specified, any error result in remounting |
| 343 |
* the volume as read-only. |
| 344 |
*/ |
| 345 |
if ((sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) && !sb_rdonly(sb)) { |
| 346 |
zonefs_warn(sb, "remounting filesystem read-only\n"); |
| 347 |
sb->s_flags |= SB_RDONLY; |
| 348 |
} |
| 349 |
|
| 350 |
/* |
| 351 |
* Update block usage stats and the inode size to prevent access to |
| 352 |
* invalid data. |
| 353 |
*/ |
| 354 |
zonefs_update_stats(inode, data_size); |
| 355 |
zonefs_i_size_write(inode, data_size); |
| 356 |
z->z_wpoffset = data_size; |
| 357 |
zonefs_inode_account_active(inode); |
| 358 |
} |
| 359 |
|
| 360 |
/* |
| 361 |
* When an file IO error occurs, check the file zone to see if there is a change |
| 362 |
* in the zone condition (e.g. offline or read-only). For a failed write to a |
| 363 |
* sequential zone, the zone write pointer position must also be checked to |
| 364 |
* eventually correct the file size and zonefs inode write pointer offset |
| 365 |
* (which can be out of sync with the drive due to partial write failures). |
| 366 |
*/ |
| 367 |
void __zonefs_io_error(struct inode *inode, bool write) |
| 368 |
{ |
| 369 |
struct zonefs_zone *z = zonefs_inode_zone(inode); |
| 370 |
struct super_block *sb = inode->i_sb; |
| 371 |
unsigned int noio_flag; |
| 372 |
struct blk_zone zone; |
| 373 |
int ret; |
| 374 |
|
| 375 |
/* |
| 376 |
* Conventional zone have no write pointer and cannot become read-only |
| 377 |
* or offline. So simply fake a report for a single or aggregated zone |
| 378 |
* and let zonefs_handle_io_error() correct the zone inode information |
| 379 |
* according to the mount options. |
| 380 |
*/ |
| 381 |
if (!zonefs_zone_is_seq(z)) { |
| 382 |
zone.start = z->z_sector; |
| 383 |
zone.len = z->z_size >> SECTOR_SHIFT; |
| 384 |
zone.wp = zone.start + zone.len; |
| 385 |
zone.type = BLK_ZONE_TYPE_CONVENTIONAL; |
| 386 |
zone.cond = BLK_ZONE_COND_NOT_WP; |
| 387 |
zone.capacity = zone.len; |
| 388 |
goto handle_io_error; |
| 389 |
} |
| 390 |
|
| 391 |
/* |
| 392 |
* Memory allocations in blkdev_report_zones() can trigger a memory |
| 393 |
* reclaim which may in turn cause a recursion into zonefs as well as |
| 394 |
* struct request allocations for the same device. The former case may |
| 395 |
* end up in a deadlock on the inode truncate mutex, while the latter |
| 396 |
* may prevent IO forward progress. Executing the report zones under |
| 397 |
* the GFP_NOIO context avoids both problems. |
| 398 |
*/ |
| 399 |
noio_flag = memalloc_noio_save(); |
| 400 |
ret = blkdev_report_zones(sb->s_bdev, z->z_sector, 1, |
| 401 |
zonefs_io_error_cb, &zone); |
| 402 |
memalloc_noio_restore(noio_flag); |
| 403 |
|
| 404 |
if (ret != 1) { |
| 405 |
zonefs_err(sb, "Get inode %lu zone information failed %d\n", |
| 406 |
inode->i_ino, ret); |
| 407 |
zonefs_warn(sb, "remounting filesystem read-only\n"); |
| 408 |
sb->s_flags |= SB_RDONLY; |
| 409 |
return; |
| 410 |
} |
| 411 |
|
| 412 |
handle_io_error: |
| 413 |
zonefs_handle_io_error(inode, &zone, write); |
| 414 |
} |
| 415 |
|
| 416 |
static struct kmem_cache *zonefs_inode_cachep; |
| 417 |
|
| 418 |
static struct inode *zonefs_alloc_inode(struct super_block *sb) |
| 419 |
{ |
| 420 |
struct zonefs_inode_info *zi; |
| 421 |
|
| 422 |
zi = alloc_inode_sb(sb, zonefs_inode_cachep, GFP_KERNEL); |
| 423 |
if (!zi) |
| 424 |
return NULL; |
| 425 |
|
| 426 |
inode_init_once(&zi->i_vnode); |
| 427 |
mutex_init(&zi->i_truncate_mutex); |
| 428 |
zi->i_wr_refcnt = 0; |
| 429 |
|
| 430 |
return &zi->i_vnode; |
| 431 |
} |
| 432 |
|
| 433 |
static void zonefs_free_inode(struct inode *inode) |
| 434 |
{ |
| 435 |
kmem_cache_free(zonefs_inode_cachep, ZONEFS_I(inode)); |
| 436 |
} |
| 437 |
|
| 438 |
/* |
| 439 |
* File system stat. |
| 440 |
*/ |
| 441 |
static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf) |
| 442 |
{ |
| 443 |
struct super_block *sb = dentry->d_sb; |
| 444 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 445 |
enum zonefs_ztype t; |
| 446 |
|
| 447 |
buf->f_type = ZONEFS_MAGIC; |
| 448 |
buf->f_bsize = sb->s_blocksize; |
| 449 |
buf->f_namelen = ZONEFS_NAME_MAX; |
| 450 |
|
| 451 |
spin_lock(&sbi->s_lock); |
| 452 |
|
| 453 |
buf->f_blocks = sbi->s_blocks; |
| 454 |
if (WARN_ON(sbi->s_used_blocks > sbi->s_blocks)) |
| 455 |
buf->f_bfree = 0; |
| 456 |
else |
| 457 |
buf->f_bfree = buf->f_blocks - sbi->s_used_blocks; |
| 458 |
buf->f_bavail = buf->f_bfree; |
| 459 |
|
| 460 |
for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) { |
| 461 |
if (sbi->s_zgroup[t].g_nr_zones) |
| 462 |
buf->f_files += sbi->s_zgroup[t].g_nr_zones + 1; |
| 463 |
} |
| 464 |
buf->f_ffree = 0; |
| 465 |
|
| 466 |
spin_unlock(&sbi->s_lock); |
| 467 |
|
| 468 |
buf->f_fsid = uuid_to_fsid(sbi->s_uuid.b); |
| 469 |
|
| 470 |
return 0; |
| 471 |
} |
| 472 |
|
| 473 |
enum { |
| 474 |
Opt_errors, Opt_explicit_open, |
| 475 |
}; |
| 476 |
|
| 477 |
struct zonefs_context { |
| 478 |
unsigned long s_mount_opts; |
| 479 |
}; |
| 480 |
|
| 481 |
static const struct constant_table zonefs_param_errors[] = { |
| 482 |
{"remount-ro", ZONEFS_MNTOPT_ERRORS_RO}, |
| 483 |
{"zone-ro", ZONEFS_MNTOPT_ERRORS_ZRO}, |
| 484 |
{"zone-offline", ZONEFS_MNTOPT_ERRORS_ZOL}, |
| 485 |
{"repair", ZONEFS_MNTOPT_ERRORS_REPAIR}, |
| 486 |
{} |
| 487 |
}; |
| 488 |
|
| 489 |
static const struct fs_parameter_spec zonefs_param_spec[] = { |
| 490 |
fsparam_enum ("errors", Opt_errors, zonefs_param_errors), |
| 491 |
fsparam_flag ("explicit-open", Opt_explicit_open), |
| 492 |
{} |
| 493 |
}; |
| 494 |
|
| 495 |
static int zonefs_parse_param(struct fs_context *fc, struct fs_parameter *param) |
| 496 |
{ |
| 497 |
struct zonefs_context *ctx = fc->fs_private; |
| 498 |
struct fs_parse_result result; |
| 499 |
int opt; |
| 500 |
|
| 501 |
opt = fs_parse(fc, zonefs_param_spec, param, &result); |
| 502 |
if (opt < 0) |
| 503 |
return opt; |
| 504 |
|
| 505 |
switch (opt) { |
| 506 |
case Opt_errors: |
| 507 |
ctx->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK; |
| 508 |
ctx->s_mount_opts |= result.uint_32; |
| 509 |
break; |
| 510 |
case Opt_explicit_open: |
| 511 |
ctx->s_mount_opts |= ZONEFS_MNTOPT_EXPLICIT_OPEN; |
| 512 |
break; |
| 513 |
default: |
| 514 |
return -EINVAL; |
| 515 |
} |
| 516 |
|
| 517 |
return 0; |
| 518 |
} |
| 519 |
|
| 520 |
static int zonefs_show_options(struct seq_file *seq, struct dentry *root) |
| 521 |
{ |
| 522 |
struct zonefs_sb_info *sbi = ZONEFS_SB(root->d_sb); |
| 523 |
|
| 524 |
if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) |
| 525 |
seq_puts(seq, ",errors=remount-ro"); |
| 526 |
if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO) |
| 527 |
seq_puts(seq, ",errors=zone-ro"); |
| 528 |
if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL) |
| 529 |
seq_puts(seq, ",errors=zone-offline"); |
| 530 |
if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_REPAIR) |
| 531 |
seq_puts(seq, ",errors=repair"); |
| 532 |
|
| 533 |
return 0; |
| 534 |
} |
| 535 |
|
| 536 |
static int zonefs_inode_setattr(struct mnt_idmap *idmap, |
| 537 |
struct dentry *dentry, struct iattr *iattr) |
| 538 |
{ |
| 539 |
struct inode *inode = d_inode(dentry); |
| 540 |
int ret; |
| 541 |
|
| 542 |
if (unlikely(IS_IMMUTABLE(inode))) |
| 543 |
return -EPERM; |
| 544 |
|
| 545 |
ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr); |
| 546 |
if (ret) |
| 547 |
return ret; |
| 548 |
|
| 549 |
/* |
| 550 |
* Since files and directories cannot be created nor deleted, do not |
| 551 |
* allow setting any write attributes on the sub-directories grouping |
| 552 |
* files by zone type. |
| 553 |
*/ |
| 554 |
if ((iattr->ia_valid & ATTR_MODE) && S_ISDIR(inode->i_mode) && |
| 555 |
(iattr->ia_mode & 0222)) |
| 556 |
return -EPERM; |
| 557 |
|
| 558 |
if (((iattr->ia_valid & ATTR_UID) && |
| 559 |
!uid_eq(iattr->ia_uid, inode->i_uid)) || |
| 560 |
((iattr->ia_valid & ATTR_GID) && |
| 561 |
!gid_eq(iattr->ia_gid, inode->i_gid))) { |
| 562 |
ret = dquot_transfer(&nop_mnt_idmap, inode, iattr); |
| 563 |
if (ret) |
| 564 |
return ret; |
| 565 |
} |
| 566 |
|
| 567 |
if (iattr->ia_valid & ATTR_SIZE) { |
| 568 |
ret = zonefs_file_truncate(inode, iattr->ia_size); |
| 569 |
if (ret) |
| 570 |
return ret; |
| 571 |
} |
| 572 |
|
| 573 |
setattr_copy(&nop_mnt_idmap, inode, iattr); |
| 574 |
|
| 575 |
if (S_ISREG(inode->i_mode)) { |
| 576 |
struct zonefs_zone *z = zonefs_inode_zone(inode); |
| 577 |
|
| 578 |
z->z_mode = inode->i_mode; |
| 579 |
z->z_uid = inode->i_uid; |
| 580 |
z->z_gid = inode->i_gid; |
| 581 |
} |
| 582 |
|
| 583 |
return 0; |
| 584 |
} |
| 585 |
|
| 586 |
static const struct inode_operations zonefs_file_inode_operations = { |
| 587 |
.setattr = zonefs_inode_setattr, |
| 588 |
}; |
| 589 |
|
| 590 |
static long zonefs_fname_to_fno(const struct qstr *fname) |
| 591 |
{ |
| 592 |
const char *name = fname->name; |
| 593 |
unsigned int len = fname->len; |
| 594 |
long fno = 0, shift = 1; |
| 595 |
const char *rname; |
| 596 |
char c = *name; |
| 597 |
unsigned int i; |
| 598 |
|
| 599 |
/* |
| 600 |
* File names are always a base-10 number string without any |
| 601 |
* leading 0s. |
| 602 |
*/ |
| 603 |
if (!isdigit(c)) |
| 604 |
return -ENOENT; |
| 605 |
|
| 606 |
if (len > 1 && c == '0') |
| 607 |
return -ENOENT; |
| 608 |
|
| 609 |
if (len == 1) |
| 610 |
return c - '0'; |
| 611 |
|
| 612 |
for (i = 0, rname = name + len - 1; i < len; i++, rname--) { |
| 613 |
c = *rname; |
| 614 |
if (!isdigit(c)) |
| 615 |
return -ENOENT; |
| 616 |
fno += (c - '0') * shift; |
| 617 |
shift *= 10; |
| 618 |
} |
| 619 |
|
| 620 |
return fno; |
| 621 |
} |
| 622 |
|
| 623 |
static struct inode *zonefs_get_file_inode(struct inode *dir, |
| 624 |
struct dentry *dentry) |
| 625 |
{ |
| 626 |
struct zonefs_zone_group *zgroup = dir->i_private; |
| 627 |
struct super_block *sb = dir->i_sb; |
| 628 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 629 |
struct zonefs_zone *z; |
| 630 |
struct inode *inode; |
| 631 |
ino_t ino; |
| 632 |
long fno; |
| 633 |
|
| 634 |
/* Get the file number from the file name */ |
| 635 |
fno = zonefs_fname_to_fno(&dentry->d_name); |
| 636 |
if (fno < 0) |
| 637 |
return ERR_PTR(fno); |
| 638 |
|
| 639 |
if (!zgroup->g_nr_zones || fno >= zgroup->g_nr_zones) |
| 640 |
return ERR_PTR(-ENOENT); |
| 641 |
|
| 642 |
z = &zgroup->g_zones[fno]; |
| 643 |
ino = z->z_sector >> sbi->s_zone_sectors_shift; |
| 644 |
inode = iget_locked(sb, ino); |
| 645 |
if (!inode) |
| 646 |
return ERR_PTR(-ENOMEM); |
| 647 |
if (!(inode_state_read_once(inode) & I_NEW)) { |
| 648 |
WARN_ON_ONCE(inode->i_private != z); |
| 649 |
return inode; |
| 650 |
} |
| 651 |
|
| 652 |
inode->i_ino = ino; |
| 653 |
inode->i_mode = z->z_mode; |
| 654 |
inode_set_mtime_to_ts(inode, |
| 655 |
inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, inode_get_ctime(dir)))); |
| 656 |
inode->i_uid = z->z_uid; |
| 657 |
inode->i_gid = z->z_gid; |
| 658 |
inode->i_size = z->z_wpoffset; |
| 659 |
inode->i_blocks = z->z_capacity >> SECTOR_SHIFT; |
| 660 |
inode->i_private = z; |
| 661 |
|
| 662 |
inode->i_op = &zonefs_file_inode_operations; |
| 663 |
inode->i_fop = &zonefs_file_operations; |
| 664 |
inode->i_mapping->a_ops = &zonefs_file_aops; |
| 665 |
mapping_set_large_folios(inode->i_mapping); |
| 666 |
|
| 667 |
/* Update the inode access rights depending on the zone condition */ |
| 668 |
zonefs_inode_update_mode(inode); |
| 669 |
|
| 670 |
unlock_new_inode(inode); |
| 671 |
|
| 672 |
return inode; |
| 673 |
} |
| 674 |
|
| 675 |
static struct inode *zonefs_get_zgroup_inode(struct super_block *sb, |
| 676 |
enum zonefs_ztype ztype) |
| 677 |
{ |
| 678 |
struct inode *root = d_inode(sb->s_root); |
| 679 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 680 |
struct inode *inode; |
| 681 |
ino_t ino = bdev_nr_zones(sb->s_bdev) + ztype + 1; |
| 682 |
|
| 683 |
inode = iget_locked(sb, ino); |
| 684 |
if (!inode) |
| 685 |
return ERR_PTR(-ENOMEM); |
| 686 |
if (!(inode_state_read_once(inode) & I_NEW)) |
| 687 |
return inode; |
| 688 |
|
| 689 |
inode->i_ino = ino; |
| 690 |
inode_init_owner(&nop_mnt_idmap, inode, root, S_IFDIR | 0555); |
| 691 |
inode->i_size = sbi->s_zgroup[ztype].g_nr_zones; |
| 692 |
inode_set_mtime_to_ts(inode, |
| 693 |
inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, inode_get_ctime(root)))); |
| 694 |
inode->i_private = &sbi->s_zgroup[ztype]; |
| 695 |
set_nlink(inode, 2); |
| 696 |
|
| 697 |
inode->i_op = &zonefs_dir_inode_operations; |
| 698 |
inode->i_fop = &zonefs_dir_operations; |
| 699 |
|
| 700 |
unlock_new_inode(inode); |
| 701 |
|
| 702 |
return inode; |
| 703 |
} |
| 704 |
|
| 705 |
|
| 706 |
static struct inode *zonefs_get_dir_inode(struct inode *dir, |
| 707 |
struct dentry *dentry) |
| 708 |
{ |
| 709 |
struct super_block *sb = dir->i_sb; |
| 710 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 711 |
const char *name = dentry->d_name.name; |
| 712 |
enum zonefs_ztype ztype; |
| 713 |
|
| 714 |
/* |
| 715 |
* We only need to check for the "seq" directory and |
| 716 |
* the "cnv" directory if we have conventional zones. |
| 717 |
*/ |
| 718 |
if (dentry->d_name.len != 3) |
| 719 |
return ERR_PTR(-ENOENT); |
| 720 |
|
| 721 |
for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) { |
| 722 |
if (sbi->s_zgroup[ztype].g_nr_zones && |
| 723 |
memcmp(name, zonefs_zgroup_name(ztype), 3) == 0) |
| 724 |
break; |
| 725 |
} |
| 726 |
if (ztype == ZONEFS_ZTYPE_MAX) |
| 727 |
return ERR_PTR(-ENOENT); |
| 728 |
|
| 729 |
return zonefs_get_zgroup_inode(sb, ztype); |
| 730 |
} |
| 731 |
|
| 732 |
static struct dentry *zonefs_lookup(struct inode *dir, struct dentry *dentry, |
| 733 |
unsigned int flags) |
| 734 |
{ |
| 735 |
struct inode *inode; |
| 736 |
|
| 737 |
if (dentry->d_name.len > ZONEFS_NAME_MAX) |
| 738 |
return ERR_PTR(-ENAMETOOLONG); |
| 739 |
|
| 740 |
if (dir == d_inode(dir->i_sb->s_root)) |
| 741 |
inode = zonefs_get_dir_inode(dir, dentry); |
| 742 |
else |
| 743 |
inode = zonefs_get_file_inode(dir, dentry); |
| 744 |
|
| 745 |
return d_splice_alias(inode, dentry); |
| 746 |
} |
| 747 |
|
| 748 |
static int zonefs_readdir_root(struct file *file, struct dir_context *ctx) |
| 749 |
{ |
| 750 |
struct inode *inode = file_inode(file); |
| 751 |
struct super_block *sb = inode->i_sb; |
| 752 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 753 |
enum zonefs_ztype ztype = ZONEFS_ZTYPE_CNV; |
| 754 |
ino_t base_ino = bdev_nr_zones(sb->s_bdev) + 1; |
| 755 |
|
| 756 |
if (ctx->pos >= inode->i_size) |
| 757 |
return 0; |
| 758 |
|
| 759 |
if (!dir_emit_dots(file, ctx)) |
| 760 |
return 0; |
| 761 |
|
| 762 |
if (ctx->pos == 2) { |
| 763 |
if (!sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones) |
| 764 |
ztype = ZONEFS_ZTYPE_SEQ; |
| 765 |
|
| 766 |
if (!dir_emit(ctx, zonefs_zgroup_name(ztype), 3, |
| 767 |
base_ino + ztype, DT_DIR)) |
| 768 |
return 0; |
| 769 |
ctx->pos++; |
| 770 |
} |
| 771 |
|
| 772 |
if (ctx->pos == 3 && ztype != ZONEFS_ZTYPE_SEQ) { |
| 773 |
ztype = ZONEFS_ZTYPE_SEQ; |
| 774 |
if (!dir_emit(ctx, zonefs_zgroup_name(ztype), 3, |
| 775 |
base_ino + ztype, DT_DIR)) |
| 776 |
return 0; |
| 777 |
ctx->pos++; |
| 778 |
} |
| 779 |
|
| 780 |
return 0; |
| 781 |
} |
| 782 |
|
| 783 |
static int zonefs_readdir_zgroup(struct file *file, |
| 784 |
struct dir_context *ctx) |
| 785 |
{ |
| 786 |
struct inode *inode = file_inode(file); |
| 787 |
struct zonefs_zone_group *zgroup = inode->i_private; |
| 788 |
struct super_block *sb = inode->i_sb; |
| 789 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 790 |
struct zonefs_zone *z; |
| 791 |
int fname_len; |
| 792 |
char *fname; |
| 793 |
ino_t ino; |
| 794 |
int f; |
| 795 |
|
| 796 |
/* |
| 797 |
* The size of zone group directories is equal to the number |
| 798 |
* of zone files in the group and does note include the "." and |
| 799 |
* ".." entries. Hence the "+ 2" here. |
| 800 |
*/ |
| 801 |
if (ctx->pos >= inode->i_size + 2) |
| 802 |
return 0; |
| 803 |
|
| 804 |
if (!dir_emit_dots(file, ctx)) |
| 805 |
return 0; |
| 806 |
|
| 807 |
fname = kmalloc(ZONEFS_NAME_MAX, GFP_KERNEL); |
| 808 |
if (!fname) |
| 809 |
return -ENOMEM; |
| 810 |
|
| 811 |
for (f = ctx->pos - 2; f < zgroup->g_nr_zones; f++) { |
| 812 |
z = &zgroup->g_zones[f]; |
| 813 |
ino = z->z_sector >> sbi->s_zone_sectors_shift; |
| 814 |
fname_len = snprintf(fname, ZONEFS_NAME_MAX - 1, "%u", f); |
| 815 |
if (!dir_emit(ctx, fname, fname_len, ino, DT_REG)) |
| 816 |
break; |
| 817 |
ctx->pos++; |
| 818 |
} |
| 819 |
|
| 820 |
kfree(fname); |
| 821 |
|
| 822 |
return 0; |
| 823 |
} |
| 824 |
|
| 825 |
static int zonefs_readdir(struct file *file, struct dir_context *ctx) |
| 826 |
{ |
| 827 |
struct inode *inode = file_inode(file); |
| 828 |
|
| 829 |
if (inode == d_inode(inode->i_sb->s_root)) |
| 830 |
return zonefs_readdir_root(file, ctx); |
| 831 |
|
| 832 |
return zonefs_readdir_zgroup(file, ctx); |
| 833 |
} |
| 834 |
|
| 835 |
const struct inode_operations zonefs_dir_inode_operations = { |
| 836 |
.lookup = zonefs_lookup, |
| 837 |
.setattr = zonefs_inode_setattr, |
| 838 |
}; |
| 839 |
|
| 840 |
const struct file_operations zonefs_dir_operations = { |
| 841 |
.llseek = generic_file_llseek, |
| 842 |
.read = generic_read_dir, |
| 843 |
.iterate_shared = zonefs_readdir, |
| 844 |
}; |
| 845 |
|
| 846 |
struct zonefs_zone_data { |
| 847 |
struct super_block *sb; |
| 848 |
unsigned int nr_zones[ZONEFS_ZTYPE_MAX]; |
| 849 |
sector_t cnv_zone_start; |
| 850 |
struct blk_zone *zones; |
| 851 |
}; |
| 852 |
|
| 853 |
static int zonefs_get_zone_info_cb(struct blk_zone *zone, unsigned int idx, |
| 854 |
void *data) |
| 855 |
{ |
| 856 |
struct zonefs_zone_data *zd = data; |
| 857 |
struct super_block *sb = zd->sb; |
| 858 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 859 |
|
| 860 |
/* |
| 861 |
* We do not care about the first zone: it contains the super block |
| 862 |
* and not exposed as a file. |
| 863 |
*/ |
| 864 |
if (!idx) |
| 865 |
return 0; |
| 866 |
|
| 867 |
/* |
| 868 |
* Count the number of zones that will be exposed as files. |
| 869 |
* For sequential zones, we always have as many files as zones. |
| 870 |
* FOr conventional zones, the number of files depends on if we have |
| 871 |
* conventional zones aggregation enabled. |
| 872 |
*/ |
| 873 |
switch (zone->type) { |
| 874 |
case BLK_ZONE_TYPE_CONVENTIONAL: |
| 875 |
if (sbi->s_features & ZONEFS_F_AGGRCNV) { |
| 876 |
/* One file per set of contiguous conventional zones */ |
| 877 |
if (!(sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones) || |
| 878 |
zone->start != zd->cnv_zone_start) |
| 879 |
sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++; |
| 880 |
zd->cnv_zone_start = zone->start + zone->len; |
| 881 |
} else { |
| 882 |
/* One file per zone */ |
| 883 |
sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++; |
| 884 |
} |
| 885 |
break; |
| 886 |
case BLK_ZONE_TYPE_SEQWRITE_REQ: |
| 887 |
case BLK_ZONE_TYPE_SEQWRITE_PREF: |
| 888 |
sbi->s_zgroup[ZONEFS_ZTYPE_SEQ].g_nr_zones++; |
| 889 |
break; |
| 890 |
default: |
| 891 |
zonefs_err(zd->sb, "Unsupported zone type 0x%x\n", |
| 892 |
zone->type); |
| 893 |
return -EIO; |
| 894 |
} |
| 895 |
|
| 896 |
memcpy(&zd->zones[idx], zone, sizeof(struct blk_zone)); |
| 897 |
|
| 898 |
return 0; |
| 899 |
} |
| 900 |
|
| 901 |
static int zonefs_get_zone_info(struct zonefs_zone_data *zd) |
| 902 |
{ |
| 903 |
struct block_device *bdev = zd->sb->s_bdev; |
| 904 |
int ret; |
| 905 |
|
| 906 |
zd->zones = kvzalloc_objs(struct blk_zone, bdev_nr_zones(bdev)); |
| 907 |
if (!zd->zones) |
| 908 |
return -ENOMEM; |
| 909 |
|
| 910 |
/* Get zones information from the device */ |
| 911 |
ret = blkdev_report_zones(bdev, 0, BLK_ALL_ZONES, |
| 912 |
zonefs_get_zone_info_cb, zd); |
| 913 |
if (ret < 0) { |
| 914 |
zonefs_err(zd->sb, "Zone report failed %d\n", ret); |
| 915 |
return ret; |
| 916 |
} |
| 917 |
|
| 918 |
if (ret != bdev_nr_zones(bdev)) { |
| 919 |
zonefs_err(zd->sb, "Invalid zone report (%d/%u zones)\n", |
| 920 |
ret, bdev_nr_zones(bdev)); |
| 921 |
return -EIO; |
| 922 |
} |
| 923 |
|
| 924 |
return 0; |
| 925 |
} |
| 926 |
|
| 927 |
static inline void zonefs_free_zone_info(struct zonefs_zone_data *zd) |
| 928 |
{ |
| 929 |
kvfree(zd->zones); |
| 930 |
} |
| 931 |
|
| 932 |
/* |
| 933 |
* Create a zone group and populate it with zone files. |
| 934 |
*/ |
| 935 |
static int zonefs_init_zgroup(struct super_block *sb, |
| 936 |
struct zonefs_zone_data *zd, |
| 937 |
enum zonefs_ztype ztype) |
| 938 |
{ |
| 939 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 940 |
struct zonefs_zone_group *zgroup = &sbi->s_zgroup[ztype]; |
| 941 |
struct blk_zone *zone, *next, *end; |
| 942 |
struct zonefs_zone *z; |
| 943 |
unsigned int n = 0; |
| 944 |
int ret; |
| 945 |
|
| 946 |
/* Allocate the zone group. If it is empty, we have nothing to do. */ |
| 947 |
if (!zgroup->g_nr_zones) |
| 948 |
return 0; |
| 949 |
|
| 950 |
zgroup->g_zones = kvzalloc_objs(struct zonefs_zone, zgroup->g_nr_zones); |
| 951 |
if (!zgroup->g_zones) |
| 952 |
return -ENOMEM; |
| 953 |
|
| 954 |
/* |
| 955 |
* Initialize the zone groups using the device zone information. |
| 956 |
* We always skip the first zone as it contains the super block |
| 957 |
* and is not use to back a file. |
| 958 |
*/ |
| 959 |
end = zd->zones + bdev_nr_zones(sb->s_bdev); |
| 960 |
for (zone = &zd->zones[1]; zone < end; zone = next) { |
| 961 |
|
| 962 |
next = zone + 1; |
| 963 |
if (zonefs_zone_type(zone) != ztype) |
| 964 |
continue; |
| 965 |
|
| 966 |
if (WARN_ON_ONCE(n >= zgroup->g_nr_zones)) |
| 967 |
return -EINVAL; |
| 968 |
|
| 969 |
/* |
| 970 |
* For conventional zones, contiguous zones can be aggregated |
| 971 |
* together to form larger files. Note that this overwrites the |
| 972 |
* length of the first zone of the set of contiguous zones |
| 973 |
* aggregated together. If one offline or read-only zone is |
| 974 |
* found, assume that all zones aggregated have the same |
| 975 |
* condition. |
| 976 |
*/ |
| 977 |
if (ztype == ZONEFS_ZTYPE_CNV && |
| 978 |
(sbi->s_features & ZONEFS_F_AGGRCNV)) { |
| 979 |
for (; next < end; next++) { |
| 980 |
if (zonefs_zone_type(next) != ztype) |
| 981 |
break; |
| 982 |
zone->len += next->len; |
| 983 |
zone->capacity += next->capacity; |
| 984 |
if (next->cond == BLK_ZONE_COND_READONLY && |
| 985 |
zone->cond != BLK_ZONE_COND_OFFLINE) |
| 986 |
zone->cond = BLK_ZONE_COND_READONLY; |
| 987 |
else if (next->cond == BLK_ZONE_COND_OFFLINE) |
| 988 |
zone->cond = BLK_ZONE_COND_OFFLINE; |
| 989 |
} |
| 990 |
} |
| 991 |
|
| 992 |
z = &zgroup->g_zones[n]; |
| 993 |
if (ztype == ZONEFS_ZTYPE_CNV) |
| 994 |
z->z_flags |= ZONEFS_ZONE_CNV; |
| 995 |
z->z_sector = zone->start; |
| 996 |
z->z_size = zone->len << SECTOR_SHIFT; |
| 997 |
if (z->z_size > bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT && |
| 998 |
!(sbi->s_features & ZONEFS_F_AGGRCNV)) { |
| 999 |
zonefs_err(sb, |
| 1000 |
"Invalid zone size %llu (device zone sectors %llu)\n", |
| 1001 |
z->z_size, |
| 1002 |
bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT); |
| 1003 |
return -EINVAL; |
| 1004 |
} |
| 1005 |
|
| 1006 |
z->z_capacity = min_t(loff_t, MAX_LFS_FILESIZE, |
| 1007 |
zone->capacity << SECTOR_SHIFT); |
| 1008 |
z->z_wpoffset = zonefs_check_zone_condition(sb, z, zone); |
| 1009 |
|
| 1010 |
z->z_mode = S_IFREG | sbi->s_perm; |
| 1011 |
z->z_uid = sbi->s_uid; |
| 1012 |
z->z_gid = sbi->s_gid; |
| 1013 |
|
| 1014 |
/* |
| 1015 |
* Let zonefs_inode_update_mode() know that we will need |
| 1016 |
* special initialization of the inode mode the first time |
| 1017 |
* it is accessed. |
| 1018 |
*/ |
| 1019 |
z->z_flags |= ZONEFS_ZONE_INIT_MODE; |
| 1020 |
|
| 1021 |
sb->s_maxbytes = max(z->z_capacity, sb->s_maxbytes); |
| 1022 |
sbi->s_blocks += z->z_capacity >> sb->s_blocksize_bits; |
| 1023 |
sbi->s_used_blocks += z->z_wpoffset >> sb->s_blocksize_bits; |
| 1024 |
|
| 1025 |
/* |
| 1026 |
* For sequential zones, make sure that any open zone is closed |
| 1027 |
* first to ensure that the initial number of open zones is 0, |
| 1028 |
* in sync with the open zone accounting done when the mount |
| 1029 |
* option ZONEFS_MNTOPT_EXPLICIT_OPEN is used. |
| 1030 |
*/ |
| 1031 |
if (ztype == ZONEFS_ZTYPE_SEQ && |
| 1032 |
(zone->cond == BLK_ZONE_COND_IMP_OPEN || |
| 1033 |
zone->cond == BLK_ZONE_COND_EXP_OPEN)) { |
| 1034 |
ret = zonefs_zone_mgmt(sb, z, REQ_OP_ZONE_CLOSE); |
| 1035 |
if (ret) |
| 1036 |
return ret; |
| 1037 |
} |
| 1038 |
|
| 1039 |
zonefs_account_active(sb, z); |
| 1040 |
|
| 1041 |
n++; |
| 1042 |
} |
| 1043 |
|
| 1044 |
if (WARN_ON_ONCE(n != zgroup->g_nr_zones)) |
| 1045 |
return -EINVAL; |
| 1046 |
|
| 1047 |
zonefs_info(sb, "Zone group \"%s\" has %u file%s\n", |
| 1048 |
zonefs_zgroup_name(ztype), |
| 1049 |
zgroup->g_nr_zones, |
| 1050 |
str_plural(zgroup->g_nr_zones)); |
| 1051 |
|
| 1052 |
return 0; |
| 1053 |
} |
| 1054 |
|
| 1055 |
static void zonefs_free_zgroups(struct super_block *sb) |
| 1056 |
{ |
| 1057 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 1058 |
enum zonefs_ztype ztype; |
| 1059 |
|
| 1060 |
if (!sbi) |
| 1061 |
return; |
| 1062 |
|
| 1063 |
for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) { |
| 1064 |
kvfree(sbi->s_zgroup[ztype].g_zones); |
| 1065 |
sbi->s_zgroup[ztype].g_zones = NULL; |
| 1066 |
} |
| 1067 |
} |
| 1068 |
|
| 1069 |
/* |
| 1070 |
* Create a zone group and populate it with zone files. |
| 1071 |
*/ |
| 1072 |
static int zonefs_init_zgroups(struct super_block *sb) |
| 1073 |
{ |
| 1074 |
struct zonefs_zone_data zd; |
| 1075 |
enum zonefs_ztype ztype; |
| 1076 |
int ret; |
| 1077 |
|
| 1078 |
/* First get the device zone information */ |
| 1079 |
memset(&zd, 0, sizeof(struct zonefs_zone_data)); |
| 1080 |
zd.sb = sb; |
| 1081 |
ret = zonefs_get_zone_info(&zd); |
| 1082 |
if (ret) |
| 1083 |
goto cleanup; |
| 1084 |
|
| 1085 |
/* Allocate and initialize the zone groups */ |
| 1086 |
for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) { |
| 1087 |
ret = zonefs_init_zgroup(sb, &zd, ztype); |
| 1088 |
if (ret) { |
| 1089 |
zonefs_info(sb, |
| 1090 |
"Zone group \"%s\" initialization failed\n", |
| 1091 |
zonefs_zgroup_name(ztype)); |
| 1092 |
break; |
| 1093 |
} |
| 1094 |
} |
| 1095 |
|
| 1096 |
cleanup: |
| 1097 |
zonefs_free_zone_info(&zd); |
| 1098 |
if (ret) |
| 1099 |
zonefs_free_zgroups(sb); |
| 1100 |
|
| 1101 |
return ret; |
| 1102 |
} |
| 1103 |
|
| 1104 |
/* |
| 1105 |
* Read super block information from the device. |
| 1106 |
*/ |
| 1107 |
static int zonefs_read_super(struct super_block *sb) |
| 1108 |
{ |
| 1109 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 1110 |
struct zonefs_super *super; |
| 1111 |
u32 crc, stored_crc; |
| 1112 |
int ret; |
| 1113 |
|
| 1114 |
super = kmalloc(ZONEFS_SUPER_SIZE, GFP_KERNEL); |
| 1115 |
if (!super) |
| 1116 |
return -ENOMEM; |
| 1117 |
|
| 1118 |
ret = bdev_rw_virt(sb->s_bdev, 0, super, ZONEFS_SUPER_SIZE, |
| 1119 |
REQ_OP_READ); |
| 1120 |
if (ret) |
| 1121 |
goto free_super; |
| 1122 |
|
| 1123 |
ret = -EINVAL; |
| 1124 |
if (le32_to_cpu(super->s_magic) != ZONEFS_MAGIC) |
| 1125 |
goto free_super; |
| 1126 |
|
| 1127 |
stored_crc = le32_to_cpu(super->s_crc); |
| 1128 |
super->s_crc = 0; |
| 1129 |
crc = crc32(~0U, (unsigned char *)super, sizeof(struct zonefs_super)); |
| 1130 |
if (crc != stored_crc) { |
| 1131 |
zonefs_err(sb, "Invalid checksum (Expected 0x%08x, got 0x%08x)", |
| 1132 |
crc, stored_crc); |
| 1133 |
goto free_super; |
| 1134 |
} |
| 1135 |
|
| 1136 |
sbi->s_features = le64_to_cpu(super->s_features); |
| 1137 |
if (sbi->s_features & ~ZONEFS_F_DEFINED_FEATURES) { |
| 1138 |
zonefs_err(sb, "Unknown features set 0x%llx\n", |
| 1139 |
sbi->s_features); |
| 1140 |
goto free_super; |
| 1141 |
} |
| 1142 |
|
| 1143 |
if (sbi->s_features & ZONEFS_F_UID) { |
| 1144 |
sbi->s_uid = make_kuid(current_user_ns(), |
| 1145 |
le32_to_cpu(super->s_uid)); |
| 1146 |
if (!uid_valid(sbi->s_uid)) { |
| 1147 |
zonefs_err(sb, "Invalid UID feature\n"); |
| 1148 |
goto free_super; |
| 1149 |
} |
| 1150 |
} |
| 1151 |
|
| 1152 |
if (sbi->s_features & ZONEFS_F_GID) { |
| 1153 |
sbi->s_gid = make_kgid(current_user_ns(), |
| 1154 |
le32_to_cpu(super->s_gid)); |
| 1155 |
if (!gid_valid(sbi->s_gid)) { |
| 1156 |
zonefs_err(sb, "Invalid GID feature\n"); |
| 1157 |
goto free_super; |
| 1158 |
} |
| 1159 |
} |
| 1160 |
|
| 1161 |
if (sbi->s_features & ZONEFS_F_PERM) |
| 1162 |
sbi->s_perm = le32_to_cpu(super->s_perm); |
| 1163 |
|
| 1164 |
if (memchr_inv(super->s_reserved, 0, sizeof(super->s_reserved))) { |
| 1165 |
zonefs_err(sb, "Reserved area is being used\n"); |
| 1166 |
goto free_super; |
| 1167 |
} |
| 1168 |
|
| 1169 |
import_uuid(&sbi->s_uuid, super->s_uuid); |
| 1170 |
ret = 0; |
| 1171 |
|
| 1172 |
free_super: |
| 1173 |
kfree(super); |
| 1174 |
return ret; |
| 1175 |
} |
| 1176 |
|
| 1177 |
static const struct super_operations zonefs_sops = { |
| 1178 |
.alloc_inode = zonefs_alloc_inode, |
| 1179 |
.free_inode = zonefs_free_inode, |
| 1180 |
.statfs = zonefs_statfs, |
| 1181 |
.show_options = zonefs_show_options, |
| 1182 |
}; |
| 1183 |
|
| 1184 |
static int zonefs_get_zgroup_inodes(struct super_block *sb) |
| 1185 |
{ |
| 1186 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 1187 |
struct inode *dir_inode; |
| 1188 |
enum zonefs_ztype ztype; |
| 1189 |
|
| 1190 |
for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) { |
| 1191 |
if (!sbi->s_zgroup[ztype].g_nr_zones) |
| 1192 |
continue; |
| 1193 |
|
| 1194 |
dir_inode = zonefs_get_zgroup_inode(sb, ztype); |
| 1195 |
if (IS_ERR(dir_inode)) |
| 1196 |
return PTR_ERR(dir_inode); |
| 1197 |
|
| 1198 |
sbi->s_zgroup[ztype].g_inode = dir_inode; |
| 1199 |
} |
| 1200 |
|
| 1201 |
return 0; |
| 1202 |
} |
| 1203 |
|
| 1204 |
static void zonefs_release_zgroup_inodes(struct super_block *sb) |
| 1205 |
{ |
| 1206 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 1207 |
enum zonefs_ztype ztype; |
| 1208 |
|
| 1209 |
if (!sbi) |
| 1210 |
return; |
| 1211 |
|
| 1212 |
for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) { |
| 1213 |
if (sbi->s_zgroup[ztype].g_inode) { |
| 1214 |
iput(sbi->s_zgroup[ztype].g_inode); |
| 1215 |
sbi->s_zgroup[ztype].g_inode = NULL; |
| 1216 |
} |
| 1217 |
} |
| 1218 |
} |
| 1219 |
|
| 1220 |
/* |
| 1221 |
* Check that the device is zoned. If it is, get the list of zones and create |
| 1222 |
* sub-directories and files according to the device zone configuration and |
| 1223 |
* format options. |
| 1224 |
*/ |
| 1225 |
static int zonefs_fill_super(struct super_block *sb, struct fs_context *fc) |
| 1226 |
{ |
| 1227 |
struct zonefs_sb_info *sbi; |
| 1228 |
struct zonefs_context *ctx = fc->fs_private; |
| 1229 |
struct inode *inode; |
| 1230 |
enum zonefs_ztype ztype; |
| 1231 |
int ret; |
| 1232 |
|
| 1233 |
if (!bdev_is_zoned(sb->s_bdev)) { |
| 1234 |
zonefs_err(sb, "Not a zoned block device\n"); |
| 1235 |
return -EINVAL; |
| 1236 |
} |
| 1237 |
|
| 1238 |
/* |
| 1239 |
* Initialize super block information: the maximum file size is updated |
| 1240 |
* when the zone files are created so that the format option |
| 1241 |
* ZONEFS_F_AGGRCNV which increases the maximum file size of a file |
| 1242 |
* beyond the zone size is taken into account. |
| 1243 |
*/ |
| 1244 |
sbi = kzalloc_obj(*sbi); |
| 1245 |
if (!sbi) |
| 1246 |
return -ENOMEM; |
| 1247 |
|
| 1248 |
spin_lock_init(&sbi->s_lock); |
| 1249 |
sb->s_fs_info = sbi; |
| 1250 |
sb->s_magic = ZONEFS_MAGIC; |
| 1251 |
sb->s_maxbytes = 0; |
| 1252 |
sb->s_op = &zonefs_sops; |
| 1253 |
sb->s_time_gran = 1; |
| 1254 |
|
| 1255 |
/* |
| 1256 |
* The block size is set to the device zone write granularity to ensure |
| 1257 |
* that write operations are always aligned according to the device |
| 1258 |
* interface constraints. |
| 1259 |
*/ |
| 1260 |
sb_set_blocksize(sb, bdev_zone_write_granularity(sb->s_bdev)); |
| 1261 |
sbi->s_zone_sectors_shift = ilog2(bdev_zone_sectors(sb->s_bdev)); |
| 1262 |
sbi->s_uid = GLOBAL_ROOT_UID; |
| 1263 |
sbi->s_gid = GLOBAL_ROOT_GID; |
| 1264 |
sbi->s_perm = 0640; |
| 1265 |
sbi->s_mount_opts = ctx->s_mount_opts; |
| 1266 |
|
| 1267 |
atomic_set(&sbi->s_wro_seq_files, 0); |
| 1268 |
sbi->s_max_wro_seq_files = bdev_max_open_zones(sb->s_bdev); |
| 1269 |
atomic_set(&sbi->s_active_seq_files, 0); |
| 1270 |
sbi->s_max_active_seq_files = bdev_max_active_zones(sb->s_bdev); |
| 1271 |
|
| 1272 |
ret = zonefs_read_super(sb); |
| 1273 |
if (ret) |
| 1274 |
return ret; |
| 1275 |
|
| 1276 |
zonefs_info(sb, "Mounting %u zones", bdev_nr_zones(sb->s_bdev)); |
| 1277 |
|
| 1278 |
if (!sbi->s_max_wro_seq_files && |
| 1279 |
!sbi->s_max_active_seq_files && |
| 1280 |
sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) { |
| 1281 |
zonefs_info(sb, |
| 1282 |
"No open and active zone limits. Ignoring explicit_open mount option\n"); |
| 1283 |
sbi->s_mount_opts &= ~ZONEFS_MNTOPT_EXPLICIT_OPEN; |
| 1284 |
} |
| 1285 |
|
| 1286 |
/* Initialize the zone groups */ |
| 1287 |
ret = zonefs_init_zgroups(sb); |
| 1288 |
if (ret) |
| 1289 |
goto cleanup; |
| 1290 |
|
| 1291 |
/* Create the root directory inode */ |
| 1292 |
ret = -ENOMEM; |
| 1293 |
inode = new_inode(sb); |
| 1294 |
if (!inode) |
| 1295 |
goto cleanup; |
| 1296 |
|
| 1297 |
inode->i_ino = bdev_nr_zones(sb->s_bdev); |
| 1298 |
inode->i_mode = S_IFDIR | 0555; |
| 1299 |
simple_inode_init_ts(inode); |
| 1300 |
inode->i_op = &zonefs_dir_inode_operations; |
| 1301 |
inode->i_fop = &zonefs_dir_operations; |
| 1302 |
inode->i_size = 2; |
| 1303 |
set_nlink(inode, 2); |
| 1304 |
for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) { |
| 1305 |
if (sbi->s_zgroup[ztype].g_nr_zones) { |
| 1306 |
inc_nlink(inode); |
| 1307 |
inode->i_size++; |
| 1308 |
} |
| 1309 |
} |
| 1310 |
|
| 1311 |
sb->s_root = d_make_root(inode); |
| 1312 |
if (!sb->s_root) |
| 1313 |
goto cleanup; |
| 1314 |
|
| 1315 |
/* |
| 1316 |
* Take a reference on the zone groups directory inodes |
| 1317 |
* to keep them in the inode cache. |
| 1318 |
*/ |
| 1319 |
ret = zonefs_get_zgroup_inodes(sb); |
| 1320 |
if (ret) |
| 1321 |
goto cleanup; |
| 1322 |
|
| 1323 |
ret = zonefs_sysfs_register(sb); |
| 1324 |
if (ret) |
| 1325 |
goto cleanup; |
| 1326 |
|
| 1327 |
return 0; |
| 1328 |
|
| 1329 |
cleanup: |
| 1330 |
zonefs_release_zgroup_inodes(sb); |
| 1331 |
zonefs_free_zgroups(sb); |
| 1332 |
|
| 1333 |
return ret; |
| 1334 |
} |
| 1335 |
|
| 1336 |
static void zonefs_kill_super(struct super_block *sb) |
| 1337 |
{ |
| 1338 |
struct zonefs_sb_info *sbi = ZONEFS_SB(sb); |
| 1339 |
|
| 1340 |
/* Release the reference on the zone group directory inodes */ |
| 1341 |
zonefs_release_zgroup_inodes(sb); |
| 1342 |
|
| 1343 |
kill_block_super(sb); |
| 1344 |
|
| 1345 |
zonefs_sysfs_unregister(sb); |
| 1346 |
zonefs_free_zgroups(sb); |
| 1347 |
kfree(sbi); |
| 1348 |
} |
| 1349 |
|
| 1350 |
static void zonefs_free_fc(struct fs_context *fc) |
| 1351 |
{ |
| 1352 |
struct zonefs_context *ctx = fc->fs_private; |
| 1353 |
|
| 1354 |
kfree(ctx); |
| 1355 |
} |
| 1356 |
|
| 1357 |
static int zonefs_get_tree(struct fs_context *fc) |
| 1358 |
{ |
| 1359 |
return get_tree_bdev(fc, zonefs_fill_super); |
| 1360 |
} |
| 1361 |
|
| 1362 |
static int zonefs_reconfigure(struct fs_context *fc) |
| 1363 |
{ |
| 1364 |
struct zonefs_context *ctx = fc->fs_private; |
| 1365 |
struct super_block *sb = fc->root->d_sb; |
| 1366 |
struct zonefs_sb_info *sbi = sb->s_fs_info; |
| 1367 |
|
| 1368 |
sync_filesystem(fc->root->d_sb); |
| 1369 |
/* Copy new options from ctx into sbi. */ |
| 1370 |
sbi->s_mount_opts = ctx->s_mount_opts; |
| 1371 |
|
| 1372 |
return 0; |
| 1373 |
} |
| 1374 |
|
| 1375 |
static const struct fs_context_operations zonefs_context_ops = { |
| 1376 |
.parse_param = zonefs_parse_param, |
| 1377 |
.get_tree = zonefs_get_tree, |
| 1378 |
.reconfigure = zonefs_reconfigure, |
| 1379 |
.free = zonefs_free_fc, |
| 1380 |
}; |
| 1381 |
|
| 1382 |
/* |
| 1383 |
* Set up the filesystem mount context. |
| 1384 |
*/ |
| 1385 |
static int zonefs_init_fs_context(struct fs_context *fc) |
| 1386 |
{ |
| 1387 |
struct zonefs_context *ctx; |
| 1388 |
|
| 1389 |
ctx = kzalloc_obj(struct zonefs_context); |
| 1390 |
if (!ctx) |
| 1391 |
return -ENOMEM; |
| 1392 |
ctx->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO; |
| 1393 |
fc->ops = &zonefs_context_ops; |
| 1394 |
fc->fs_private = ctx; |
| 1395 |
|
| 1396 |
return 0; |
| 1397 |
} |
| 1398 |
|
| 1399 |
/* |
| 1400 |
* File system definition and registration. |
| 1401 |
*/ |
| 1402 |
static struct file_system_type zonefs_type = { |
| 1403 |
.owner = THIS_MODULE, |
| 1404 |
.name = "zonefs", |
| 1405 |
.kill_sb = zonefs_kill_super, |
| 1406 |
.fs_flags = FS_REQUIRES_DEV, |
| 1407 |
.init_fs_context = zonefs_init_fs_context, |
| 1408 |
.parameters = zonefs_param_spec, |
| 1409 |
}; |
| 1410 |
|
| 1411 |
static int __init zonefs_init_inodecache(void) |
| 1412 |
{ |
| 1413 |
zonefs_inode_cachep = kmem_cache_create("zonefs_inode_cache", |
| 1414 |
sizeof(struct zonefs_inode_info), 0, |
| 1415 |
SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT, |
| 1416 |
NULL); |
| 1417 |
if (zonefs_inode_cachep == NULL) |
| 1418 |
return -ENOMEM; |
| 1419 |
return 0; |
| 1420 |
} |
| 1421 |
|
| 1422 |
static void zonefs_destroy_inodecache(void) |
| 1423 |
{ |
| 1424 |
/* |
| 1425 |
* Make sure all delayed rcu free inodes are flushed before we |
| 1426 |
* destroy the inode cache. |
| 1427 |
*/ |
| 1428 |
rcu_barrier(); |
| 1429 |
kmem_cache_destroy(zonefs_inode_cachep); |
| 1430 |
} |
| 1431 |
|
| 1432 |
static int __init zonefs_init(void) |
| 1433 |
{ |
| 1434 |
int ret; |
| 1435 |
|
| 1436 |
BUILD_BUG_ON(sizeof(struct zonefs_super) != ZONEFS_SUPER_SIZE); |
| 1437 |
|
| 1438 |
ret = zonefs_init_inodecache(); |
| 1439 |
if (ret) |
| 1440 |
return ret; |
| 1441 |
|
| 1442 |
ret = zonefs_sysfs_init(); |
| 1443 |
if (ret) |
| 1444 |
goto destroy_inodecache; |
| 1445 |
|
| 1446 |
ret = register_filesystem(&zonefs_type); |
| 1447 |
if (ret) |
| 1448 |
goto sysfs_exit; |
| 1449 |
|
| 1450 |
return 0; |
| 1451 |
|
| 1452 |
sysfs_exit: |
| 1453 |
zonefs_sysfs_exit(); |
| 1454 |
destroy_inodecache: |
| 1455 |
zonefs_destroy_inodecache(); |
| 1456 |
|
| 1457 |
return ret; |
| 1458 |
} |
| 1459 |
|
| 1460 |
static void __exit zonefs_exit(void) |
| 1461 |
{ |
| 1462 |
unregister_filesystem(&zonefs_type); |
| 1463 |
zonefs_sysfs_exit(); |
| 1464 |
zonefs_destroy_inodecache(); |
| 1465 |
} |
| 1466 |
|
| 1467 |
MODULE_AUTHOR("Damien Le Moal"); |
| 1468 |
MODULE_DESCRIPTION("Zone file system for zoned block devices"); |
| 1469 |
MODULE_LICENSE("GPL"); |
| 1470 |
MODULE_ALIAS_FS("zonefs"); |
| 1471 |
module_init(zonefs_init); |
| 1472 |
module_exit(zonefs_exit); |