1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 4 * All Rights Reserved. 5 */ 6 #ifndef __XFS_BUF_H__ 7 #define __XFS_BUF_H__ 8 9 #include <linux/list.h> 10 #include <linux/types.h> 11 #include <linux/spinlock.h> 12 #include <linux/mm.h> 13 #include <linux/fs.h> 14 #include <linux/dax.h> 15 #include <linux/uio.h> 16 #include <linux/list_lru.h> 17 18 extern struct kmem_cache *xfs_buf_cache; 19 20 /* 21 * Base types 22 */ 23 struct xfs_buf; 24 25 #define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL)) 26 27 #define XBF_READ (1u << 0) /* buffer intended for reading from device */ 28 #define XBF_WRITE (1u << 1) /* buffer intended for writing to device */ 29 #define XBF_READ_AHEAD (1u << 2) /* asynchronous read-ahead */ 30 #define XBF_NO_IOACCT (1u << 3) /* bypass I/O accounting (non-LRU bufs) */ 31 #define XBF_ASYNC (1u << 4) /* initiator will not wait for completion */ 32 #define XBF_DONE (1u << 5) /* all pages in the buffer uptodate */ 33 #define XBF_STALE (1u << 6) /* buffer has been staled, do not find it */ 34 #define XBF_WRITE_FAIL (1u << 7) /* async writes have failed on this buffer */ 35 36 /* buffer type flags for write callbacks */ 37 #define _XBF_LOGRECOVERY (1u << 18)/* log recovery buffer */ 38 39 /* flags used only internally */ 40 #define _XBF_PAGES (1u << 20)/* backed by refcounted pages */ 41 #define _XBF_KMEM (1u << 21)/* backed by heap memory */ 42 #define _XBF_DELWRI_Q (1u << 22)/* buffer on a delwri queue */ 43 44 /* flags used only as arguments to access routines */ 45 /* 46 * Online fsck is scanning the buffer cache for live buffers. Do not warn 47 * about length mismatches during lookups and do not return stale buffers. 48 */ 49 #define XBF_LIVESCAN (1u << 28) 50 #define XBF_INCORE (1u << 29)/* lookup only, return if found in cache */ 51 #define XBF_TRYLOCK (1u << 30)/* lock requested, but do not wait */ 52 #define XBF_UNMAPPED (1u << 31)/* do not map the buffer */ 53 54 55 typedef unsigned int xfs_buf_flags_t; 56 57 #define XFS_BUF_FLAGS \ 58 { XBF_READ, "READ" }, \ 59 { XBF_WRITE, "WRITE" }, \ 60 { XBF_READ_AHEAD, "READ_AHEAD" }, \ 61 { XBF_NO_IOACCT, "NO_IOACCT" }, \ 62 { XBF_ASYNC, "ASYNC" }, \ 63 { XBF_DONE, "DONE" }, \ 64 { XBF_STALE, "STALE" }, \ 65 { XBF_WRITE_FAIL, "WRITE_FAIL" }, \ 66 { _XBF_LOGRECOVERY, "LOG_RECOVERY" }, \ 67 { _XBF_PAGES, "PAGES" }, \ 68 { _XBF_KMEM, "KMEM" }, \ 69 { _XBF_DELWRI_Q, "DELWRI_Q" }, \ 70 /* The following interface flags should never be set */ \ 71 { XBF_LIVESCAN, "LIVESCAN" }, \ 72 { XBF_INCORE, "INCORE" }, \ 73 { XBF_TRYLOCK, "TRYLOCK" }, \ 74 { XBF_UNMAPPED, "UNMAPPED" } 75 76 /* 77 * Internal state flags. 78 */ 79 #define XFS_BSTATE_DISPOSE (1 << 0) /* buffer being discarded */ 80 #define XFS_BSTATE_IN_FLIGHT (1 << 1) /* I/O in flight */ 81 82 struct xfs_buf_cache { 83 spinlock_t bc_lock; 84 struct rhashtable bc_hash; 85 }; 86 87 int xfs_buf_cache_init(struct xfs_buf_cache *bch); 88 void xfs_buf_cache_destroy(struct xfs_buf_cache *bch); 89 90 /* 91 * The xfs_buftarg contains 2 notions of "sector size" - 92 * 93 * 1) The metadata sector size, which is the minimum unit and 94 * alignment of IO which will be performed by metadata operations. 95 * 2) The device logical sector size 96 * 97 * The first is specified at mkfs time, and is stored on-disk in the 98 * superblock's sb_sectsize. 99 * 100 * The latter is derived from the underlying device, and controls direct IO 101 * alignment constraints. 102 */ 103 struct xfs_buftarg { 104 dev_t bt_dev; 105 struct file *bt_bdev_file; 106 struct block_device *bt_bdev; 107 struct dax_device *bt_daxdev; 108 struct file *bt_file; 109 u64 bt_dax_part_off; 110 struct xfs_mount *bt_mount; 111 unsigned int bt_meta_sectorsize; 112 size_t bt_meta_sectormask; 113 size_t bt_logical_sectorsize; 114 size_t bt_logical_sectormask; 115 116 /* LRU control structures */ 117 struct shrinker *bt_shrinker; 118 struct list_lru bt_lru; 119 120 struct percpu_counter bt_io_count; 121 struct ratelimit_state bt_ioerror_rl; 122 123 /* Atomic write unit values */ 124 unsigned int bt_bdev_awu_min; 125 unsigned int bt_bdev_awu_max; 126 127 /* built-in cache, if we're not using the perag one */ 128 struct xfs_buf_cache bt_cache[]; 129 }; 130 131 #define XB_PAGES 2 132 133 struct xfs_buf_map { 134 xfs_daddr_t bm_bn; /* block number for I/O */ 135 int bm_len; /* size of I/O */ 136 unsigned int bm_flags; 137 }; 138 139 /* 140 * Online fsck is scanning the buffer cache for live buffers. Do not warn 141 * about length mismatches during lookups and do not return stale buffers. 142 */ 143 #define XBM_LIVESCAN (1U << 0) 144 145 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \ 146 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) }; 147 148 struct xfs_buf_ops { 149 char *name; 150 union { 151 __be32 magic[2]; /* v4 and v5 on disk magic values */ 152 __be16 magic16[2]; /* v4 and v5 on disk magic values */ 153 }; 154 void (*verify_read)(struct xfs_buf *); 155 void (*verify_write)(struct xfs_buf *); 156 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp); 157 }; 158 159 struct xfs_buf { 160 /* 161 * first cacheline holds all the fields needed for an uncontended cache 162 * hit to be fully processed. The semaphore straddles the cacheline 163 * boundary, but the counter and lock sits on the first cacheline, 164 * which is the only bit that is touched if we hit the semaphore 165 * fast-path on locking. 166 */ 167 struct rhash_head b_rhash_head; /* pag buffer hash node */ 168 169 xfs_daddr_t b_rhash_key; /* buffer cache index */ 170 int b_length; /* size of buffer in BBs */ 171 unsigned int b_hold; /* reference count */ 172 atomic_t b_lru_ref; /* lru reclaim ref count */ 173 xfs_buf_flags_t b_flags; /* status flags */ 174 struct semaphore b_sema; /* semaphore for lockables */ 175 176 /* 177 * concurrent access to b_lru and b_lru_flags are protected by 178 * bt_lru_lock and not by b_sema 179 */ 180 struct list_head b_lru; /* lru list */ 181 spinlock_t b_lock; /* internal state lock */ 182 unsigned int b_state; /* internal state flags */ 183 wait_queue_head_t b_waiters; /* unpin waiters */ 184 struct list_head b_list; 185 struct xfs_perag *b_pag; 186 struct xfs_mount *b_mount; 187 struct xfs_buftarg *b_target; /* buffer target (device) */ 188 void *b_addr; /* virtual address of buffer */ 189 struct work_struct b_ioend_work; 190 struct completion b_iowait; /* queue for I/O waiters */ 191 struct xfs_buf_log_item *b_log_item; 192 struct list_head b_li_list; /* Log items list head */ 193 struct xfs_trans *b_transp; 194 struct page **b_pages; /* array of page pointers */ 195 struct page *b_page_array[XB_PAGES]; /* inline pages */ 196 struct xfs_buf_map *b_maps; /* compound buffer map */ 197 struct xfs_buf_map __b_map; /* inline compound buffer map */ 198 int b_map_count; 199 atomic_t b_pin_count; /* pin count */ 200 unsigned int b_page_count; /* size of page array */ 201 unsigned int b_offset; /* page offset of b_addr, 202 only for _XBF_KMEM buffers */ 203 int b_error; /* error code on I/O */ 204 void (*b_iodone)(struct xfs_buf *bp); 205 206 /* 207 * async write failure retry count. Initialised to zero on the first 208 * failure, then when it exceeds the maximum configured without a 209 * success the write is considered to be failed permanently and the 210 * iodone handler will take appropriate action. 211 * 212 * For retry timeouts, we record the jiffy of the first failure. This 213 * means that we can change the retry timeout for buffers already under 214 * I/O and thus avoid getting stuck in a retry loop with a long timeout. 215 * 216 * last_error is used to ensure that we are getting repeated errors, not 217 * different errors. e.g. a block device might change ENOSPC to EIO when 218 * a failure timeout occurs, so we want to re-initialise the error 219 * retry behaviour appropriately when that happens. 220 */ 221 int b_retries; 222 unsigned long b_first_retry_time; /* in jiffies */ 223 int b_last_error; 224 225 const struct xfs_buf_ops *b_ops; 226 struct rcu_head b_rcu; 227 }; 228 229 /* Finding and Reading Buffers */ 230 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map, 231 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp); 232 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map, 233 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp, 234 const struct xfs_buf_ops *ops, xfs_failaddr_t fa); 235 void xfs_buf_readahead_map(struct xfs_buftarg *target, 236 struct xfs_buf_map *map, int nmaps, 237 const struct xfs_buf_ops *ops); 238 239 static inline int 240 xfs_buf_incore( 241 struct xfs_buftarg *target, 242 xfs_daddr_t blkno, 243 size_t numblks, 244 xfs_buf_flags_t flags, 245 struct xfs_buf **bpp) 246 { 247 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); 248 249 return xfs_buf_get_map(target, &map, 1, XBF_INCORE | flags, bpp); 250 } 251 252 static inline int 253 xfs_buf_get( 254 struct xfs_buftarg *target, 255 xfs_daddr_t blkno, 256 size_t numblks, 257 struct xfs_buf **bpp) 258 { 259 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); 260 261 return xfs_buf_get_map(target, &map, 1, 0, bpp); 262 } 263 264 static inline int 265 xfs_buf_read( 266 struct xfs_buftarg *target, 267 xfs_daddr_t blkno, 268 size_t numblks, 269 xfs_buf_flags_t flags, 270 struct xfs_buf **bpp, 271 const struct xfs_buf_ops *ops) 272 { 273 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); 274 275 return xfs_buf_read_map(target, &map, 1, flags, bpp, ops, 276 __builtin_return_address(0)); 277 } 278 279 static inline void 280 xfs_buf_readahead( 281 struct xfs_buftarg *target, 282 xfs_daddr_t blkno, 283 size_t numblks, 284 const struct xfs_buf_ops *ops) 285 { 286 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); 287 return xfs_buf_readahead_map(target, &map, 1, ops); 288 } 289 290 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks, 291 xfs_buf_flags_t flags, struct xfs_buf **bpp); 292 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr, 293 size_t numblks, xfs_buf_flags_t flags, struct xfs_buf **bpp, 294 const struct xfs_buf_ops *ops); 295 int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags); 296 void xfs_buf_hold(struct xfs_buf *bp); 297 298 /* Releasing Buffers */ 299 extern void xfs_buf_rele(struct xfs_buf *); 300 301 /* Locking and Unlocking Buffers */ 302 extern int xfs_buf_trylock(struct xfs_buf *); 303 extern void xfs_buf_lock(struct xfs_buf *); 304 extern void xfs_buf_unlock(struct xfs_buf *); 305 #define xfs_buf_islocked(bp) \ 306 ((bp)->b_sema.count <= 0) 307 308 static inline void xfs_buf_relse(struct xfs_buf *bp) 309 { 310 xfs_buf_unlock(bp); 311 xfs_buf_rele(bp); 312 } 313 314 /* Buffer Read and Write Routines */ 315 extern int xfs_bwrite(struct xfs_buf *bp); 316 317 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error, 318 xfs_failaddr_t failaddr); 319 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address) 320 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa); 321 void xfs_buf_ioend_fail(struct xfs_buf *); 322 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize); 323 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa); 324 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address) 325 326 /* Buffer Utility Routines */ 327 extern void *xfs_buf_offset(struct xfs_buf *, size_t); 328 extern void xfs_buf_stale(struct xfs_buf *bp); 329 330 /* Delayed Write Buffer Routines */ 331 extern void xfs_buf_delwri_cancel(struct list_head *); 332 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *); 333 void xfs_buf_delwri_queue_here(struct xfs_buf *bp, struct list_head *bl); 334 extern int xfs_buf_delwri_submit(struct list_head *); 335 extern int xfs_buf_delwri_submit_nowait(struct list_head *); 336 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *); 337 338 static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp) 339 { 340 return bp->b_maps[0].bm_bn; 341 } 342 343 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref); 344 345 /* 346 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer 347 * up with a reference count of 0 so it will be tossed from the cache when 348 * released. 349 */ 350 static inline void xfs_buf_oneshot(struct xfs_buf *bp) 351 { 352 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1) 353 return; 354 atomic_set(&bp->b_lru_ref, 0); 355 } 356 357 static inline int xfs_buf_ispinned(struct xfs_buf *bp) 358 { 359 return atomic_read(&bp->b_pin_count); 360 } 361 362 static inline int 363 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset) 364 { 365 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length), 366 cksum_offset); 367 } 368 369 static inline void 370 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset) 371 { 372 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), 373 cksum_offset); 374 } 375 376 /* 377 * Handling of buftargs. 378 */ 379 struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp, 380 struct file *bdev_file); 381 extern void xfs_free_buftarg(struct xfs_buftarg *); 382 extern void xfs_buftarg_wait(struct xfs_buftarg *); 383 extern void xfs_buftarg_drain(struct xfs_buftarg *); 384 extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int); 385 386 #define xfs_getsize_buftarg(buftarg) block_size((buftarg)->bt_bdev) 387 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev) 388 389 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops); 390 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic); 391 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic); 392 393 /* for xfs_buf_mem.c only: */ 394 int xfs_init_buftarg(struct xfs_buftarg *btp, size_t logical_sectorsize, 395 const char *descr); 396 void xfs_destroy_buftarg(struct xfs_buftarg *btp); 397 398 #endif /* __XFS_BUF_H__ */ 399