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