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/buffer_head.h> 16 #include <linux/uio.h> 17 #include <linux/list_lru.h> 18 19 /* 20 * Base types 21 */ 22 23 #define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL)) 24 25 typedef enum { 26 XBRW_READ = 1, /* transfer into target memory */ 27 XBRW_WRITE = 2, /* transfer from target memory */ 28 XBRW_ZERO = 3, /* Zero target memory */ 29 } xfs_buf_rw_t; 30 31 #define XBF_READ (1 << 0) /* buffer intended for reading from device */ 32 #define XBF_WRITE (1 << 1) /* buffer intended for writing to device */ 33 #define XBF_READ_AHEAD (1 << 2) /* asynchronous read-ahead */ 34 #define XBF_NO_IOACCT (1 << 3) /* bypass I/O accounting (non-LRU bufs) */ 35 #define XBF_ASYNC (1 << 4) /* initiator will not wait for completion */ 36 #define XBF_DONE (1 << 5) /* all pages in the buffer uptodate */ 37 #define XBF_STALE (1 << 6) /* buffer has been staled, do not find it */ 38 #define XBF_WRITE_FAIL (1 << 24)/* async writes have failed on this buffer */ 39 40 /* I/O hints for the BIO layer */ 41 #define XBF_SYNCIO (1 << 10)/* treat this buffer as synchronous I/O */ 42 #define XBF_FUA (1 << 11)/* force cache write through mode */ 43 #define XBF_FLUSH (1 << 12)/* flush the disk cache before a write */ 44 45 /* flags used only as arguments to access routines */ 46 #define XBF_TRYLOCK (1 << 16)/* lock requested, but do not wait */ 47 #define XBF_UNMAPPED (1 << 17)/* do not map the buffer */ 48 49 /* flags used only internally */ 50 #define _XBF_PAGES (1 << 20)/* backed by refcounted pages */ 51 #define _XBF_KMEM (1 << 21)/* backed by heap memory */ 52 #define _XBF_DELWRI_Q (1 << 22)/* buffer on a delwri queue */ 53 #define _XBF_COMPOUND (1 << 23)/* compound buffer */ 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_SYNCIO, "SYNCIO" }, \ 67 { XBF_FUA, "FUA" }, \ 68 { XBF_FLUSH, "FLUSH" }, \ 69 { XBF_TRYLOCK, "TRYLOCK" }, /* should never be set */\ 70 { XBF_UNMAPPED, "UNMAPPED" }, /* ditto */\ 71 { _XBF_PAGES, "PAGES" }, \ 72 { _XBF_KMEM, "KMEM" }, \ 73 { _XBF_DELWRI_Q, "DELWRI_Q" }, \ 74 { _XBF_COMPOUND, "COMPOUND" } 75 76 77 /* 78 * Internal state flags. 79 */ 80 #define XFS_BSTATE_DISPOSE (1 << 0) /* buffer being discarded */ 81 #define XFS_BSTATE_IN_FLIGHT (1 << 1) /* I/O in flight */ 82 83 /* 84 * The xfs_buftarg contains 2 notions of "sector size" - 85 * 86 * 1) The metadata sector size, which is the minimum unit and 87 * alignment of IO which will be performed by metadata operations. 88 * 2) The device logical sector size 89 * 90 * The first is specified at mkfs time, and is stored on-disk in the 91 * superblock's sb_sectsize. 92 * 93 * The latter is derived from the underlying device, and controls direct IO 94 * alignment constraints. 95 */ 96 typedef struct xfs_buftarg { 97 dev_t bt_dev; 98 struct block_device *bt_bdev; 99 struct dax_device *bt_daxdev; 100 struct xfs_mount *bt_mount; 101 unsigned int bt_meta_sectorsize; 102 size_t bt_meta_sectormask; 103 size_t bt_logical_sectorsize; 104 size_t bt_logical_sectormask; 105 106 /* LRU control structures */ 107 struct shrinker bt_shrinker; 108 struct list_lru bt_lru; 109 110 struct percpu_counter bt_io_count; 111 } xfs_buftarg_t; 112 113 struct xfs_buf; 114 typedef void (*xfs_buf_iodone_t)(struct xfs_buf *); 115 116 117 #define XB_PAGES 2 118 119 struct xfs_buf_map { 120 xfs_daddr_t bm_bn; /* block number for I/O */ 121 int bm_len; /* size of I/O */ 122 }; 123 124 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \ 125 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) }; 126 127 struct xfs_buf_ops { 128 char *name; 129 void (*verify_read)(struct xfs_buf *); 130 void (*verify_write)(struct xfs_buf *); 131 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp); 132 }; 133 134 typedef struct xfs_buf { 135 /* 136 * first cacheline holds all the fields needed for an uncontended cache 137 * hit to be fully processed. The semaphore straddles the cacheline 138 * boundary, but the counter and lock sits on the first cacheline, 139 * which is the only bit that is touched if we hit the semaphore 140 * fast-path on locking. 141 */ 142 struct rhash_head b_rhash_head; /* pag buffer hash node */ 143 xfs_daddr_t b_bn; /* block number of buffer */ 144 int b_length; /* size of buffer in BBs */ 145 atomic_t b_hold; /* reference count */ 146 atomic_t b_lru_ref; /* lru reclaim ref count */ 147 xfs_buf_flags_t b_flags; /* status flags */ 148 struct semaphore b_sema; /* semaphore for lockables */ 149 150 /* 151 * concurrent access to b_lru and b_lru_flags are protected by 152 * bt_lru_lock and not by b_sema 153 */ 154 struct list_head b_lru; /* lru list */ 155 spinlock_t b_lock; /* internal state lock */ 156 unsigned int b_state; /* internal state flags */ 157 int b_io_error; /* internal IO error state */ 158 wait_queue_head_t b_waiters; /* unpin waiters */ 159 struct list_head b_list; 160 struct xfs_perag *b_pag; /* contains rbtree root */ 161 xfs_buftarg_t *b_target; /* buffer target (device) */ 162 void *b_addr; /* virtual address of buffer */ 163 struct work_struct b_ioend_work; 164 struct workqueue_struct *b_ioend_wq; /* I/O completion wq */ 165 xfs_buf_iodone_t b_iodone; /* I/O completion function */ 166 struct completion b_iowait; /* queue for I/O waiters */ 167 void *b_log_item; 168 struct list_head b_li_list; /* Log items list head */ 169 struct xfs_trans *b_transp; 170 struct page **b_pages; /* array of page pointers */ 171 struct page *b_page_array[XB_PAGES]; /* inline pages */ 172 struct xfs_buf_map *b_maps; /* compound buffer map */ 173 struct xfs_buf_map __b_map; /* inline compound buffer map */ 174 int b_map_count; 175 int b_io_length; /* IO size in BBs */ 176 atomic_t b_pin_count; /* pin count */ 177 atomic_t b_io_remaining; /* #outstanding I/O requests */ 178 unsigned int b_page_count; /* size of page array */ 179 unsigned int b_offset; /* page offset in first page */ 180 int b_error; /* error code on I/O */ 181 182 /* 183 * async write failure retry count. Initialised to zero on the first 184 * failure, then when it exceeds the maximum configured without a 185 * success the write is considered to be failed permanently and the 186 * iodone handler will take appropriate action. 187 * 188 * For retry timeouts, we record the jiffie of the first failure. This 189 * means that we can change the retry timeout for buffers already under 190 * I/O and thus avoid getting stuck in a retry loop with a long timeout. 191 * 192 * last_error is used to ensure that we are getting repeated errors, not 193 * different errors. e.g. a block device might change ENOSPC to EIO when 194 * a failure timeout occurs, so we want to re-initialise the error 195 * retry behaviour appropriately when that happens. 196 */ 197 int b_retries; 198 unsigned long b_first_retry_time; /* in jiffies */ 199 int b_last_error; 200 201 const struct xfs_buf_ops *b_ops; 202 203 #ifdef XFS_BUF_LOCK_TRACKING 204 int b_last_holder; 205 #endif 206 } xfs_buf_t; 207 208 /* Finding and Reading Buffers */ 209 struct xfs_buf *xfs_buf_incore(struct xfs_buftarg *target, 210 xfs_daddr_t blkno, size_t numblks, 211 xfs_buf_flags_t flags); 212 213 struct xfs_buf *_xfs_buf_alloc(struct xfs_buftarg *target, 214 struct xfs_buf_map *map, int nmaps, 215 xfs_buf_flags_t flags); 216 217 static inline struct xfs_buf * 218 xfs_buf_alloc( 219 struct xfs_buftarg *target, 220 xfs_daddr_t blkno, 221 size_t numblks, 222 xfs_buf_flags_t flags) 223 { 224 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); 225 return _xfs_buf_alloc(target, &map, 1, flags); 226 } 227 228 struct xfs_buf *xfs_buf_get_map(struct xfs_buftarg *target, 229 struct xfs_buf_map *map, int nmaps, 230 xfs_buf_flags_t flags); 231 struct xfs_buf *xfs_buf_read_map(struct xfs_buftarg *target, 232 struct xfs_buf_map *map, int nmaps, 233 xfs_buf_flags_t flags, 234 const struct xfs_buf_ops *ops); 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 struct xfs_buf * 240 xfs_buf_get( 241 struct xfs_buftarg *target, 242 xfs_daddr_t blkno, 243 size_t numblks, 244 xfs_buf_flags_t flags) 245 { 246 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); 247 return xfs_buf_get_map(target, &map, 1, flags); 248 } 249 250 static inline struct xfs_buf * 251 xfs_buf_read( 252 struct xfs_buftarg *target, 253 xfs_daddr_t blkno, 254 size_t numblks, 255 xfs_buf_flags_t flags, 256 const struct xfs_buf_ops *ops) 257 { 258 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); 259 return xfs_buf_read_map(target, &map, 1, flags, ops); 260 } 261 262 static inline void 263 xfs_buf_readahead( 264 struct xfs_buftarg *target, 265 xfs_daddr_t blkno, 266 size_t numblks, 267 const struct xfs_buf_ops *ops) 268 { 269 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); 270 return xfs_buf_readahead_map(target, &map, 1, ops); 271 } 272 273 void xfs_buf_set_empty(struct xfs_buf *bp, size_t numblks); 274 int xfs_buf_associate_memory(struct xfs_buf *bp, void *mem, size_t length); 275 276 struct xfs_buf *xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks, 277 int flags); 278 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr, 279 size_t numblks, int flags, struct xfs_buf **bpp, 280 const struct xfs_buf_ops *ops); 281 void xfs_buf_hold(struct xfs_buf *bp); 282 283 /* Releasing Buffers */ 284 extern void xfs_buf_free(xfs_buf_t *); 285 extern void xfs_buf_rele(xfs_buf_t *); 286 287 /* Locking and Unlocking Buffers */ 288 extern int xfs_buf_trylock(xfs_buf_t *); 289 extern void xfs_buf_lock(xfs_buf_t *); 290 extern void xfs_buf_unlock(xfs_buf_t *); 291 #define xfs_buf_islocked(bp) \ 292 ((bp)->b_sema.count <= 0) 293 294 /* Buffer Read and Write Routines */ 295 extern int xfs_bwrite(struct xfs_buf *bp); 296 extern void xfs_buf_ioend(struct xfs_buf *bp); 297 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error, 298 xfs_failaddr_t failaddr); 299 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address) 300 extern void xfs_buf_ioerror_alert(struct xfs_buf *, const char *func); 301 extern void xfs_buf_submit(struct xfs_buf *bp); 302 extern int xfs_buf_submit_wait(struct xfs_buf *bp); 303 extern void xfs_buf_iomove(xfs_buf_t *, size_t, size_t, void *, 304 xfs_buf_rw_t); 305 #define xfs_buf_zero(bp, off, len) \ 306 xfs_buf_iomove((bp), (off), (len), NULL, XBRW_ZERO) 307 308 /* Buffer Utility Routines */ 309 extern void *xfs_buf_offset(struct xfs_buf *, size_t); 310 extern void xfs_buf_stale(struct xfs_buf *bp); 311 312 /* Delayed Write Buffer Routines */ 313 extern void xfs_buf_delwri_cancel(struct list_head *); 314 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *); 315 extern int xfs_buf_delwri_submit(struct list_head *); 316 extern int xfs_buf_delwri_submit_nowait(struct list_head *); 317 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *); 318 319 /* Buffer Daemon Setup Routines */ 320 extern int xfs_buf_init(void); 321 extern void xfs_buf_terminate(void); 322 323 /* 324 * These macros use the IO block map rather than b_bn. b_bn is now really 325 * just for the buffer cache index for cached buffers. As IO does not use b_bn 326 * anymore, uncached buffers do not use b_bn at all and hence must modify the IO 327 * map directly. Uncached buffers are not allowed to be discontiguous, so this 328 * is safe to do. 329 * 330 * In future, uncached buffers will pass the block number directly to the io 331 * request function and hence these macros will go away at that point. 332 */ 333 #define XFS_BUF_ADDR(bp) ((bp)->b_maps[0].bm_bn) 334 #define XFS_BUF_SET_ADDR(bp, bno) ((bp)->b_maps[0].bm_bn = (xfs_daddr_t)(bno)) 335 336 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref); 337 338 /* 339 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer 340 * up with a reference count of 0 so it will be tossed from the cache when 341 * released. 342 */ 343 static inline void xfs_buf_oneshot(struct xfs_buf *bp) 344 { 345 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1) 346 return; 347 atomic_set(&bp->b_lru_ref, 0); 348 } 349 350 static inline int xfs_buf_ispinned(struct xfs_buf *bp) 351 { 352 return atomic_read(&bp->b_pin_count); 353 } 354 355 static inline void xfs_buf_relse(xfs_buf_t *bp) 356 { 357 xfs_buf_unlock(bp); 358 xfs_buf_rele(bp); 359 } 360 361 static inline int 362 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset) 363 { 364 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length), 365 cksum_offset); 366 } 367 368 static inline void 369 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset) 370 { 371 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), 372 cksum_offset); 373 } 374 375 /* 376 * Handling of buftargs. 377 */ 378 extern xfs_buftarg_t *xfs_alloc_buftarg(struct xfs_mount *, 379 struct block_device *, struct dax_device *); 380 extern void xfs_free_buftarg(struct xfs_buftarg *); 381 extern void xfs_wait_buftarg(xfs_buftarg_t *); 382 extern int xfs_setsize_buftarg(xfs_buftarg_t *, unsigned int); 383 384 #define xfs_getsize_buftarg(buftarg) block_size((buftarg)->bt_bdev) 385 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev) 386 387 #endif /* __XFS_BUF_H__ */ 388