1/* SPDX-License-Identifier: GPL-2.0-or-later */ 2/* Network filesystem support services. 3 * 4 * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved. 5 * Written by David Howells (dhowells@redhat.com) 6 * 7 * See: 8 * 9 * Documentation/filesystems/netfs_library.rst 10 * 11 * for a description of the network filesystem interface declared here. 12 */ 13 14#ifndef _LINUX_NETFS_H 15#define _LINUX_NETFS_H 16 17#include <linux/workqueue.h> 18#include <linux/fs.h> 19#include <linux/pagemap.h> 20#include <linux/uio.h> 21 22enum netfs_sreq_ref_trace; 23 24/* 25 * Overload PG_private_2 to give us PG_fscache - this is used to indicate that 26 * a page is currently backed by a local disk cache 27 */ 28#define folio_test_fscache(folio) folio_test_private_2(folio) 29#define PageFsCache(page) PagePrivate2((page)) 30#define SetPageFsCache(page) SetPagePrivate2((page)) 31#define ClearPageFsCache(page) ClearPagePrivate2((page)) 32#define TestSetPageFsCache(page) TestSetPagePrivate2((page)) 33#define TestClearPageFsCache(page) TestClearPagePrivate2((page)) 34 35/** 36 * folio_start_fscache - Start an fscache write on a folio. 37 * @folio: The folio. 38 * 39 * Call this function before writing a folio to a local cache. Starting a 40 * second write before the first one finishes is not allowed. 41 */ 42static inline void folio_start_fscache(struct folio *folio) 43{ 44 VM_BUG_ON_FOLIO(folio_test_private_2(folio), folio); 45 folio_get(folio); 46 folio_set_private_2(folio); 47} 48 49/** 50 * folio_end_fscache - End an fscache write on a folio. 51 * @folio: The folio. 52 * 53 * Call this function after the folio has been written to the local cache. 54 * This will wake any sleepers waiting on this folio. 55 */ 56static inline void folio_end_fscache(struct folio *folio) 57{ 58 folio_end_private_2(folio); 59} 60 61/** 62 * folio_wait_fscache - Wait for an fscache write on this folio to end. 63 * @folio: The folio. 64 * 65 * If this folio is currently being written to a local cache, wait for 66 * the write to finish. Another write may start after this one finishes, 67 * unless the caller holds the folio lock. 68 */ 69static inline void folio_wait_fscache(struct folio *folio) 70{ 71 folio_wait_private_2(folio); 72} 73 74/** 75 * folio_wait_fscache_killable - Wait for an fscache write on this folio to end. 76 * @folio: The folio. 77 * 78 * If this folio is currently being written to a local cache, wait 79 * for the write to finish or for a fatal signal to be received. 80 * Another write may start after this one finishes, unless the caller 81 * holds the folio lock. 82 * 83 * Return: 84 * - 0 if successful. 85 * - -EINTR if a fatal signal was encountered. 86 */ 87static inline int folio_wait_fscache_killable(struct folio *folio) 88{ 89 return folio_wait_private_2_killable(folio); 90} 91 92static inline void set_page_fscache(struct page *page) 93{ 94 folio_start_fscache(page_folio(page)); 95} 96 97static inline void end_page_fscache(struct page *page) 98{ 99 folio_end_private_2(page_folio(page)); 100} 101 102static inline void wait_on_page_fscache(struct page *page) 103{ 104 folio_wait_private_2(page_folio(page)); 105} 106 107static inline int wait_on_page_fscache_killable(struct page *page) 108{ 109 return folio_wait_private_2_killable(page_folio(page)); 110} 111 112/* Marks used on xarray-based buffers */ 113#define NETFS_BUF_PUT_MARK XA_MARK_0 /* - Page needs putting */ 114#define NETFS_BUF_PAGECACHE_MARK XA_MARK_1 /* - Page needs wb/dirty flag wrangling */ 115 116enum netfs_io_source { 117 NETFS_FILL_WITH_ZEROES, 118 NETFS_DOWNLOAD_FROM_SERVER, 119 NETFS_READ_FROM_CACHE, 120 NETFS_INVALID_READ, 121 NETFS_UPLOAD_TO_SERVER, 122 NETFS_WRITE_TO_CACHE, 123 NETFS_INVALID_WRITE, 124} __mode(byte); 125 126typedef void (*netfs_io_terminated_t)(void *priv, ssize_t transferred_or_error, 127 bool was_async); 128 129/* 130 * Per-inode context. This wraps the VFS inode. 131 */ 132struct netfs_inode { 133 struct inode inode; /* The VFS inode */ 134 const struct netfs_request_ops *ops; 135#if IS_ENABLED(CONFIG_FSCACHE) 136 struct fscache_cookie *cache; 137#endif 138 loff_t remote_i_size; /* Size of the remote file */ 139 loff_t zero_point; /* Size after which we assume there's no data 140 * on the server */ 141 unsigned long flags; 142#define NETFS_ICTX_ODIRECT 0 /* The file has DIO in progress */ 143#define NETFS_ICTX_UNBUFFERED 1 /* I/O should not use the pagecache */ 144#define NETFS_ICTX_WRITETHROUGH 2 /* Write-through caching */ 145#define NETFS_ICTX_NO_WRITE_STREAMING 3 /* Don't engage in write-streaming */ 146}; 147 148/* 149 * A netfs group - for instance a ceph snap. This is marked on dirty pages and 150 * pages marked with a group must be flushed before they can be written under 151 * the domain of another group. 152 */ 153struct netfs_group { 154 refcount_t ref; 155 void (*free)(struct netfs_group *netfs_group); 156}; 157 158/* 159 * Information about a dirty page (attached only if necessary). 160 * folio->private 161 */ 162struct netfs_folio { 163 struct netfs_group *netfs_group; /* Filesystem's grouping marker (or NULL). */ 164 unsigned int dirty_offset; /* Write-streaming dirty data offset */ 165 unsigned int dirty_len; /* Write-streaming dirty data length */ 166}; 167#define NETFS_FOLIO_INFO 0x1UL /* OR'd with folio->private. */ 168 169static inline struct netfs_folio *netfs_folio_info(struct folio *folio) 170{ 171 void *priv = folio_get_private(folio); 172 173 if ((unsigned long)priv & NETFS_FOLIO_INFO) 174 return (struct netfs_folio *)((unsigned long)priv & ~NETFS_FOLIO_INFO); 175 return NULL; 176} 177 178static inline struct netfs_group *netfs_folio_group(struct folio *folio) 179{ 180 struct netfs_folio *finfo; 181 void *priv = folio_get_private(folio); 182 183 finfo = netfs_folio_info(folio); 184 if (finfo) 185 return finfo->netfs_group; 186 return priv; 187} 188 189/* 190 * Resources required to do operations on a cache. 191 */ 192struct netfs_cache_resources { 193 const struct netfs_cache_ops *ops; 194 void *cache_priv; 195 void *cache_priv2; 196 unsigned int debug_id; /* Cookie debug ID */ 197 unsigned int inval_counter; /* object->inval_counter at begin_op */ 198}; 199 200/* 201 * Descriptor for a single component subrequest. Each operation represents an 202 * individual read/write from/to a server, a cache, a journal, etc.. 203 * 204 * The buffer iterator is persistent for the life of the subrequest struct and 205 * the pages it points to can be relied on to exist for the duration. 206 */ 207struct netfs_io_subrequest { 208 struct netfs_io_request *rreq; /* Supervising I/O request */ 209 struct work_struct work; 210 struct list_head rreq_link; /* Link in rreq->subrequests */ 211 struct iov_iter io_iter; /* Iterator for this subrequest */ 212 loff_t start; /* Where to start the I/O */ 213 size_t len; /* Size of the I/O */ 214 size_t transferred; /* Amount of data transferred */ 215 refcount_t ref; 216 short error; /* 0 or error that occurred */ 217 unsigned short debug_index; /* Index in list (for debugging output) */ 218 unsigned int max_nr_segs; /* 0 or max number of segments in an iterator */ 219 enum netfs_io_source source; /* Where to read from/write to */ 220 unsigned long flags; 221#define NETFS_SREQ_COPY_TO_CACHE 0 /* Set if should copy the data to the cache */ 222#define NETFS_SREQ_CLEAR_TAIL 1 /* Set if the rest of the read should be cleared */ 223#define NETFS_SREQ_SHORT_IO 2 /* Set if the I/O was short */ 224#define NETFS_SREQ_SEEK_DATA_READ 3 /* Set if ->read() should SEEK_DATA first */ 225#define NETFS_SREQ_NO_PROGRESS 4 /* Set if we didn't manage to read any data */ 226#define NETFS_SREQ_ONDEMAND 5 /* Set if it's from on-demand read mode */ 227}; 228 229enum netfs_io_origin { 230 NETFS_READAHEAD, /* This read was triggered by readahead */ 231 NETFS_READPAGE, /* This read is a synchronous read */ 232 NETFS_READ_FOR_WRITE, /* This read is to prepare a write */ 233 NETFS_WRITEBACK, /* This write was triggered by writepages */ 234 NETFS_WRITETHROUGH, /* This write was made by netfs_perform_write() */ 235 NETFS_LAUNDER_WRITE, /* This is triggered by ->launder_folio() */ 236 NETFS_UNBUFFERED_WRITE, /* This is an unbuffered write */ 237 NETFS_DIO_READ, /* This is a direct I/O read */ 238 NETFS_DIO_WRITE, /* This is a direct I/O write */ 239 nr__netfs_io_origin 240} __mode(byte); 241 242/* 243 * Descriptor for an I/O helper request. This is used to make multiple I/O 244 * operations to a variety of data stores and then stitch the result together. 245 */ 246struct netfs_io_request { 247 union { 248 struct work_struct work; 249 struct rcu_head rcu; 250 }; 251 struct inode *inode; /* The file being accessed */ 252 struct address_space *mapping; /* The mapping being accessed */ 253 struct kiocb *iocb; /* AIO completion vector */ 254 struct netfs_cache_resources cache_resources; 255 struct list_head proc_link; /* Link in netfs_iorequests */ 256 struct list_head subrequests; /* Contributory I/O operations */ 257 struct iov_iter iter; /* Unencrypted-side iterator */ 258 struct iov_iter io_iter; /* I/O (Encrypted-side) iterator */ 259 void *netfs_priv; /* Private data for the netfs */ 260 struct bio_vec *direct_bv; /* DIO buffer list (when handling iovec-iter) */ 261 unsigned int direct_bv_count; /* Number of elements in direct_bv[] */ 262 unsigned int debug_id; 263 unsigned int rsize; /* Maximum read size (0 for none) */ 264 unsigned int wsize; /* Maximum write size (0 for none) */ 265 unsigned int subreq_counter; /* Next subreq->debug_index */ 266 atomic_t nr_outstanding; /* Number of ops in progress */ 267 atomic_t nr_copy_ops; /* Number of copy-to-cache ops in progress */ 268 size_t submitted; /* Amount submitted for I/O so far */ 269 size_t len; /* Length of the request */ 270 size_t upper_len; /* Length can be extended to here */ 271 size_t transferred; /* Amount to be indicated as transferred */ 272 short error; /* 0 or error that occurred */ 273 enum netfs_io_origin origin; /* Origin of the request */ 274 bool direct_bv_unpin; /* T if direct_bv[] must be unpinned */ 275 loff_t i_size; /* Size of the file */ 276 loff_t start; /* Start position */ 277 pgoff_t no_unlock_folio; /* Don't unlock this folio after read */ 278 refcount_t ref; 279 unsigned long flags; 280#define NETFS_RREQ_INCOMPLETE_IO 0 /* Some ioreqs terminated short or with error */ 281#define NETFS_RREQ_COPY_TO_CACHE 1 /* Need to write to the cache */ 282#define NETFS_RREQ_NO_UNLOCK_FOLIO 2 /* Don't unlock no_unlock_folio on completion */ 283#define NETFS_RREQ_DONT_UNLOCK_FOLIOS 3 /* Don't unlock the folios on completion */ 284#define NETFS_RREQ_FAILED 4 /* The request failed */ 285#define NETFS_RREQ_IN_PROGRESS 5 /* Unlocked when the request completes */ 286#define NETFS_RREQ_WRITE_TO_CACHE 7 /* Need to write to the cache */ 287#define NETFS_RREQ_UPLOAD_TO_SERVER 8 /* Need to write to the server */ 288#define NETFS_RREQ_NONBLOCK 9 /* Don't block if possible (O_NONBLOCK) */ 289#define NETFS_RREQ_BLOCKED 10 /* We blocked */ 290 const struct netfs_request_ops *netfs_ops; 291 void (*cleanup)(struct netfs_io_request *req); 292}; 293 294/* 295 * Operations the network filesystem can/must provide to the helpers. 296 */ 297struct netfs_request_ops { 298 unsigned int io_request_size; /* Alloc size for netfs_io_request struct */ 299 unsigned int io_subrequest_size; /* Alloc size for netfs_io_subrequest struct */ 300 int (*init_request)(struct netfs_io_request *rreq, struct file *file); 301 void (*free_request)(struct netfs_io_request *rreq); 302 void (*free_subrequest)(struct netfs_io_subrequest *rreq); 303 304 /* Read request handling */ 305 void (*expand_readahead)(struct netfs_io_request *rreq); 306 bool (*clamp_length)(struct netfs_io_subrequest *subreq); 307 void (*issue_read)(struct netfs_io_subrequest *subreq); 308 bool (*is_still_valid)(struct netfs_io_request *rreq); 309 int (*check_write_begin)(struct file *file, loff_t pos, unsigned len, 310 struct folio **foliop, void **_fsdata); 311 void (*done)(struct netfs_io_request *rreq); 312 313 /* Modification handling */ 314 void (*update_i_size)(struct inode *inode, loff_t i_size); 315 316 /* Write request handling */ 317 void (*create_write_requests)(struct netfs_io_request *wreq, 318 loff_t start, size_t len); 319 void (*invalidate_cache)(struct netfs_io_request *wreq); 320}; 321 322/* 323 * How to handle reading from a hole. 324 */ 325enum netfs_read_from_hole { 326 NETFS_READ_HOLE_IGNORE, 327 NETFS_READ_HOLE_CLEAR, 328 NETFS_READ_HOLE_FAIL, 329}; 330 331/* 332 * Table of operations for access to a cache. 333 */ 334struct netfs_cache_ops { 335 /* End an operation */ 336 void (*end_operation)(struct netfs_cache_resources *cres); 337 338 /* Read data from the cache */ 339 int (*read)(struct netfs_cache_resources *cres, 340 loff_t start_pos, 341 struct iov_iter *iter, 342 enum netfs_read_from_hole read_hole, 343 netfs_io_terminated_t term_func, 344 void *term_func_priv); 345 346 /* Write data to the cache */ 347 int (*write)(struct netfs_cache_resources *cres, 348 loff_t start_pos, 349 struct iov_iter *iter, 350 netfs_io_terminated_t term_func, 351 void *term_func_priv); 352 353 /* Expand readahead request */ 354 void (*expand_readahead)(struct netfs_cache_resources *cres, 355 loff_t *_start, size_t *_len, loff_t i_size); 356 357 /* Prepare a read operation, shortening it to a cached/uncached 358 * boundary as appropriate. 359 */ 360 enum netfs_io_source (*prepare_read)(struct netfs_io_subrequest *subreq, 361 loff_t i_size); 362 363 /* Prepare a write operation, working out what part of the write we can 364 * actually do. 365 */ 366 int (*prepare_write)(struct netfs_cache_resources *cres, 367 loff_t *_start, size_t *_len, size_t upper_len, 368 loff_t i_size, bool no_space_allocated_yet); 369 370 /* Prepare an on-demand read operation, shortening it to a cached/uncached 371 * boundary as appropriate. 372 */ 373 enum netfs_io_source (*prepare_ondemand_read)(struct netfs_cache_resources *cres, 374 loff_t start, size_t *_len, 375 loff_t i_size, 376 unsigned long *_flags, ino_t ino); 377 378 /* Query the occupancy of the cache in a region, returning where the 379 * next chunk of data starts and how long it is. 380 */ 381 int (*query_occupancy)(struct netfs_cache_resources *cres, 382 loff_t start, size_t len, size_t granularity, 383 loff_t *_data_start, size_t *_data_len); 384}; 385 386/* High-level read API. */ 387ssize_t netfs_unbuffered_read_iter(struct kiocb *iocb, struct iov_iter *iter); 388ssize_t netfs_buffered_read_iter(struct kiocb *iocb, struct iov_iter *iter); 389ssize_t netfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter); 390 391/* High-level write API */ 392ssize_t netfs_perform_write(struct kiocb *iocb, struct iov_iter *iter, 393 struct netfs_group *netfs_group); 394ssize_t netfs_buffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *from, 395 struct netfs_group *netfs_group); 396ssize_t netfs_unbuffered_write_iter(struct kiocb *iocb, struct iov_iter *from); 397ssize_t netfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from); 398 399/* Address operations API */ 400struct readahead_control; 401void netfs_readahead(struct readahead_control *); 402int netfs_read_folio(struct file *, struct folio *); 403int netfs_write_begin(struct netfs_inode *, struct file *, 404 struct address_space *, loff_t pos, unsigned int len, 405 struct folio **, void **fsdata); 406int netfs_writepages(struct address_space *mapping, 407 struct writeback_control *wbc); 408bool netfs_dirty_folio(struct address_space *mapping, struct folio *folio); 409int netfs_unpin_writeback(struct inode *inode, struct writeback_control *wbc); 410void netfs_clear_inode_writeback(struct inode *inode, const void *aux); 411void netfs_invalidate_folio(struct folio *folio, size_t offset, size_t length); 412bool netfs_release_folio(struct folio *folio, gfp_t gfp); 413int netfs_launder_folio(struct folio *folio); 414 415/* VMA operations API. */ 416vm_fault_t netfs_page_mkwrite(struct vm_fault *vmf, struct netfs_group *netfs_group); 417 418/* (Sub)request management API. */ 419void netfs_subreq_terminated(struct netfs_io_subrequest *, ssize_t, bool); 420void netfs_get_subrequest(struct netfs_io_subrequest *subreq, 421 enum netfs_sreq_ref_trace what); 422void netfs_put_subrequest(struct netfs_io_subrequest *subreq, 423 bool was_async, enum netfs_sreq_ref_trace what); 424ssize_t netfs_extract_user_iter(struct iov_iter *orig, size_t orig_len, 425 struct iov_iter *new, 426 iov_iter_extraction_t extraction_flags); 427size_t netfs_limit_iter(const struct iov_iter *iter, size_t start_offset, 428 size_t max_size, size_t max_segs); 429struct netfs_io_subrequest *netfs_create_write_request( 430 struct netfs_io_request *wreq, enum netfs_io_source dest, 431 loff_t start, size_t len, work_func_t worker); 432void netfs_write_subrequest_terminated(void *_op, ssize_t transferred_or_error, 433 bool was_async); 434void netfs_queue_write_request(struct netfs_io_subrequest *subreq); 435 436int netfs_start_io_read(struct inode *inode); 437void netfs_end_io_read(struct inode *inode); 438int netfs_start_io_write(struct inode *inode); 439void netfs_end_io_write(struct inode *inode); 440int netfs_start_io_direct(struct inode *inode); 441void netfs_end_io_direct(struct inode *inode); 442 443/** 444 * netfs_inode - Get the netfs inode context from the inode 445 * @inode: The inode to query 446 * 447 * Get the netfs lib inode context from the network filesystem's inode. The 448 * context struct is expected to directly follow on from the VFS inode struct. 449 */ 450static inline struct netfs_inode *netfs_inode(struct inode *inode) 451{ 452 return container_of(inode, struct netfs_inode, inode); 453} 454 455/** 456 * netfs_inode_init - Initialise a netfslib inode context 457 * @ctx: The netfs inode to initialise 458 * @ops: The netfs's operations list 459 * @use_zero_point: True to use the zero_point read optimisation 460 * 461 * Initialise the netfs library context struct. This is expected to follow on 462 * directly from the VFS inode struct. 463 */ 464static inline void netfs_inode_init(struct netfs_inode *ctx, 465 const struct netfs_request_ops *ops, 466 bool use_zero_point) 467{ 468 ctx->ops = ops; 469 ctx->remote_i_size = i_size_read(&ctx->inode); 470 ctx->zero_point = LLONG_MAX; 471 ctx->flags = 0; 472#if IS_ENABLED(CONFIG_FSCACHE) 473 ctx->cache = NULL; 474#endif 475 /* ->releasepage() drives zero_point */ 476 if (use_zero_point) { 477 ctx->zero_point = ctx->remote_i_size; 478 mapping_set_release_always(ctx->inode.i_mapping); 479 } 480} 481 482/** 483 * netfs_resize_file - Note that a file got resized 484 * @ctx: The netfs inode being resized 485 * @new_i_size: The new file size 486 * @changed_on_server: The change was applied to the server 487 * 488 * Inform the netfs lib that a file got resized so that it can adjust its state. 489 */ 490static inline void netfs_resize_file(struct netfs_inode *ctx, loff_t new_i_size, 491 bool changed_on_server) 492{ 493 if (changed_on_server) 494 ctx->remote_i_size = new_i_size; 495 if (new_i_size < ctx->zero_point) 496 ctx->zero_point = new_i_size; 497} 498 499/** 500 * netfs_i_cookie - Get the cache cookie from the inode 501 * @ctx: The netfs inode to query 502 * 503 * Get the caching cookie (if enabled) from the network filesystem's inode. 504 */ 505static inline struct fscache_cookie *netfs_i_cookie(struct netfs_inode *ctx) 506{ 507#if IS_ENABLED(CONFIG_FSCACHE) 508 return ctx->cache; 509#else 510 return NULL; 511#endif 512} 513 514#endif /* _LINUX_NETFS_H */ 515