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 #include <linux/rolling_buffer.h>
22
23 enum netfs_sreq_ref_trace;
24 typedef struct mempool mempool_t;
25 struct folio_queue;
26
27 /**
28 * folio_start_private_2 - Start an fscache write on a folio. [DEPRECATED]
29 * @folio: The folio.
30 *
31 * Call this function before writing a folio to a local cache. Starting a
32 * second write before the first one finishes is not allowed.
33 *
34 * Note that this should no longer be used.
35 */
folio_start_private_2(struct folio * folio)36 static inline void folio_start_private_2(struct folio *folio)
37 {
38 VM_BUG_ON_FOLIO(folio_test_private_2(folio), folio);
39 folio_get(folio);
40 folio_set_private_2(folio);
41 }
42
43 enum netfs_io_source {
44 NETFS_SOURCE_UNKNOWN,
45 NETFS_FILL_WITH_ZEROES,
46 NETFS_DOWNLOAD_FROM_SERVER,
47 NETFS_READ_FROM_CACHE,
48 NETFS_INVALID_READ,
49 NETFS_UPLOAD_TO_SERVER,
50 NETFS_WRITE_TO_CACHE,
51 } __mode(byte);
52
53 typedef void (*netfs_io_terminated_t)(void *priv, ssize_t transferred_or_error);
54
55 /*
56 * Per-inode context. This wraps the VFS inode.
57 */
58 struct netfs_inode {
59 struct inode inode; /* The VFS inode */
60 const struct netfs_request_ops *ops;
61 #if IS_ENABLED(CONFIG_FSCACHE)
62 struct fscache_cookie *cache;
63 #endif
64 struct list_head wb_queue; /* Queue of processes wanting to do writeback */
65 loff_t _remote_i_size; /* Size of the remote file */
66 loff_t _zero_point; /* Size after which we assume there's no data
67 * on the server */
68 spinlock_t lock; /* Lock covering wb_queue */
69 atomic_t io_count; /* Number of outstanding reqs */
70 unsigned long flags;
71 #define NETFS_ICTX_ODIRECT 0 /* The file has DIO in progress */
72 #define NETFS_ICTX_UNBUFFERED 1 /* I/O should not use the pagecache */
73 #define NETFS_ICTX_WB_LOCK 2 /* Writeback serialisation lock */
74 #define NETFS_ICTX_MODIFIED_ATTR 3 /* Indicate change in mtime/ctime */
75 #define NETFS_ICTX_SINGLE_NO_UPLOAD 4 /* Monolithic payload, cache but no upload */
76 };
77
78 /*
79 * A netfs group - for instance a ceph snap. This is marked on dirty pages and
80 * pages marked with a group must be flushed before they can be written under
81 * the domain of another group.
82 */
83 struct netfs_group {
84 refcount_t ref;
85 void (*free)(struct netfs_group *netfs_group);
86 };
87
88 /*
89 * Information about a dirty page (attached only if necessary).
90 * folio->private
91 */
92 struct netfs_folio {
93 struct netfs_group *netfs_group; /* Filesystem's grouping marker (or NULL). */
94 unsigned int dirty_offset; /* Write-streaming dirty data offset */
95 unsigned int dirty_len; /* Write-streaming dirty data length */
96 };
97 #define NETFS_FOLIO_INFO 0x1UL /* OR'd with folio->private. */
98 #define NETFS_FOLIO_COPY_TO_CACHE ((struct netfs_group *)0x356UL) /* Write to the cache only */
99
netfs_is_folio_info(const void * priv)100 static inline bool netfs_is_folio_info(const void *priv)
101 {
102 return (unsigned long)priv & NETFS_FOLIO_INFO;
103 }
104
__netfs_folio_info(const void * priv)105 static inline struct netfs_folio *__netfs_folio_info(const void *priv)
106 {
107 if (netfs_is_folio_info(priv))
108 return (struct netfs_folio *)((unsigned long)priv & ~NETFS_FOLIO_INFO);
109 return NULL;
110 }
111
netfs_folio_info(struct folio * folio)112 static inline struct netfs_folio *netfs_folio_info(struct folio *folio)
113 {
114 return __netfs_folio_info(folio_get_private(folio));
115 }
116
netfs_folio_group(struct folio * folio)117 static inline struct netfs_group *netfs_folio_group(struct folio *folio)
118 {
119 struct netfs_folio *finfo;
120 void *priv = folio_get_private(folio);
121
122 finfo = netfs_folio_info(folio);
123 if (finfo)
124 return finfo->netfs_group;
125 return priv;
126 }
127
128 /*
129 * Stream of I/O subrequests going to a particular destination, such as the
130 * server or the local cache. This is mainly intended for writing where we may
131 * have to write to multiple destinations concurrently.
132 */
133 struct netfs_io_stream {
134 /* Submission tracking */
135 struct netfs_io_subrequest *construct; /* Op being constructed */
136 size_t sreq_max_len; /* Maximum size of a subrequest */
137 unsigned int sreq_max_segs; /* 0 or max number of segments in an iterator */
138 unsigned int submit_off; /* Folio offset we're submitting from */
139 unsigned int submit_len; /* Amount of data left to submit */
140 unsigned int submit_extendable_to; /* Amount I/O can be rounded up to */
141 void (*prepare_write)(struct netfs_io_subrequest *subreq);
142 void (*issue_write)(struct netfs_io_subrequest *subreq);
143 /* Collection tracking */
144 struct list_head subrequests; /* Contributory I/O operations */
145 unsigned long long collected_to; /* Position we've collected results to */
146 size_t transferred; /* The amount transferred from this stream */
147 unsigned short error; /* Aggregate error for the stream */
148 enum netfs_io_source source; /* Where to read from/write to */
149 unsigned char stream_nr; /* Index of stream in parent table */
150 bool avail; /* T if stream is available */
151 bool active; /* T if stream is active */
152 bool need_retry; /* T if this stream needs retrying */
153 bool failed; /* T if this stream failed */
154 bool transferred_valid; /* T is ->transferred is valid */
155 };
156
157 /*
158 * Resources required to do operations on a cache.
159 */
160 struct netfs_cache_resources {
161 const struct netfs_cache_ops *ops;
162 void *cache_priv;
163 void *cache_priv2;
164 unsigned int debug_id; /* Cookie debug ID */
165 unsigned int inval_counter; /* object->inval_counter at begin_op */
166 };
167
168 /*
169 * Descriptor for a single component subrequest. Each operation represents an
170 * individual read/write from/to a server, a cache, a journal, etc..
171 *
172 * The buffer iterator is persistent for the life of the subrequest struct and
173 * the pages it points to can be relied on to exist for the duration.
174 */
175 struct netfs_io_subrequest {
176 struct netfs_io_request *rreq; /* Supervising I/O request */
177 struct work_struct work;
178 struct list_head rreq_link; /* Link in rreq->subrequests */
179 struct iov_iter io_iter; /* Iterator for this subrequest */
180 unsigned long long start; /* Where to start the I/O */
181 size_t len; /* Size of the I/O */
182 size_t transferred; /* Amount of data transferred */
183 refcount_t ref;
184 short error; /* 0 or error that occurred */
185 unsigned short debug_index; /* Index in list (for debugging output) */
186 unsigned int nr_segs; /* Number of segs in io_iter */
187 u8 retry_count; /* The number of retries (0 on initial pass) */
188 enum netfs_io_source source; /* Where to read from/write to */
189 unsigned char stream_nr; /* I/O stream this belongs to */
190 unsigned long flags;
191 #define NETFS_SREQ_COPY_TO_CACHE 0 /* Set if should copy the data to the cache */
192 #define NETFS_SREQ_CLEAR_TAIL 1 /* Set if the rest of the read should be cleared */
193 #define NETFS_SREQ_MADE_PROGRESS 4 /* Set if we transferred at least some data */
194 #define NETFS_SREQ_BOUNDARY 6 /* Set if ends on hard boundary (eg. ceph object) */
195 #define NETFS_SREQ_HIT_EOF 7 /* Set if short due to EOF */
196 #define NETFS_SREQ_IN_PROGRESS 8 /* Unlocked when the subrequest completes */
197 #define NETFS_SREQ_NEED_RETRY 9 /* Set if the filesystem requests a retry */
198 #define NETFS_SREQ_FAILED 10 /* Set if the subreq failed unretryably */
199 };
200
201 enum netfs_io_origin {
202 NETFS_READAHEAD, /* This read was triggered by readahead */
203 NETFS_READPAGE, /* This read is a synchronous read */
204 NETFS_READ_GAPS, /* This read is a synchronous read to fill gaps */
205 NETFS_READ_SINGLE, /* This read should be treated as a single object */
206 NETFS_READ_FOR_WRITE, /* This read is to prepare a write */
207 NETFS_UNBUFFERED_READ, /* This is an unbuffered read */
208 NETFS_DIO_READ, /* This is a direct I/O read */
209 NETFS_WRITEBACK, /* This write was triggered by writepages */
210 NETFS_WRITEBACK_SINGLE, /* This monolithic write was triggered by writepages */
211 NETFS_WRITETHROUGH, /* This write was made by netfs_perform_write() */
212 NETFS_UNBUFFERED_WRITE, /* This is an unbuffered write */
213 NETFS_DIO_WRITE, /* This is a direct I/O write */
214 NETFS_PGPRIV2_COPY_TO_CACHE, /* [DEPRECATED] This is writing read data to the cache */
215 nr__netfs_io_origin
216 } __mode(byte);
217
218 /*
219 * Descriptor for an I/O helper request. This is used to make multiple I/O
220 * operations to a variety of data stores and then stitch the result together.
221 */
222 struct netfs_io_request {
223 union {
224 struct work_struct cleanup_work; /* Deferred cleanup work */
225 struct rcu_head rcu;
226 };
227 struct work_struct work; /* Result collector work */
228 struct inode *inode; /* The file being accessed */
229 struct address_space *mapping; /* The mapping being accessed */
230 struct kiocb *iocb; /* AIO completion vector */
231 struct netfs_cache_resources cache_resources;
232 struct netfs_io_request *copy_to_cache; /* Request to write just-read data to the cache */
233 #ifdef CONFIG_PROC_FS
234 struct list_head proc_link; /* Link in netfs_iorequests */
235 #endif
236 struct netfs_io_stream io_streams[2]; /* Streams of parallel I/O operations */
237 #define NR_IO_STREAMS 2 //wreq->nr_io_streams
238 struct netfs_group *group; /* Writeback group being written back */
239 struct rolling_buffer buffer; /* Unencrypted buffer */
240 #define NETFS_ROLLBUF_PUT_MARK ROLLBUF_MARK_1
241 #define NETFS_ROLLBUF_PAGECACHE_MARK ROLLBUF_MARK_2
242 wait_queue_head_t waitq; /* Processor waiter */
243 void *netfs_priv; /* Private data for the netfs */
244 void *netfs_priv2; /* Private data for the netfs */
245 struct bio_vec *direct_bv; /* DIO buffer list (when handling iovec-iter) */
246 unsigned long long submitted; /* Amount submitted for I/O so far */
247 unsigned long long len; /* Length of the request */
248 size_t transferred; /* Amount to be indicated as transferred */
249 long error; /* 0 or error that occurred */
250 unsigned long long i_size; /* Size of the file */
251 unsigned long long start; /* Start position */
252 atomic64_t issued_to; /* Write issuer folio cursor */
253 unsigned long long collected_to; /* Point we've collected to */
254 unsigned long long cleaned_to; /* Position we've cleaned folios to */
255 unsigned long long abandon_to; /* Position to abandon folios to */
256 const struct folio *no_unlock_folio; /* Don't unlock this folio after read */
257 gfp_t gfp; /* GFP flags to use */
258 unsigned int direct_bv_count; /* Number of elements in direct_bv[] */
259 unsigned int debug_id;
260 unsigned int rsize; /* Maximum read size (0 for none) */
261 unsigned int wsize; /* Maximum write size (0 for none) */
262 atomic_t subreq_counter; /* Next subreq->debug_index */
263 unsigned int nr_group_rel; /* Number of refs to release on ->group */
264 spinlock_t lock; /* Lock for queuing subreqs */
265 unsigned char front_folio_order; /* Order (size) of front folio */
266 enum netfs_io_origin origin; /* Origin of the request */
267 bool direct_bv_unpin; /* T if direct_bv[] must be unpinned */
268 refcount_t ref;
269 unsigned long flags;
270 #define NETFS_RREQ_IN_PROGRESS 0 /* Unlocked when the request completes (has ref) */
271 #define NETFS_RREQ_ALL_QUEUED 1 /* All subreqs are now queued */
272 #define NETFS_RREQ_PAUSE 2 /* Pause subrequest generation */
273 #define NETFS_RREQ_FAILED 3 /* The request failed */
274 #define NETFS_RREQ_RETRYING 4 /* Set if we're in the retry path */
275 #define NETFS_RREQ_SHORT_TRANSFER 5 /* Set if we have a short transfer */
276 #define NETFS_RREQ_OFFLOAD_COLLECTION 8 /* Offload collection to workqueue */
277 #define NETFS_RREQ_NO_UNLOCK_FOLIO 9 /* Don't unlock no_unlock_folio on completion */
278 #define NETFS_RREQ_FOLIO_COPY_TO_CACHE 10 /* Copy current folio to cache from read */
279 #define NETFS_RREQ_UPLOAD_TO_SERVER 11 /* Need to write to the server */
280 #define NETFS_RREQ_USE_IO_ITER 12 /* Use ->io_iter rather than ->i_pages */
281 #define NETFS_RREQ_USE_PGPRIV2 31 /* [DEPRECATED] Use PG_private_2 to mark
282 * write to cache on read */
283 const struct netfs_request_ops *netfs_ops;
284 };
285
286 /*
287 * Operations the network filesystem can/must provide to the helpers.
288 */
289 struct netfs_request_ops {
290 mempool_t *request_pool;
291 mempool_t *subrequest_pool;
292 int (*init_request)(struct netfs_io_request *rreq, struct file *file);
293 void (*free_request)(struct netfs_io_request *rreq);
294 void (*free_subrequest)(struct netfs_io_subrequest *rreq);
295
296 /* Read request handling */
297 void (*expand_readahead)(struct netfs_io_request *rreq);
298 int (*prepare_read)(struct netfs_io_subrequest *subreq);
299 void (*issue_read)(struct netfs_io_subrequest *subreq);
300 bool (*is_still_valid)(struct netfs_io_request *rreq);
301 int (*check_write_begin)(struct file *file, loff_t pos, unsigned len,
302 struct folio **foliop, void **_fsdata);
303 void (*done)(struct netfs_io_request *rreq);
304
305 /* Modification handling */
306 void (*update_i_size)(struct inode *inode, loff_t i_size);
307 void (*post_modify)(struct inode *inode);
308
309 /* Write request handling */
310 void (*begin_writeback)(struct netfs_io_request *wreq);
311 void (*prepare_write)(struct netfs_io_subrequest *subreq);
312 void (*issue_write)(struct netfs_io_subrequest *subreq);
313 void (*retry_request)(struct netfs_io_request *wreq, struct netfs_io_stream *stream);
314 void (*invalidate_cache)(struct netfs_io_request *wreq);
315 };
316
317 /*
318 * How to handle reading from a hole.
319 */
320 enum netfs_read_from_hole {
321 NETFS_READ_HOLE_IGNORE,
322 NETFS_READ_HOLE_FAIL,
323 };
324
325 /*
326 * Table of operations for access to a cache.
327 */
328 struct netfs_cache_ops {
329 /* End an operation */
330 void (*end_operation)(struct netfs_cache_resources *cres);
331
332 /* Read data from the cache */
333 int (*read)(struct netfs_cache_resources *cres,
334 loff_t start_pos,
335 struct iov_iter *iter,
336 enum netfs_read_from_hole read_hole,
337 netfs_io_terminated_t term_func,
338 void *term_func_priv);
339
340 /* Write data to the cache */
341 int (*write)(struct netfs_cache_resources *cres,
342 loff_t start_pos,
343 struct iov_iter *iter,
344 netfs_io_terminated_t term_func,
345 void *term_func_priv);
346
347 /* Write data to the cache from a netfs subrequest. */
348 void (*issue_write)(struct netfs_io_subrequest *subreq);
349
350 /* Expand readahead request */
351 void (*expand_readahead)(struct netfs_cache_resources *cres,
352 unsigned long long *_start,
353 unsigned long long *_len,
354 unsigned long long i_size);
355
356 /* Prepare a read operation, shortening it to a cached/uncached
357 * boundary as appropriate.
358 */
359 enum netfs_io_source (*prepare_read)(struct netfs_io_subrequest *subreq,
360 unsigned long long i_size);
361
362 /* Prepare a write subrequest, working out if we're allowed to do it
363 * and finding out the maximum amount of data to gather before
364 * attempting to submit. If we're not permitted to do it, the
365 * subrequest should be marked failed.
366 */
367 void (*prepare_write_subreq)(struct netfs_io_subrequest *subreq);
368
369 /* Prepare a write operation, working out what part of the write we can
370 * actually do.
371 */
372 int (*prepare_write)(struct netfs_cache_resources *cres,
373 loff_t *_start, size_t *_len, size_t upper_len,
374 loff_t i_size, bool no_space_allocated_yet);
375
376 /* Query the occupancy of the cache in a region, returning where the
377 * next chunk of data starts and how long it is.
378 */
379 int (*query_occupancy)(struct netfs_cache_resources *cres,
380 loff_t start, size_t len, size_t granularity,
381 loff_t *_data_start, size_t *_data_len);
382 };
383
384 /* High-level read API. */
385 ssize_t netfs_unbuffered_read_iter_locked(struct kiocb *iocb, struct iov_iter *iter);
386 ssize_t netfs_unbuffered_read_iter(struct kiocb *iocb, struct iov_iter *iter);
387 ssize_t netfs_buffered_read_iter(struct kiocb *iocb, struct iov_iter *iter);
388 ssize_t netfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter);
389
390 /* High-level write API */
391 ssize_t netfs_perform_write(struct kiocb *iocb, struct iov_iter *iter,
392 struct netfs_group *netfs_group);
393 ssize_t netfs_buffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *from,
394 struct netfs_group *netfs_group);
395 ssize_t netfs_unbuffered_write_iter(struct kiocb *iocb, struct iov_iter *from);
396 ssize_t netfs_unbuffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *iter,
397 struct netfs_group *netfs_group);
398 ssize_t netfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from);
399
400 /* Single, monolithic object read/write API. */
401 void netfs_single_mark_inode_dirty(struct inode *inode);
402 ssize_t netfs_read_single(struct inode *inode, struct file *file, struct iov_iter *iter);
403 int netfs_writeback_single(struct address_space *mapping,
404 struct writeback_control *wbc,
405 struct iov_iter *iter);
406
407 /* Address operations API */
408 struct readahead_control;
409 void netfs_readahead(struct readahead_control *);
410 int netfs_read_folio(struct file *, struct folio *);
411 int netfs_write_begin(struct netfs_inode *, struct file *,
412 struct address_space *, loff_t pos, unsigned int len,
413 struct folio **, void **fsdata);
414 int netfs_writepages(struct address_space *mapping,
415 struct writeback_control *wbc);
416 bool netfs_dirty_folio(struct address_space *mapping, struct folio *folio);
417 int netfs_unpin_writeback(struct inode *inode, struct writeback_control *wbc);
418 void netfs_clear_inode_writeback(struct inode *inode, const void *aux);
419 void netfs_invalidate_folio(struct folio *folio, size_t offset, size_t length);
420 bool netfs_release_folio(struct folio *folio, gfp_t gfp);
421
422 /* VMA operations API. */
423 vm_fault_t netfs_page_mkwrite(struct vm_fault *vmf, struct netfs_group *netfs_group);
424
425 /* (Sub)request management API. */
426 void netfs_read_subreq_progress(struct netfs_io_subrequest *subreq);
427 void netfs_read_subreq_terminated(struct netfs_io_subrequest *subreq);
428 void netfs_get_subrequest(struct netfs_io_subrequest *subreq,
429 enum netfs_sreq_ref_trace what);
430 void netfs_put_subrequest(struct netfs_io_subrequest *subreq,
431 enum netfs_sreq_ref_trace what);
432 ssize_t netfs_extract_user_iter(struct iov_iter *orig, size_t orig_len,
433 struct iov_iter *new,
434 iov_iter_extraction_t extraction_flags);
435 size_t netfs_limit_iter(const struct iov_iter *iter, size_t start_offset,
436 size_t max_size, size_t max_segs);
437 void netfs_prepare_write_failed(struct netfs_io_subrequest *subreq);
438 void netfs_write_subrequest_terminated(void *_op, ssize_t transferred_or_error);
439
440 int netfs_start_io_read(struct inode *inode);
441 void netfs_end_io_read(struct inode *inode);
442 int netfs_start_io_write(struct inode *inode);
443 void netfs_end_io_write(struct inode *inode);
444 int netfs_start_io_direct(struct inode *inode);
445 void netfs_end_io_direct(struct inode *inode);
446
447 /* Miscellaneous APIs. */
448 struct folio_queue *netfs_folioq_alloc(unsigned int rreq_id, gfp_t gfp,
449 unsigned int trace /*enum netfs_folioq_trace*/);
450 void netfs_folioq_free(struct folio_queue *folioq,
451 unsigned int trace /*enum netfs_trace_folioq*/);
452
453 /* Buffer wrangling helpers API. */
454 int netfs_alloc_folioq_buffer(struct address_space *mapping,
455 struct folio_queue **_buffer,
456 size_t *_cur_size, ssize_t size, gfp_t gfp);
457 void netfs_free_folioq_buffer(struct folio_queue *fq);
458
459 /* Writeback exclusion API. */
460 bool netfs_wb_begin(struct netfs_inode *ictx, bool nowait);
461 void netfs_wb_end(struct netfs_inode *ictx);
462
463 /**
464 * netfs_inode - Get the netfs inode context from the inode
465 * @inode: The inode to query
466 *
467 * Get the netfs lib inode context from the network filesystem's inode. The
468 * context struct is expected to directly follow on from the VFS inode struct.
469 */
netfs_inode(struct inode * inode)470 static inline struct netfs_inode *netfs_inode(struct inode *inode)
471 {
472 return container_of(inode, struct netfs_inode, inode);
473 }
474
475 /**
476 * netfs_read_remote_i_size - Read remote_i_size safely
477 * @inode: The inode to access
478 *
479 * Read remote_i_size safely without the potential for tearing on 32-bit
480 * arches.
481 *
482 * NOTE: in a 32bit arch with a preemptable kernel and an UP compile the
483 * i_size_read/write must be atomic with respect to the local cpu (unlike with
484 * preempt disabled), but they don't need to be atomic with respect to other
485 * cpus like in true SMP (so they need either to either locally disable irq
486 * around the read or for example on x86 they can be still implemented as a
487 * cmpxchg8b without the need of the lock prefix). For SMP compiles and 64bit
488 * archs it makes no difference if preempt is enabled or not.
489 */
netfs_read_remote_i_size(const struct inode * inode)490 static inline unsigned long long netfs_read_remote_i_size(const struct inode *inode)
491 {
492 const struct netfs_inode *ictx = container_of(inode, struct netfs_inode, inode);
493 unsigned long long remote_i_size;
494
495 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
496 unsigned int seq;
497
498 do {
499 seq = read_seqcount_begin(&inode->i_size_seqcount);
500 remote_i_size = ictx->_remote_i_size;
501 } while (read_seqcount_retry(&inode->i_size_seqcount, seq));
502 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
503 preempt_disable();
504 remote_i_size = ictx->_remote_i_size;
505 preempt_enable();
506 #else
507 /* Pairs with smp_store_release() in netfs_write_remote_i_size() */
508 remote_i_size = smp_load_acquire(&ictx->_remote_i_size);
509 #endif
510 return remote_i_size;
511 }
512
513 /*
514 * netfs_write_remote_i_size - Set remote_i_size safely
515 * @inode: The inode to access
516 * @remote_i_size: The new value for the size of the file on the server
517 *
518 * Set remote_i_size safely without the potential for tearing on 32-bit arches.
519 *
520 * Context: The caller must hold inode->i_lock.
521 *
522 * NOTE: unlike netfs_read_remote_i_size(), netfs_write_remote_i_size() does
523 * need locking around it (normally i_rwsem), otherwise on 32bit/SMP an update
524 * of i_size_seqcount can be lost, resulting in subsequent i_size_read() calls
525 * spinning forever.
526 */
netfs_write_remote_i_size(struct inode * inode,unsigned long long remote_i_size)527 static inline void netfs_write_remote_i_size(struct inode *inode,
528 unsigned long long remote_i_size)
529 {
530 struct netfs_inode *ictx = netfs_inode(inode);
531
532 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
533 write_seqcount_begin(&inode->i_size_seqcount);
534 ictx->_remote_i_size = remote_i_size;
535 write_seqcount_end(&inode->i_size_seqcount);
536 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
537 preempt_disable();
538 ictx->_remote_i_size = remote_i_size;
539 preempt_enable();
540 #else
541 /*
542 * Pairs with smp_load_acquire() in netfs_read_remote_i_size() to
543 * ensure changes related to inode size (such as page contents) are
544 * visible before we see the changed inode size.
545 */
546 smp_store_release(&ictx->_remote_i_size, remote_i_size);
547 #endif
548 }
549
550 /**
551 * netfs_read_zero_point - Read zero_point safely
552 * @inode: The inode to access
553 *
554 * Read zero_point safely without the potential for tearing on 32-bit
555 * arches.
556 *
557 * NOTE: in a 32bit arch with a preemptable kernel and an UP compile the
558 * i_size_read/write must be atomic with respect to the local cpu (unlike with
559 * preempt disabled), but they don't need to be atomic with respect to other
560 * cpus like in true SMP (so they need either to either locally disable irq
561 * around the read or for example on x86 they can be still implemented as a
562 * cmpxchg8b without the need of the lock prefix). For SMP compiles and 64bit
563 * archs it makes no difference if preempt is enabled or not.
564 */
netfs_read_zero_point(const struct inode * inode)565 static inline unsigned long long netfs_read_zero_point(const struct inode *inode)
566 {
567 struct netfs_inode *ictx = container_of(inode, struct netfs_inode, inode);
568 unsigned long long zero_point;
569
570 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
571 unsigned int seq;
572
573 do {
574 seq = read_seqcount_begin(&inode->i_size_seqcount);
575 zero_point = ictx->_zero_point;
576 } while (read_seqcount_retry(&inode->i_size_seqcount, seq));
577 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
578 preempt_disable();
579 zero_point = ictx->_zero_point;
580 preempt_enable();
581 #else
582 /* Pairs with smp_store_release() in netfs_write_zero_point() */
583 zero_point = smp_load_acquire(&ictx->_zero_point);
584 #endif
585 return zero_point;
586 }
587
588 /*
589 * netfs_write_zero_point - Set zero_point safely
590 * @inode: The inode to access
591 * @zero_point: The new value for the point beyond which the server has no data
592 *
593 * Set zero_point safely without the potential for tearing on 32-bit arches.
594 *
595 * Context: The caller must hold inode->i_lock.
596 *
597 * NOTE: unlike netfs_read_zero_point(), netfs_write_zero_point() does need
598 * locking around it (normally i_rwsem), otherwise on 32bit/SMP an update of
599 * i_size_seqcount can be lost, resulting in subsequent read calls spinning
600 * forever.
601 */
netfs_write_zero_point(struct inode * inode,unsigned long long zero_point)602 static inline void netfs_write_zero_point(struct inode *inode,
603 unsigned long long zero_point)
604 {
605 struct netfs_inode *ictx = netfs_inode(inode);
606
607 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
608 write_seqcount_begin(&inode->i_size_seqcount);
609 ictx->_zero_point = zero_point;
610 write_seqcount_end(&inode->i_size_seqcount);
611 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
612 preempt_disable();
613 ictx->_zero_point = zero_point;
614 preempt_enable();
615 #else
616 /*
617 * Pairs with smp_load_acquire() in netfs_read_zero_point() to
618 * ensure changes related to inode size (such as page contents) are
619 * visible before we see the changed inode size.
620 */
621 smp_store_release(&ictx->_zero_point, zero_point);
622 #endif
623 }
624
625 /**
626 * netfs_read_sizes - Read remote_i_size and zero_point safely
627 * @inode: The inode to access
628 * @i_size: Where to return the local file size.
629 * @remote_i_size: Where to return the size of the file on the server
630 * @zero_point: Where to return the the point beyond which the server has no data
631 *
632 * Read remote_i_size and zero_point safely without the potential for tearing
633 * on 32-bit arches.
634 *
635 * NOTE: in a 32bit arch with a preemptable kernel and an UP compile the
636 * i_size_read/write must be atomic with respect to the local cpu (unlike with
637 * preempt disabled), but they don't need to be atomic with respect to other
638 * cpus like in true SMP (so they need either to either locally disable irq
639 * around the read or for example on x86 they can be still implemented as a
640 * cmpxchg8b without the need of the lock prefix). For SMP compiles and 64bit
641 * archs it makes no difference if preempt is enabled or not.
642 */
netfs_read_sizes(const struct inode * inode,unsigned long long * i_size,unsigned long long * remote_i_size,unsigned long long * zero_point)643 static inline void netfs_read_sizes(const struct inode *inode,
644 unsigned long long *i_size,
645 unsigned long long *remote_i_size,
646 unsigned long long *zero_point)
647 {
648 const struct netfs_inode *ictx = container_of(inode, struct netfs_inode, inode);
649 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
650 unsigned int seq;
651
652 do {
653 seq = read_seqcount_begin(&inode->i_size_seqcount);
654 *i_size = inode->i_size;
655 *remote_i_size = ictx->_remote_i_size;
656 *zero_point = ictx->_zero_point;
657 } while (read_seqcount_retry(&inode->i_size_seqcount, seq));
658 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
659 preempt_disable();
660 *i_size = inode->i_size;
661 *remote_i_size = ictx->_remote_i_size;
662 *zero_point = ictx->_zero_point;
663 preempt_enable();
664 #else
665 /* Pairs with smp_store_release() in i_size_write() */
666 *i_size = smp_load_acquire(&inode->i_size);
667 /* Pairs with smp_store_release() in netfs_write_remote_i_size() */
668 *remote_i_size = smp_load_acquire(&ictx->_remote_i_size);
669 /* Pairs with smp_store_release() in netfs_write_zero_point() */
670 *zero_point = smp_load_acquire(&ictx->_zero_point);
671 #endif
672 }
673
674 /*
675 * netfs_write_sizes - Set i_size, remote_i_size and zero_point safely
676 * @inode: The inode to access
677 * @i_size: The new value for the local size of the file
678 * @remote_i_size: The new value for the size of the file on the server
679 * @zero_point: The new value for the point beyond which the server has no data
680 *
681 * Set both remote_i_size and zero_point safely without the potential for
682 * tearing on 32-bit arches.
683 *
684 * Context: The caller must hold inode->i_lock.
685 *
686 * NOTE: unlike netfs_read_zero_point(), netfs_write_zero_point() does need
687 * locking around it (normally i_rwsem), otherwise on 32bit/SMP an update of
688 * i_size_seqcount can be lost, resulting in subsequent read calls spinning
689 * forever.
690 */
netfs_write_sizes(struct inode * inode,unsigned long long i_size,unsigned long long remote_i_size,unsigned long long zero_point)691 static inline void netfs_write_sizes(struct inode *inode,
692 unsigned long long i_size,
693 unsigned long long remote_i_size,
694 unsigned long long zero_point)
695 {
696 struct netfs_inode *ictx = netfs_inode(inode);
697
698 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
699 write_seqcount_begin(&inode->i_size_seqcount);
700 inode->i_size = i_size;
701 ictx->_remote_i_size = remote_i_size;
702 ictx->_zero_point = zero_point;
703 write_seqcount_end(&inode->i_size_seqcount);
704 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
705 preempt_disable();
706 inode->i_size = i_size;
707 ictx->_remote_i_size = remote_i_size;
708 ictx->_zero_point = zero_point;
709 preempt_enable();
710 #else
711 /*
712 * Pairs with smp_load_acquire() in i_size_read(),
713 * netfs_read_remote_i_size() and netfs_read_zero_point() to ensure
714 * changes related to inode size (such as page contents) are visible
715 * before we see the changed inode size.
716 */
717 smp_store_release(&inode->i_size, i_size);
718 smp_store_release(&ictx->_remote_i_size, remote_i_size);
719 smp_store_release(&ictx->_zero_point, zero_point);
720 #endif
721 }
722
723 /**
724 * netfs_inode_init - Initialise a netfslib inode context
725 * @ctx: The netfs inode to initialise
726 * @ops: The netfs's operations list
727 * @use_zero_point: True to use the zero_point read optimisation
728 *
729 * Initialise the netfs library context struct. This is expected to follow on
730 * directly from the VFS inode struct.
731 */
netfs_inode_init(struct netfs_inode * ctx,const struct netfs_request_ops * ops,bool use_zero_point)732 static inline void netfs_inode_init(struct netfs_inode *ctx,
733 const struct netfs_request_ops *ops,
734 bool use_zero_point)
735 {
736 ctx->ops = ops;
737 ctx->_remote_i_size = i_size_read(&ctx->inode);
738 ctx->_zero_point = LLONG_MAX;
739 ctx->flags = 0;
740 atomic_set(&ctx->io_count, 0);
741 #if IS_ENABLED(CONFIG_FSCACHE)
742 ctx->cache = NULL;
743 #endif
744 INIT_LIST_HEAD(&ctx->wb_queue);
745 spin_lock_init(&ctx->lock);
746 /* ->releasepage() drives zero_point */
747 if (use_zero_point) {
748 ctx->_zero_point = ctx->_remote_i_size;
749 mapping_set_release_always(ctx->inode.i_mapping);
750 }
751 }
752
753 /**
754 * netfs_resize_file - Note that a file got resized
755 * @ictx: The netfs inode being resized
756 * @new_i_size: The new file size
757 * @changed_on_server: The change was applied to the server
758 *
759 * Inform the netfs lib that a file got resized so that it can adjust its state.
760 */
netfs_resize_file(struct netfs_inode * ictx,unsigned long long new_i_size,bool changed_on_server)761 static inline void netfs_resize_file(struct netfs_inode *ictx,
762 unsigned long long new_i_size,
763 bool changed_on_server)
764 {
765 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
766 struct inode *inode = &ictx->inode;
767
768 preempt_disable();
769 write_seqcount_begin(&inode->i_size_seqcount);
770 if (changed_on_server)
771 ictx->_remote_i_size = new_i_size;
772 if (new_i_size < ictx->_zero_point)
773 ictx->_zero_point = new_i_size;
774 write_seqcount_end(&inode->i_size_seqcount);
775 preempt_enable();
776 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
777 preempt_disable();
778 if (changed_on_server)
779 ictx->_remote_i_size = new_i_size;
780 if (new_i_size < ictx->_zero_point)
781 ictx->_zero_point = new_i_size;
782 preempt_enable();
783 #else
784 /*
785 * Pairs with smp_load_acquire() in netfs_read_remote_i_size and
786 * netfs_read_zero_point() to ensure changes related to inode size
787 * (such as page contents) are visible before we see the changed inode
788 * size.
789 */
790 if (changed_on_server)
791 smp_store_release(&ictx->_remote_i_size, new_i_size);
792 if (new_i_size < ictx->_zero_point)
793 smp_store_release(&ictx->_zero_point, new_i_size);
794 #endif
795 }
796
797 /**
798 * netfs_i_cookie - Get the cache cookie from the inode
799 * @ctx: The netfs inode to query
800 *
801 * Get the caching cookie (if enabled) from the network filesystem's inode.
802 */
netfs_i_cookie(struct netfs_inode * ctx)803 static inline struct fscache_cookie *netfs_i_cookie(struct netfs_inode *ctx)
804 {
805 #if IS_ENABLED(CONFIG_FSCACHE)
806 return ctx->cache;
807 #else
808 return NULL;
809 #endif
810 }
811
812 /**
813 * netfs_wait_for_outstanding_io - Wait for outstanding I/O to complete
814 * @inode: The netfs inode to wait on
815 *
816 * Wait for outstanding I/O requests of any type to complete. This is intended
817 * to be called from inode eviction routines. This makes sure that any
818 * resources held by those requests are cleaned up before we let the inode get
819 * cleaned up.
820 */
netfs_wait_for_outstanding_io(struct inode * inode)821 static inline void netfs_wait_for_outstanding_io(struct inode *inode)
822 {
823 struct netfs_inode *ictx = netfs_inode(inode);
824
825 wait_var_event(&ictx->io_count, atomic_read(&ictx->io_count) == 0);
826 }
827
828 #endif /* _LINUX_NETFS_H */
829