4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #ifndef LLITE_INTERNAL_H
34 #define LLITE_INTERNAL_H
35 #include "../include/lustre_debug.h"
36 #include "../include/lustre_ver.h"
37 #include "../include/lustre_disk.h" /* for s2sbi */
38 #include "../include/lustre_eacl.h"
40 /* for struct cl_lock_descr and struct cl_io */
41 #include "../include/cl_object.h"
42 #include "../include/lustre_lmv.h"
43 #include "../include/lustre_mdc.h"
44 #include "../include/lustre_intent.h"
45 #include <linux/compat.h>
46 #include <linux/xattr.h>
47 #include <linux/posix_acl_xattr.h>
48 #include "vvp_internal.h"
49 #include "range_lock.h"
55 #ifndef VM_FAULT_RETRY
56 #define VM_FAULT_RETRY 0
59 /** Only used on client-side for indicating the tail of dir hash/offset. */
60 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
61 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
63 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
64 #define LUSTRE_FPRIVATE(file) ((file)->private_data)
66 struct ll_dentry_data {
67 struct lookup_intent *lld_it;
68 unsigned int lld_sa_generation;
69 unsigned int lld_invalid:1;
70 unsigned int lld_nfs_dentry:1;
71 struct rcu_head lld_rcu_head;
74 #define ll_d2d(de) ((struct ll_dentry_data *)((de)->d_fsdata))
76 #define LLI_INODE_MAGIC 0x111d0de5
77 #define LLI_INODE_DEAD 0xdeadd00d
79 struct ll_getname_data {
80 struct dir_context ctx;
81 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
82 struct lu_fid lgd_fid; /* target fid we are looking for */
83 int lgd_found; /* inode matched? */
87 struct lu_env *lg_env;
89 struct cl_lock *lg_lock;
94 /* MDS has an authority for the Size-on-MDS attributes. */
95 LLIF_MDS_SIZE_LOCK = (1 << 0),
96 /* Epoch close is postponed. */
97 LLIF_EPOCH_PENDING = (1 << 1),
98 /* DONE WRITING is allowed. */
99 LLIF_DONE_WRITING = (1 << 2),
100 /* Sizeon-on-MDS attributes are changed. An attribute update needs to
103 LLIF_SOM_DIRTY = (1 << 3),
104 /* File data is modified. */
105 LLIF_DATA_MODIFIED = (1 << 4),
106 /* File is being restored */
107 LLIF_FILE_RESTORING = (1 << 5),
108 /* Xattr cache is attached to the file */
109 LLIF_XATTR_CACHE = (1 << 6),
112 struct ll_inode_info {
113 __u32 lli_inode_magic;
118 struct posix_acl *lli_posix_acl;
120 /* identifying fields for both metadata and data stacks. */
121 struct lu_fid lli_fid;
122 /* master inode fid for stripe directory */
123 struct lu_fid lli_pfid;
125 struct list_head lli_close_list;
127 /* handle is to be sent to MDS later on done_writing and setattr.
128 * Open handle data are needed for the recovery to reconstruct
129 * the inode state on the MDS. XXX: recovery is not ready yet.
131 struct obd_client_handle *lli_pending_och;
133 /* We need all three because every inode may be opened in different
136 struct obd_client_handle *lli_mds_read_och;
137 struct obd_client_handle *lli_mds_write_och;
138 struct obd_client_handle *lli_mds_exec_och;
139 __u64 lli_open_fd_read_count;
140 __u64 lli_open_fd_write_count;
141 __u64 lli_open_fd_exec_count;
142 /* Protects access to och pointers and their usage counters */
143 struct mutex lli_och_mutex;
145 struct inode lli_vfs_inode;
147 /* the most recent timestamps obtained from mds */
151 spinlock_t lli_agl_lock;
153 /* Try to make the d::member and f::member are aligned. Before using
154 * these members, make clear whether it is directory or not.
159 /* serialize normal readdir and statahead-readdir. */
160 struct mutex d_readdir_mutex;
162 /* metadata statahead */
163 /* since parent-child threads can share the same @file
164 * struct, "opendir_key" is the token when dir close for
165 * case of parent exit before child -- it is me should
166 * cleanup the dir readahead.
169 struct ll_statahead_info *d_sai;
170 /* protect statahead stuff. */
171 spinlock_t d_sa_lock;
172 /* "opendir_pid" is the token when lookup/revalidate
173 * -- I am the owner of dir statahead.
176 /* stat will try to access statahead entries or start
177 * statahead if this flag is set, and this flag will be
178 * set upon dir open, and cleared when dir is closed,
179 * statahead hit ratio is too low, or start statahead
182 unsigned int d_sa_enabled:1;
183 /* directory stripe information */
184 struct lmv_stripe_md *d_lsm_md;
185 /* striped directory size */
186 loff_t d_stripe_size;
187 /* striped directory nlink */
188 __u64 d_stripe_nlink;
191 #define lli_readdir_mutex u.d.d_readdir_mutex
192 #define lli_opendir_key u.d.d_opendir_key
193 #define lli_sai u.d.d_sai
194 #define lli_sa_lock u.d.d_sa_lock
195 #define lli_sa_enabled u.d.d_sa_enabled
196 #define lli_opendir_pid u.d.d_opendir_pid
197 #define lli_lsm_md u.d.d_lsm_md
198 #define lli_stripe_dir_size u.d.d_stripe_size
199 #define lli_stripe_dir_nlink u.d.d_stripe_nlink
201 /* for non-directory */
203 struct mutex f_size_mutex;
204 char *f_symlink_name;
207 * struct rw_semaphore {
208 * signed long count; // align d.d_def_acl
209 * spinlock_t wait_lock; // align d.d_sa_lock
210 * struct list_head wait_list;
213 struct rw_semaphore f_trunc_sem;
214 struct range_lock_tree f_write_tree;
216 struct rw_semaphore f_glimpse_sem;
217 unsigned long f_glimpse_time;
218 struct list_head f_agl_list;
221 /* for writepage() only to communicate to fsync */
225 * whenever a process try to read/write the file, the
226 * jobid of the process will be saved here, and it'll
227 * be packed into the write PRC when flush later.
229 * so the read/write statistics for jobid will not be
230 * accurate if the file is shared by different jobs.
232 char f_jobid[JOBSTATS_JOBID_SIZE];
235 #define lli_size_mutex u.f.f_size_mutex
236 #define lli_symlink_name u.f.f_symlink_name
237 #define lli_maxbytes u.f.f_maxbytes
238 #define lli_trunc_sem u.f.f_trunc_sem
239 #define lli_write_tree u.f.f_write_tree
240 #define lli_glimpse_sem u.f.f_glimpse_sem
241 #define lli_glimpse_time u.f.f_glimpse_time
242 #define lli_agl_list u.f.f_agl_list
243 #define lli_agl_index u.f.f_agl_index
244 #define lli_async_rc u.f.f_async_rc
245 #define lli_jobid u.f.f_jobid
249 /* XXX: For following frequent used members, although they maybe special
250 * used for non-directory object, it is some time-wasting to check
251 * whether the object is directory or not before using them. On the
252 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
253 * the "ll_inode_info" size even if moving those members into u.f.
254 * So keep them out side.
256 * In the future, if more members are added only for directory,
257 * some of the following members can be moved into u.f.
260 struct cl_object *lli_clob;
262 /* mutex to request for layout lock exclusively. */
263 struct mutex lli_layout_mutex;
264 /* Layout version, protected by lli_layout_lock */
265 __u32 lli_layout_gen;
266 spinlock_t lli_layout_lock;
268 struct rw_semaphore lli_xattrs_list_rwsem;
269 struct mutex lli_xattrs_enq_lock;
270 struct list_head lli_xattrs;/* ll_xattr_entry->xe_list */
273 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
277 spin_lock(&lli->lli_layout_lock);
278 gen = lli->lli_layout_gen;
279 spin_unlock(&lli->lli_layout_lock);
284 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
286 spin_lock(&lli->lli_layout_lock);
287 lli->lli_layout_gen = gen;
288 spin_unlock(&lli->lli_layout_lock);
291 int ll_xattr_cache_destroy(struct inode *inode);
293 int ll_xattr_cache_get(struct inode *inode, const char *name,
294 char *buffer, size_t size, __u64 valid);
297 * Locking to guarantee consistency of non-atomic updates to long long i_size,
298 * consistency between file size and KMS.
300 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
303 void ll_inode_size_lock(struct inode *inode);
304 void ll_inode_size_unlock(struct inode *inode);
306 /* FIXME: replace the name of this with LL_I to conform to kernel stuff */
307 /* static inline struct ll_inode_info *LL_I(struct inode *inode) */
308 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
310 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
313 /* default to about 40meg of readahead on a given system. That much tied
314 * up in 512k readahead requests serviced at 40ms each is about 1GB/s.
316 #define SBI_DEFAULT_READAHEAD_MAX (40UL << (20 - PAGE_SHIFT))
318 /* default to read-ahead full files smaller than 2MB on the second read */
319 #define SBI_DEFAULT_READAHEAD_WHOLE_MAX (2UL << (20 - PAGE_SHIFT))
324 RA_STAT_DISTANT_READPAGE,
325 RA_STAT_MISS_IN_WINDOW,
326 RA_STAT_FAILED_GRAB_PAGE,
327 RA_STAT_FAILED_MATCH,
332 RA_STAT_MAX_IN_FLIGHT,
333 RA_STAT_WRONG_GRAB_PAGE,
334 RA_STAT_FAILED_REACH_END,
339 atomic_t ra_cur_pages;
340 unsigned long ra_max_pages;
341 unsigned long ra_max_pages_per_file;
342 unsigned long ra_max_read_ahead_whole_pages;
345 /* ra_io_arg will be filled in the beginning of ll_readahead with
346 * ras_lock, then the following ll_read_ahead_pages will read RA
347 * pages according to this arg, all the items in this structure are
348 * counted by page index.
351 unsigned long ria_start; /* start offset of read-ahead*/
352 unsigned long ria_end; /* end offset of read-ahead*/
353 /* If stride read pattern is detected, ria_stoff means where
354 * stride read is started. Note: for normal read-ahead, the
355 * value here is meaningless, and also it will not be accessed
358 /* ria_length and ria_pages are the length and pages length in the
359 * stride I/O mode. And they will also be used to check whether
360 * it is stride I/O read-ahead in the read-ahead pages
362 unsigned long ria_length;
363 unsigned long ria_pages;
366 /* LL_HIST_MAX=32 causes an overflow */
367 #define LL_HIST_MAX 28
368 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
369 #define LL_PROCESS_HIST_MAX 10
370 struct per_process_info {
372 struct obd_histogram pp_r_hist;
373 struct obd_histogram pp_w_hist;
376 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
377 struct ll_rw_extents_info {
378 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
381 #define LL_OFFSET_HIST_MAX 100
382 struct ll_rw_process_info {
385 loff_t rw_range_start;
387 loff_t rw_last_file_pos;
389 size_t rw_smallest_extent;
390 size_t rw_largest_extent;
391 struct ll_file_data *rw_last_file;
394 enum stats_track_type {
395 STATS_TRACK_ALL = 0, /* track all processes */
396 STATS_TRACK_PID, /* track process with this pid */
397 STATS_TRACK_PPID, /* track processes with this ppid */
398 STATS_TRACK_GID, /* track processes with this gid */
402 /* flags for sbi->ll_flags */
403 #define LL_SBI_NOLCK 0x01 /* DLM locking disabled (directio-only) */
404 #define LL_SBI_CHECKSUM 0x02 /* checksum each page as it's written */
405 #define LL_SBI_FLOCK 0x04
406 #define LL_SBI_USER_XATTR 0x08 /* support user xattr */
407 #define LL_SBI_ACL 0x10 /* support ACL */
408 /* LL_SBI_RMT_CLIENT 0x40 remote client */
409 #define LL_SBI_MDS_CAPA 0x80 /* support mds capa, obsolete */
410 #define LL_SBI_OSS_CAPA 0x100 /* support oss capa, obsolete */
411 #define LL_SBI_LOCALFLOCK 0x200 /* Local flocks support by kernel */
412 #define LL_SBI_LRU_RESIZE 0x400 /* lru resize support */
413 #define LL_SBI_LAZYSTATFS 0x800 /* lazystatfs mount option */
414 #define LL_SBI_SOM_PREVIEW 0x1000 /* SOM preview mount option */
415 #define LL_SBI_32BIT_API 0x2000 /* generate 32 bit inodes. */
416 #define LL_SBI_64BIT_HASH 0x4000 /* support 64-bits dir hash/offset */
417 #define LL_SBI_AGL_ENABLED 0x8000 /* enable agl */
418 #define LL_SBI_VERBOSE 0x10000 /* verbose mount/umount */
419 #define LL_SBI_LAYOUT_LOCK 0x20000 /* layout lock support */
420 #define LL_SBI_USER_FID2PATH 0x40000 /* allow fid2path by unprivileged users */
421 #define LL_SBI_XATTR_CACHE 0x80000 /* support for xattr cache */
422 #define LL_SBI_NOROOTSQUASH 0x100000 /* do not apply root squash */
424 #define LL_SBI_FLAGS { \
449 /* this protects pglist and ra_info. It isn't safe to
450 * grab from interrupt contexts
453 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
454 spinlock_t ll_process_lock; /* ll_rw_process_info */
455 struct obd_uuid ll_sb_uuid;
456 struct obd_export *ll_md_exp;
457 struct obd_export *ll_dt_exp;
458 struct dentry *ll_debugfs_entry;
459 struct lu_fid ll_root_fid; /* root object fid */
462 unsigned int ll_umounting:1,
463 ll_xattr_cache_enabled:1;
464 struct lustre_client_ocd ll_lco;
466 struct ll_close_queue *ll_lcq;
468 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
471 * Used to track "unstable" pages on a client, and maintain a
472 * LRU list of clean pages. An "unstable" page is defined as
473 * any page which is sent to a server as part of a bulk request,
474 * but is uncommitted to stable storage.
476 struct cl_client_cache *ll_cache;
478 struct lprocfs_stats *ll_ra_stats;
480 struct ll_ra_info ll_ra_info;
481 unsigned int ll_namelen;
482 struct file_operations *ll_fop;
484 unsigned int ll_md_brw_pages; /* readdir pages per RPC */
486 struct lu_site *ll_site;
487 struct cl_device *ll_cl;
489 struct ll_rw_extents_info ll_rw_extents_info;
490 int ll_extent_process_count;
491 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
492 unsigned int ll_offset_process_count;
493 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
494 unsigned int ll_rw_offset_entry_count;
495 int ll_stats_track_id;
496 enum stats_track_type ll_stats_track_type;
499 /* metadata stat-ahead */
500 unsigned int ll_sa_max; /* max statahead RPCs */
501 atomic_t ll_sa_total; /* statahead thread started
504 atomic_t ll_sa_wrong; /* statahead thread stopped for
507 atomic_t ll_sa_running; /* running statahead thread
510 atomic_t ll_agl_total; /* AGL thread started count */
512 dev_t ll_sdev_orig; /* save s_dev before assign for
516 struct root_squash_info ll_squash;
518 __kernel_fsid_t ll_fsid;
519 struct kobject ll_kobj; /* sysfs object */
520 struct super_block *ll_sb; /* struct super_block (for sysfs code)*/
521 struct completion ll_kobj_unregister;
525 * per file-descriptor read-ahead data.
527 struct ll_readahead_state {
530 * index of the last page that read(2) needed and that wasn't in the
531 * cache. Used by ras_update() to detect seeks.
533 * XXX nikita: if access seeks into cached region, Lustre doesn't see
536 unsigned long ras_last_readpage;
538 * number of pages read after last read-ahead window reset. As window
539 * is reset on each seek, this is effectively a number of consecutive
540 * accesses. Maybe ->ras_accessed_in_window is better name.
542 * XXX nikita: window is also reset (by ras_update()) when Lustre
543 * believes that memory pressure evicts read-ahead pages. In that
544 * case, it probably doesn't make sense to expand window to
545 * PTLRPC_MAX_BRW_PAGES on the third access.
547 unsigned long ras_consecutive_pages;
549 * number of read requests after the last read-ahead window reset
550 * As window is reset on each seek, this is effectively the number
551 * on consecutive read request and is used to trigger read-ahead.
553 unsigned long ras_consecutive_requests;
555 * Parameters of current read-ahead window. Handled by
556 * ras_update(). On the initial access to the file or after a seek,
557 * window is reset to 0. After 3 consecutive accesses, window is
558 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
559 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
561 unsigned long ras_window_start, ras_window_len;
563 * Where next read-ahead should start at. This lies within read-ahead
564 * window. Read-ahead window is read in pieces rather than at once
565 * because: 1. lustre limits total number of pages under read-ahead by
566 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
567 * not covered by DLM lock.
569 unsigned long ras_next_readahead;
571 * Total number of ll_file_read requests issued, reads originating
572 * due to mmap are not counted in this total. This value is used to
573 * trigger full file read-ahead after multiple reads to a small file.
575 unsigned long ras_requests;
577 * Page index with respect to the current request, these value
578 * will not be accurate when dealing with reads issued via mmap.
580 unsigned long ras_request_index;
582 * The following 3 items are used for detecting the stride I/O
584 * In stride I/O mode,
585 * ...............|-----data-----|****gap*****|--------|******|....
586 * offset |-stride_pages-|-stride_gap-|
587 * ras_stride_offset = offset;
588 * ras_stride_length = stride_pages + stride_gap;
589 * ras_stride_pages = stride_pages;
590 * Note: all these three items are counted by pages.
592 unsigned long ras_stride_length;
593 unsigned long ras_stride_pages;
594 pgoff_t ras_stride_offset;
596 * number of consecutive stride request count, and it is similar as
597 * ras_consecutive_requests, but used for stride I/O mode.
598 * Note: only more than 2 consecutive stride request are detected,
599 * stride read-ahead will be enable
601 unsigned long ras_consecutive_stride_requests;
604 extern struct kmem_cache *ll_file_data_slab;
605 struct lustre_handle;
606 struct ll_file_data {
607 struct ll_readahead_state fd_ras;
608 struct ll_grouplock fd_grouplock;
612 /* openhandle if lease exists for this file.
613 * Borrow lli->lli_och_mutex to protect assignment
615 struct obd_client_handle *fd_lease_och;
616 struct obd_client_handle *fd_och;
617 struct file *fd_file;
618 /* Indicate whether need to report failure when close.
619 * true: failure is known, not report again.
620 * false: unknown failure, should report.
622 bool fd_write_failed;
623 rwlock_t fd_lock; /* protect lcc list */
624 struct list_head fd_lccs; /* list of ll_cl_context */
627 struct lov_stripe_md;
629 extern struct dentry *llite_root;
630 extern struct kset *llite_kset;
632 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
634 return &lli->lli_vfs_inode;
637 __u32 ll_i2suppgid(struct inode *i);
638 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2);
640 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
642 #if BITS_PER_LONG == 32
644 #elif defined(CONFIG_COMPAT)
645 return unlikely(in_compat_syscall() || (sbi->ll_flags & LL_SBI_32BIT_API));
647 return unlikely(sbi->ll_flags & LL_SBI_32BIT_API);
651 void ll_ras_enter(struct file *f);
653 /* llite/lcommon_misc.c */
654 int cl_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp);
655 int cl_ocd_update(struct obd_device *host,
656 struct obd_device *watched,
657 enum obd_notify_event ev, void *owner, void *data);
658 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
659 struct ll_grouplock *cg);
660 void cl_put_grouplock(struct ll_grouplock *cg);
662 /* llite/lproc_llite.c */
663 int ldebugfs_register_mountpoint(struct dentry *parent,
664 struct super_block *sb, char *osc, char *mdc);
665 void ldebugfs_unregister_mountpoint(struct ll_sb_info *sbi);
666 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count);
667 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars);
668 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
669 struct ll_file_data *file, loff_t pos,
670 size_t count, int rw);
673 extern const struct file_operations ll_dir_operations;
674 extern const struct inode_operations ll_dir_inode_operations;
675 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
676 struct dir_context *ctx);
677 int ll_get_mdt_idx(struct inode *inode);
678 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
679 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
680 __u64 offset, struct ll_dir_chain *chain);
681 void ll_release_page(struct inode *inode, struct page *page, bool remove);
684 extern const struct inode_operations ll_special_inode_operations;
686 int ll_objects_destroy(struct ptlrpc_request *request,
688 struct inode *ll_iget(struct super_block *sb, ino_t hash,
689 struct lustre_md *lic);
690 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
691 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
692 void *data, int flag);
693 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
694 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
697 int ll_writepage(struct page *page, struct writeback_control *wbc);
698 int ll_writepages(struct address_space *, struct writeback_control *wbc);
699 int ll_readpage(struct file *file, struct page *page);
700 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
701 int ll_readahead(const struct lu_env *env, struct cl_io *io,
702 struct cl_page_list *queue, struct ll_readahead_state *ras,
704 struct ll_cl_context *ll_cl_find(struct file *file);
705 void ll_cl_add(struct file *file, const struct lu_env *env, struct cl_io *io);
706 void ll_cl_remove(struct file *file, const struct lu_env *env);
708 extern const struct address_space_operations ll_aops;
711 extern struct file_operations ll_file_operations;
712 extern struct file_operations ll_file_operations_flock;
713 extern struct file_operations ll_file_operations_noflock;
714 extern const struct inode_operations ll_file_inode_operations;
715 int ll_have_md_lock(struct inode *inode, __u64 *bits,
716 enum ldlm_mode l_req_mode);
717 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
718 struct lustre_handle *lockh, __u64 flags,
719 enum ldlm_mode mode);
720 int ll_file_open(struct inode *inode, struct file *file);
721 int ll_file_release(struct inode *inode, struct file *file);
722 int ll_glimpse_ioctl(struct ll_sb_info *sbi,
723 struct lov_stripe_md *lsm, lstat_t *st);
724 void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch);
725 int ll_release_openhandle(struct inode *, struct lookup_intent *);
726 int ll_md_real_close(struct inode *inode, fmode_t fmode);
727 void ll_ioepoch_close(struct inode *inode, struct md_op_data *op_data,
728 struct obd_client_handle **och, unsigned long flags);
729 void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data);
730 int ll_som_update(struct inode *inode, struct md_op_data *op_data);
731 int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
732 __u64 ioepoch, int sync);
733 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
734 struct lustre_handle *fh);
735 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
736 struct posix_acl *ll_get_acl(struct inode *inode, int type);
737 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
738 const char *name, int namelen);
739 int ll_get_fid_by_name(struct inode *parent, const char *name,
740 int namelen, struct lu_fid *fid);
741 int ll_inode_permission(struct inode *inode, int mask);
743 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
744 __u64 flags, struct lov_user_md *lum,
746 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
747 struct lov_mds_md **lmm, int *lmm_size,
748 struct ptlrpc_request **request);
749 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
751 int ll_dir_getstripe(struct inode *inode, void **lmmp, int *lmm_size,
752 struct ptlrpc_request **request, u64 valid);
753 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
754 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
755 int ll_fid2path(struct inode *inode, void __user *arg);
756 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
757 int ll_hsm_release(struct inode *inode);
761 int ll_d_init(struct dentry *de);
762 extern const struct dentry_operations ll_d_ops;
763 void ll_intent_drop_lock(struct lookup_intent *);
764 void ll_intent_release(struct lookup_intent *);
765 void ll_invalidate_aliases(struct inode *);
766 void ll_lookup_finish_locks(struct lookup_intent *it, struct inode *inode);
767 int ll_revalidate_it_finish(struct ptlrpc_request *request,
768 struct lookup_intent *it, struct inode *inode);
770 /* llite/llite_lib.c */
771 extern struct super_operations lustre_super_operations;
773 void ll_lli_init(struct ll_inode_info *lli);
774 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt);
775 void ll_put_super(struct super_block *sb);
776 void ll_kill_super(struct super_block *sb);
777 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
778 void ll_clear_inode(struct inode *inode);
779 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import);
780 int ll_setattr(struct dentry *de, struct iattr *attr);
781 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
782 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
783 __u64 max_age, __u32 flags);
784 int ll_update_inode(struct inode *inode, struct lustre_md *md);
785 int ll_read_inode2(struct inode *inode, void *opaque);
786 void ll_delete_inode(struct inode *inode);
787 int ll_iocontrol(struct inode *inode, struct file *file,
788 unsigned int cmd, unsigned long arg);
789 int ll_flush_ctx(struct inode *inode);
790 void ll_umount_begin(struct super_block *sb);
791 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
792 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
793 void ll_dirty_page_discard_warn(struct page *page, int ioret);
794 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
795 struct super_block *, struct lookup_intent *);
796 int ll_obd_statfs(struct inode *inode, void __user *arg);
797 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
798 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
799 int ll_process_config(struct lustre_cfg *lcfg);
802 LUSTRE_OPC_MKDIR = 0,
803 LUSTRE_OPC_SYMLINK = 1,
804 LUSTRE_OPC_MKNOD = 2,
805 LUSTRE_OPC_CREATE = 3,
809 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
810 struct inode *i1, struct inode *i2,
811 const char *name, int namelen,
812 int mode, __u32 opc, void *data);
813 void ll_finish_md_op_data(struct md_op_data *op_data);
814 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
815 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen);
816 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
817 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req);
819 /* llite/llite_nfs.c */
820 extern const struct export_operations lustre_export_operations;
821 __u32 get_uuid2int(const char *name, int len);
822 void get_uuid2fsid(const char *name, int len, __kernel_fsid_t *fsid);
823 struct inode *search_inode_for_lustre(struct super_block *sb,
824 const struct lu_fid *fid);
825 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
827 /* llite/symlink.c */
828 extern const struct inode_operations ll_fast_symlink_inode_operations;
830 /* llite/llite_close.c */
831 struct ll_close_queue {
833 struct list_head lcq_head;
834 wait_queue_head_t lcq_waitq;
835 struct completion lcq_comp;
839 void vvp_write_pending(struct vvp_object *club, struct vvp_page *page);
840 void vvp_write_complete(struct vvp_object *club, struct vvp_page *page);
843 * IO arguments for various VFS I/O interfaces.
847 enum vvp_io_subtype via_io_subtype;
851 struct kiocb *via_iocb;
852 struct iov_iter *via_iter;
855 struct pipe_inode_info *via_pipe;
856 unsigned int via_flags;
861 struct ll_cl_context {
862 struct list_head lcc_list;
864 const struct lu_env *lcc_env;
865 struct cl_io *lcc_io;
866 struct cl_page *lcc_page;
869 struct ll_thread_info {
870 struct vvp_io_args lti_args;
871 struct ra_io_arg lti_ria;
872 struct ll_cl_context lti_io_ctx;
875 extern struct lu_context_key ll_thread_key;
876 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
878 struct ll_thread_info *lti;
880 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
885 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env,
886 enum vvp_io_subtype type)
888 struct vvp_io_args *via = &ll_env_info(env)->lti_args;
890 via->via_io_subtype = type;
895 void ll_queue_done_writing(struct inode *inode, unsigned long flags);
896 void ll_close_thread_shutdown(struct ll_close_queue *lcq);
897 int ll_close_thread_start(struct ll_close_queue **lcq_ret);
899 /* llite/llite_mmap.c */
901 int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last);
902 int ll_file_mmap(struct file *file, struct vm_area_struct *vma);
903 void policy_from_vma(ldlm_policy_data_t *policy, struct vm_area_struct *vma,
904 unsigned long addr, size_t count);
905 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
908 static inline void ll_invalidate_page(struct page *vmpage)
910 struct address_space *mapping = vmpage->mapping;
911 loff_t offset = vmpage->index << PAGE_SHIFT;
913 LASSERT(PageLocked(vmpage));
918 * truncate_complete_page() calls
919 * a_ops->invalidatepage()->cl_page_delete()->vvp_page_delete().
921 ll_teardown_mmaps(mapping, offset, offset + PAGE_SIZE);
922 truncate_complete_page(mapping, vmpage);
925 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
927 /* don't need an addref as the sb_info should be holding one */
928 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
930 return ll_s2sbi(sb)->ll_dt_exp;
933 /* don't need an addref as the sb_info should be holding one */
934 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
936 return ll_s2sbi(sb)->ll_md_exp;
939 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
941 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
948 /* FIXME: replace the name of this with LL_SB to conform to kernel stuff */
949 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
951 return ll_s2sbi(inode->i_sb);
954 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
956 return ll_s2dtexp(inode->i_sb);
959 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
961 return ll_s2mdexp(inode->i_sb);
964 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
969 fid = &ll_i2info(inode)->lli_fid;
974 static inline __u64 ll_file_maxbytes(struct inode *inode)
976 return ll_i2info(inode)->lli_maxbytes;
980 extern const struct xattr_handler *ll_xattr_handlers[];
982 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
985 * Common IO arguments for various VFS I/O interfaces.
987 int cl_sb_init(struct super_block *sb);
988 int cl_sb_fini(struct super_block *sb);
989 void ll_io_init(struct cl_io *io, const struct file *file, int write);
991 void ras_update(struct ll_sb_info *sbi, struct inode *inode,
992 struct ll_readahead_state *ras, unsigned long index,
994 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
995 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
998 #define LL_SA_RPC_MIN 2
999 #define LL_SA_RPC_DEF 32
1000 #define LL_SA_RPC_MAX 8192
1002 #define LL_SA_CACHE_BIT 5
1003 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1004 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1006 /* per inode struct, for dir only */
1007 struct ll_statahead_info {
1008 struct inode *sai_inode;
1009 atomic_t sai_refcount; /* when access this struct, hold
1012 unsigned int sai_generation; /* generation for statahead */
1013 unsigned int sai_max; /* max ahead of lookup */
1014 __u64 sai_sent; /* stat requests sent count */
1015 __u64 sai_replied; /* stat requests which received
1018 __u64 sai_index; /* index of statahead entry */
1019 __u64 sai_index_wait; /* index of entry which is the
1020 * caller is waiting for
1022 __u64 sai_hit; /* hit count */
1023 __u64 sai_miss; /* miss count:
1024 * for "ls -al" case, it includes
1025 * hidden dentry miss;
1026 * for "ls -l" case, it does not
1027 * include hidden dentry miss.
1028 * "sai_miss_hidden" is used for
1031 unsigned int sai_consecutive_miss; /* consecutive miss */
1032 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1033 * is not a hidden one
1035 unsigned int sai_skip_hidden;/* skipped hidden dentry count */
1036 unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
1039 sai_agl_valid:1,/* AGL is valid for the dir */
1040 sai_in_readpage:1;/* statahead is in readdir() */
1041 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1042 struct ptlrpc_thread sai_thread; /* stat-ahead thread */
1043 struct ptlrpc_thread sai_agl_thread; /* AGL thread */
1044 struct list_head sai_entries; /* entry list */
1045 struct list_head sai_entries_received; /* entries returned */
1046 struct list_head sai_entries_stated; /* entries stated */
1047 struct list_head sai_entries_agl; /* AGL entries to be sent */
1048 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1049 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1050 atomic_t sai_cache_count; /* entry count in cache */
1053 int do_statahead_enter(struct inode *dir, struct dentry **dentry,
1055 void ll_authorize_statahead(struct inode *dir, void *key);
1056 void ll_deauthorize_statahead(struct inode *dir, void *key);
1058 blkcnt_t dirty_cnt(struct inode *inode);
1060 int cl_glimpse_size0(struct inode *inode, int agl);
1061 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1062 struct inode *inode, struct cl_object *clob, int agl);
1064 static inline int cl_glimpse_size(struct inode *inode)
1066 return cl_glimpse_size0(inode, 0);
1069 static inline int cl_agl(struct inode *inode)
1071 return cl_glimpse_size0(inode, 1);
1074 static inline int ll_glimpse_size(struct inode *inode)
1076 struct ll_inode_info *lli = ll_i2info(inode);
1079 down_read(&lli->lli_glimpse_sem);
1080 rc = cl_glimpse_size(inode);
1081 lli->lli_glimpse_time = cfs_time_current();
1082 up_read(&lli->lli_glimpse_sem);
1087 ll_statahead_mark(struct inode *dir, struct dentry *dentry)
1089 struct ll_inode_info *lli = ll_i2info(dir);
1090 struct ll_statahead_info *sai = lli->lli_sai;
1091 struct ll_dentry_data *ldd = ll_d2d(dentry);
1093 /* not the same process, don't mark */
1094 if (lli->lli_opendir_pid != current_pid())
1099 ldd->lld_sa_generation = sai->sai_generation;
1103 dentry_need_statahead(struct inode *dir, struct dentry *dentry)
1105 struct ll_inode_info *lli;
1106 struct ll_dentry_data *ldd;
1108 if (ll_i2sbi(dir)->ll_sa_max == 0)
1111 lli = ll_i2info(dir);
1114 * statahead is not allowed for this dir, there may be three causes:
1115 * 1. dir is not opened.
1116 * 2. statahead hit ratio is too low.
1117 * 3. previous stat started statahead thread failed.
1119 if (!lli->lli_sa_enabled)
1122 /* not the same process, don't statahead */
1123 if (lli->lli_opendir_pid != current_pid())
1126 ldd = ll_d2d(dentry);
1128 * When stats a dentry, the system trigger more than once "revalidate"
1129 * or "lookup", for "getattr", for "getxattr", and maybe for others.
1130 * Under patchless client mode, the operation intent is not accurate,
1131 * which maybe misguide the statahead thread. For example:
1132 * The "revalidate" call for "getattr" and "getxattr" of a dentry maybe
1133 * have the same operation intent -- "IT_GETATTR".
1134 * In fact, one dentry should has only one chance to interact with the
1135 * statahead thread, otherwise the statahead windows will be confused.
1136 * The solution is as following:
1137 * Assign "lld_sa_generation" with "sai_generation" when a dentry
1138 * "IT_GETATTR" for the first time, and the subsequent "IT_GETATTR"
1139 * will bypass interacting with statahead thread for checking:
1140 * "lld_sa_generation == lli_sai->sai_generation"
1142 if (ldd && lli->lli_sai &&
1143 ldd->lld_sa_generation == lli->lli_sai->sai_generation)
1150 ll_statahead_enter(struct inode *dir, struct dentry **dentryp, int only_unplug)
1152 if (!dentry_need_statahead(dir, *dentryp))
1155 return do_statahead_enter(dir, dentryp, only_unplug);
1158 /* llite ioctl register support routine */
1164 #define LLIOC_MAX_CMD 256
1167 * Rules to write a callback function:
1170 * @magic: Dynamic ioctl call routine will feed this value with the pointer
1171 * returned to ll_iocontrol_register. Callback functions should use this
1172 * data to check the potential collasion of ioctl cmd. If collasion is
1173 * found, callback function should return LLIOC_CONT.
1174 * @rcp: The result of ioctl command.
1177 * If @magic matches the pointer returned by ll_iocontrol_data, the
1178 * callback should return LLIOC_STOP; return LLIOC_STOP otherwise.
1180 typedef enum llioc_iter (*llioc_callback_t)(struct inode *inode,
1181 struct file *file, unsigned int cmd, unsigned long arg,
1182 void *magic, int *rcp);
1184 /* export functions */
1185 /* Register ioctl block dynamatically for a regular file.
1187 * @cmd: the array of ioctl command set
1188 * @count: number of commands in the @cmd
1189 * @cb: callback function, it will be called if an ioctl command is found to
1190 * belong to the command list @cmd.
1193 * A magic pointer will be returned if success;
1194 * otherwise, NULL will be returned.
1196 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd);
1197 void ll_iocontrol_unregister(void *magic);
1199 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1200 enum cl_fsync_mode mode, int ignore_layout);
1202 /** direct write pages */
1203 struct ll_dio_pages {
1204 /** page array to be written. we don't support
1205 * partial pages except the last one.
1207 struct page **ldp_pages;
1208 /* offset of each page */
1209 loff_t *ldp_offsets;
1210 /** if ldp_offsets is NULL, it means a sequential
1211 * pages to be written, then this is the file offset
1212 * of the first page.
1214 loff_t ldp_start_offset;
1215 /** how many bytes are to be written. */
1217 /** # of pages in the array. */
1221 static inline void cl_stats_tally(struct cl_device *dev, enum cl_req_type crt,
1224 int opc = (crt == CRT_READ) ? LPROC_LL_OSC_READ :
1227 ll_stats_ops_tally(ll_s2sbi(cl2vvp_dev(dev)->vdv_sb), opc, rc);
1230 ssize_t ll_direct_rw_pages(const struct lu_env *env, struct cl_io *io,
1231 int rw, struct inode *inode,
1232 struct ll_dio_pages *pv);
1234 static inline int ll_file_nolock(const struct file *file)
1236 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1237 struct inode *inode = file_inode(file);
1239 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1240 (ll_i2sbi(inode)->ll_flags & LL_SBI_NOLCK));
1243 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1244 struct lookup_intent *it, __u64 *bits)
1246 if (!it->it_lock_set) {
1247 struct lustre_handle handle;
1249 /* If this inode is a remote object, it will get two
1250 * separate locks in different namespaces, Master MDT,
1251 * where the name entry is, will grant LOOKUP lock,
1252 * remote MDT, where the object is, will grant
1253 * UPDATE|PERM lock. The inode will be attached to both
1254 * LOOKUP and PERM locks, so revoking either locks will
1255 * case the dcache being cleared
1257 if (it->it_remote_lock_mode) {
1258 handle.cookie = it->it_remote_lock_handle;
1259 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"%p for remote lock %#llx\n",
1260 PFID(ll_inode2fid(inode)), inode,
1262 md_set_lock_data(exp, &handle, inode, NULL);
1265 handle.cookie = it->it_lock_handle;
1267 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"%p for lock %#llx\n",
1268 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1270 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1271 it->it_lock_set = 1;
1275 *bits = it->it_lock_bits;
1278 static inline int d_lustre_invalid(const struct dentry *dentry)
1280 struct ll_dentry_data *lld = ll_d2d(dentry);
1282 return !lld || lld->lld_invalid;
1285 static inline void __d_lustre_invalidate(struct dentry *dentry)
1287 struct ll_dentry_data *lld = ll_d2d(dentry);
1290 lld->lld_invalid = 1;
1294 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1295 * ll_md_blocking_ast), unhash this dentry, and let dcache to reclaim it later;
1296 * else dput() of the last refcount will unhash this dentry and kill it.
1298 static inline void d_lustre_invalidate(struct dentry *dentry, int nested)
1300 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1302 dentry->d_parent, d_inode(dentry), d_count(dentry));
1304 spin_lock_nested(&dentry->d_lock,
1305 nested ? DENTRY_D_LOCK_NESTED : DENTRY_D_LOCK_NORMAL);
1306 __d_lustre_invalidate(dentry);
1308 * We should be careful about dentries created by d_obtain_alias().
1309 * These dentries are not put in the dentry tree, instead they are
1310 * linked to sb->s_anon through dentry->d_hash.
1311 * shrink_dcache_for_umount() shrinks the tree and sb->s_anon list.
1312 * If we unhashed such a dentry, unmount would not be able to find
1313 * it and busy inodes would be reported.
1315 if (d_count(dentry) == 0 && !(dentry->d_flags & DCACHE_DISCONNECTED))
1317 spin_unlock(&dentry->d_lock);
1320 static inline void d_lustre_revalidate(struct dentry *dentry)
1322 spin_lock(&dentry->d_lock);
1323 LASSERT(ll_d2d(dentry));
1324 ll_d2d(dentry)->lld_invalid = 0;
1325 spin_unlock(&dentry->d_lock);
1329 LL_LAYOUT_GEN_NONE = ((__u32)-2), /* layout lock was cancelled */
1330 LL_LAYOUT_GEN_EMPTY = ((__u32)-1) /* for empty layout */
1333 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1334 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1335 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1337 int ll_xattr_init(void);
1338 void ll_xattr_fini(void);
1340 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1341 struct cl_page *page, enum cl_req_type crt);
1344 int cl_setattr_ost(struct inode *inode, const struct iattr *attr);
1346 extern struct lu_env *cl_inode_fini_env;
1347 extern int cl_inode_fini_refcheck;
1349 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1350 void cl_inode_fini(struct inode *inode);
1351 int cl_local_size(struct inode *inode);
1353 __u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1354 __u32 cl_fid_build_gen(const struct lu_fid *fid);
1356 #endif /* LLITE_INTERNAL_H */