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, 2017, 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 <lustre_debug.h>
36 #include <lustre_disk.h> /* for s2sbi */
37 #include <lustre_eacl.h>
38 #include <lustre_linkea.h>
40 /* for struct cl_lock_descr and struct cl_io */
41 #include <cl_object.h>
42 #include <lustre_lmv.h>
43 #include <lustre_mdc.h>
44 #include <lustre_intent.h>
45 #include <linux/compat.h>
46 #include <linux/aio.h>
47 #include <lustre_compat.h>
48 #include <lustre_crypto.h>
50 #include "vvp_internal.h"
51 #include "range_lock.h"
58 #ifndef VM_FAULT_RETRY
59 #define VM_FAULT_RETRY 0
62 /* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
63 * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
64 #ifndef LOOKUP_CONTINUE
65 #define LOOKUP_CONTINUE LOOKUP_PARENT
68 /** Only used on client-side for indicating the tail of dir hash/offset. */
69 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
70 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
72 /* 4UL * 1024 * 1024 */
73 #define LL_MAX_BLKSIZE_BITS 22
75 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
77 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
79 struct ll_dentry_data {
80 unsigned int lld_sa_generation;
81 unsigned int lld_invalid:1;
82 unsigned int lld_nfs_dentry:1;
83 struct rcu_head lld_rcu_head;
86 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
88 #define LLI_INODE_MAGIC 0x111d0de5
89 #define LLI_INODE_DEAD 0xdeadd00d
91 struct ll_getname_data {
92 #ifdef HAVE_DIR_CONTEXT
93 struct dir_context ctx;
95 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
96 struct lu_fid lgd_fid; /* target fid we are looking for */
97 int lgd_found; /* inode matched? */
100 struct ll_grouplock {
101 struct lu_env *lg_env;
103 struct cl_lock *lg_lock;
104 unsigned long lg_gid;
107 /* See comment on trunc_sem_down_read_nowait */
108 struct ll_trunc_sem {
109 /* when positive, this is a count of readers, when -1, it indicates
110 * the semaphore is held for write, and 0 is unlocked
112 atomic_t ll_trunc_readers;
113 /* this tracks a count of waiting writers */
114 atomic_t ll_trunc_waiters;
117 struct ll_inode_info {
118 __u32 lli_inode_magic;
121 volatile unsigned long lli_flags;
122 struct posix_acl *lli_posix_acl;
124 /* identifying fields for both metadata and data stacks. */
125 struct lu_fid lli_fid;
126 /* master inode fid for stripe directory */
127 struct lu_fid lli_pfid;
129 /* We need all three because every inode may be opened in different
131 struct obd_client_handle *lli_mds_read_och;
132 struct obd_client_handle *lli_mds_write_och;
133 struct obd_client_handle *lli_mds_exec_och;
134 __u64 lli_open_fd_read_count;
135 __u64 lli_open_fd_write_count;
136 __u64 lli_open_fd_exec_count;
137 /* Protects access to och pointers and their usage counters */
138 struct mutex lli_och_mutex;
140 struct inode lli_vfs_inode;
142 /* the most recent timestamps obtained from mds */
147 spinlock_t lli_agl_lock;
149 /* Try to make the d::member and f::member are aligned. Before using
150 * these members, make clear whether it is directory or not. */
154 /* metadata statahead */
155 /* since parent-child threads can share the same @file
156 * struct, "opendir_key" is the token when dir close for
157 * case of parent exit before child -- it is me should
158 * cleanup the dir readahead. */
159 void *lli_opendir_key;
160 struct ll_statahead_info *lli_sai;
161 /* protect statahead stuff. */
162 spinlock_t lli_sa_lock;
163 /* "opendir_pid" is the token when lookup/revalid
164 * -- I am the owner of dir statahead. */
165 pid_t lli_opendir_pid;
166 /* stat will try to access statahead entries or start
167 * statahead if this flag is set, and this flag will be
168 * set upon dir open, and cleared when dir is closed,
169 * statahead hit ratio is too low, or start statahead
171 unsigned int lli_sa_enabled:1;
172 /* generation for statahead */
173 unsigned int lli_sa_generation;
174 /* rw lock protects lli_lsm_md */
175 struct rw_semaphore lli_lsm_sem;
176 /* directory stripe information */
177 struct lmv_stripe_md *lli_lsm_md;
178 /* directory default LMV */
179 struct lmv_stripe_md *lli_default_lsm_md;
182 /* for non-directory */
184 struct mutex lli_size_mutex;
185 char *lli_symlink_name;
186 struct ll_trunc_sem lli_trunc_sem;
187 struct range_lock_tree lli_write_tree;
189 struct rw_semaphore lli_glimpse_sem;
190 ktime_t lli_glimpse_time;
191 struct list_head lli_agl_list;
194 /* for writepage() only to communicate to fsync */
197 /* protect the file heat fields */
198 spinlock_t lli_heat_lock;
199 __u32 lli_heat_flags;
200 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
203 * Whenever a process try to read/write the file, the
204 * jobid of the process will be saved here, and it'll
205 * be packed into the write PRC when flush later.
207 * So the read/write statistics for jobid will not be
208 * accurate if the file is shared by different jobs.
210 char lli_jobid[LUSTRE_JOBID_SIZE];
212 struct mutex lli_pcc_lock;
213 enum lu_pcc_state_flags lli_pcc_state;
215 * @lli_pcc_generation saves the gobal PCC generation
216 * when the file was successfully attached into PCC.
217 * The flags of the PCC dataset are saved in
219 * The gobal PCC generation will be increased when add
220 * or delete a PCC backend, or change the configuration
221 * parameters for PCC.
222 * If @lli_pcc_generation is same as the gobal PCC
223 * generation, we can use the saved flags of the PCC
224 * dataset to determine whether need to try auto attach
227 __u64 lli_pcc_generation;
228 enum pcc_dataset_flags lli_pcc_dsflags;
229 struct pcc_inode *lli_pcc_inode;
231 struct mutex lli_group_mutex;
232 __u64 lli_group_users;
233 unsigned long lli_group_gid;
235 __u64 lli_attr_valid;
237 __u64 lli_lazyblocks;
241 /* XXX: For following frequent used members, although they maybe special
242 * used for non-directory object, it is some time-wasting to check
243 * whether the object is directory or not before using them. On the
244 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
245 * the "ll_inode_info" size even if moving those members into u.f.
246 * So keep them out side.
248 * In the future, if more members are added only for directory,
249 * some of the following members can be moved into u.f.
251 struct cl_object *lli_clob;
253 /* mutex to request for layout lock exclusively. */
254 struct mutex lli_layout_mutex;
255 /* Layout version, protected by lli_layout_lock */
256 __u32 lli_layout_gen;
257 spinlock_t lli_layout_lock;
259 __u32 lli_projid; /* project id */
261 struct rw_semaphore lli_xattrs_list_rwsem;
262 struct mutex lli_xattrs_enq_lock;
263 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
266 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
268 atomic_set(&sem->ll_trunc_readers, 0);
269 atomic_set(&sem->ll_trunc_waiters, 0);
272 /* This version of down read ignores waiting writers, meaning if the semaphore
273 * is already held for read, this down_read will 'join' that reader and also
274 * take the semaphore.
276 * This lets us avoid an unusual deadlock.
278 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
279 * in Lustre, in order to exclude truncates. Some of these paths then need to
280 * take the mmap_sem, while still holding the trunc_sem. The problem is that
281 * page faults hold the mmap_sem when calling in to Lustre, and then must also
282 * take the trunc_sem to exclude truncate.
284 * This means the locking order for trunc_sem and mmap_sem is sometimes AB,
285 * sometimes BA. This is almost OK because in both cases, we take the trunc
286 * sem for read, so it doesn't block.
288 * However, if a write mode user (truncate, a setattr op) arrives in the
289 * middle of this, the second reader on the truncate_sem will wait behind that
292 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
293 * to the mode in which they take the semaphore):
294 * reader (holding mmap_sem, needs truncate_sem)
296 * reader (holding truncate sem, waiting for mmap_sem)
298 * And so the readers deadlock.
300 * The solution is this modified semaphore, where this down_read ignores
301 * waiting write operations, and all waiters are woken up at once, so readers
302 * using down_read_nowait cannot get stuck behind waiting writers, regardless
303 * of the order they arrived in.
305 * down_read_nowait is only used in the page fault case, where we already hold
306 * the mmap_sem. This is because otherwise repeated read and write operations
307 * (which take the truncate sem) could prevent a truncate from ever starting.
308 * This could still happen with page faults, but without an even more complex
309 * mechanism, this is unavoidable.
313 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
315 wait_var_event(&sem->ll_trunc_readers,
316 atomic_inc_unless_negative(&sem->ll_trunc_readers));
319 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
321 wait_var_event(&sem->ll_trunc_readers,
322 atomic_read(&sem->ll_trunc_waiters) == 0 &&
323 atomic_inc_unless_negative(&sem->ll_trunc_readers));
326 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
328 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
329 atomic_read(&sem->ll_trunc_waiters))
330 wake_up_var(&sem->ll_trunc_readers);
333 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
335 atomic_inc(&sem->ll_trunc_waiters);
336 wait_var_event(&sem->ll_trunc_readers,
337 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
338 atomic_dec(&sem->ll_trunc_waiters);
341 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
343 atomic_set(&sem->ll_trunc_readers, 0);
344 wake_up_var(&sem->ll_trunc_readers);
347 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
351 spin_lock(&lli->lli_layout_lock);
352 gen = lli->lli_layout_gen;
353 spin_unlock(&lli->lli_layout_lock);
358 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
360 spin_lock(&lli->lli_layout_lock);
361 lli->lli_layout_gen = gen;
362 spin_unlock(&lli->lli_layout_lock);
366 /* File data is modified. */
367 LLIF_DATA_MODIFIED = 0,
368 /* File is being restored */
369 LLIF_FILE_RESTORING = 1,
370 /* Xattr cache is attached to the file */
371 LLIF_XATTR_CACHE = 2,
372 /* Project inherit */
373 LLIF_PROJECT_INHERIT = 3,
374 /* update atime from MDS even if it's older than local inode atime. */
375 LLIF_UPDATE_ATIME = 4,
379 static inline void ll_file_set_flag(struct ll_inode_info *lli,
380 enum ll_file_flags flag)
382 set_bit(flag, &lli->lli_flags);
385 static inline void ll_file_clear_flag(struct ll_inode_info *lli,
386 enum ll_file_flags flag)
388 clear_bit(flag, &lli->lli_flags);
391 static inline bool ll_file_test_flag(struct ll_inode_info *lli,
392 enum ll_file_flags flag)
394 return test_bit(flag, &lli->lli_flags);
397 static inline bool ll_file_test_and_clear_flag(struct ll_inode_info *lli,
398 enum ll_file_flags flag)
400 return test_and_clear_bit(flag, &lli->lli_flags);
403 int ll_xattr_cache_destroy(struct inode *inode);
405 int ll_xattr_cache_get(struct inode *inode,
411 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
413 #ifdef CONFIG_SECURITY
414 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
415 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
421 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
423 #ifdef CONFIG_SECURITY
424 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
428 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
429 const char **secctx_name, void **secctx,
431 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
434 int ll_listsecurity(struct inode *inode, char *secctx_name,
435 size_t secctx_name_size);
438 * Locking to guarantee consistency of non-atomic updates to long long i_size,
439 * consistency between file size and KMS.
441 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
444 void ll_inode_size_lock(struct inode *inode);
445 void ll_inode_size_unlock(struct inode *inode);
447 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
449 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
452 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
454 return ll_i2info(inode)->lli_pcc_inode;
457 /* default to use at least 16M for fast read if possible */
458 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
460 /* default readahead on a given system. */
461 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(64UL)
463 /* default read-ahead full files smaller than limit on the second read */
464 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
469 RA_STAT_DISTANT_READPAGE,
470 RA_STAT_MISS_IN_WINDOW,
471 RA_STAT_FAILED_GRAB_PAGE,
472 RA_STAT_FAILED_MATCH,
477 RA_STAT_MAX_IN_FLIGHT,
478 RA_STAT_WRONG_GRAB_PAGE,
479 RA_STAT_FAILED_REACH_END,
481 RA_STAT_FAILED_FAST_READ,
486 atomic_t ra_cur_pages;
487 unsigned long ra_max_pages;
488 unsigned long ra_max_pages_per_file;
489 unsigned long ra_max_read_ahead_whole_pages;
490 struct workqueue_struct *ll_readahead_wq;
492 * Max number of active works could be triggered
493 * for async readahead.
495 unsigned int ra_async_max_active;
496 /* how many async readahead triggered in flight */
497 atomic_t ra_async_inflight;
498 /* Threshold to control when to trigger async readahead */
499 unsigned long ra_async_pages_per_file_threshold;
502 /* ra_io_arg will be filled in the beginning of ll_readahead with
503 * ras_lock, then the following ll_read_ahead_pages will read RA
504 * pages according to this arg, all the items in this structure are
505 * counted by page index.
508 pgoff_t ria_start_idx; /* start offset of read-ahead*/
509 pgoff_t ria_end_idx; /* end offset of read-ahead*/
510 unsigned long ria_reserved; /* reserved pages for read-ahead */
511 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
512 bool ria_eof; /* reach end of file */
513 /* If stride read pattern is detected, ria_stoff is the byte offset
514 * where stride read is started. Note: for normal read-ahead, the
515 * value here is meaningless, and also it will not be accessed*/
517 /* ria_length and ria_bytes are the length and pages length in the
518 * stride I/O mode. And they will also be used to check whether
519 * it is stride I/O read-ahead in the read-ahead pages*/
524 /* LL_HIST_MAX=32 causes an overflow */
525 #define LL_HIST_MAX 28
526 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
527 #define LL_PROCESS_HIST_MAX 10
528 struct per_process_info {
530 struct obd_histogram pp_r_hist;
531 struct obd_histogram pp_w_hist;
534 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
535 struct ll_rw_extents_info {
536 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
539 #define LL_OFFSET_HIST_MAX 100
540 struct ll_rw_process_info {
543 loff_t rw_range_start;
545 loff_t rw_last_file_pos;
547 size_t rw_smallest_extent;
548 size_t rw_largest_extent;
549 struct ll_file_data *rw_last_file;
552 enum stats_track_type {
553 STATS_TRACK_ALL = 0, /* track all processes */
554 STATS_TRACK_PID, /* track process with this pid */
555 STATS_TRACK_PPID, /* track processes with this ppid */
556 STATS_TRACK_GID, /* track processes with this gid */
560 /* flags for sbi->ll_flags */
561 #define LL_SBI_NOLCK 0x01 /* DLM locking disabled (directio-only) */
562 #define LL_SBI_CHECKSUM 0x02 /* checksum each page as it's written */
563 #define LL_SBI_FLOCK 0x04
564 #define LL_SBI_USER_XATTR 0x08 /* support user xattr */
565 #define LL_SBI_ACL 0x10 /* support ACL */
566 /* LL_SBI_RMT_CLIENT 0x40 remote client */
567 #define LL_SBI_MDS_CAPA 0x80 /* support mds capa, obsolete */
568 #define LL_SBI_OSS_CAPA 0x100 /* support oss capa, obsolete */
569 #define LL_SBI_LOCALFLOCK 0x200 /* Local flocks support by kernel */
570 #define LL_SBI_LRU_RESIZE 0x400 /* lru resize support */
571 #define LL_SBI_LAZYSTATFS 0x800 /* lazystatfs mount option */
572 /* LL_SBI_SOM_PREVIEW 0x1000 SOM preview mount option, obsolete */
573 #define LL_SBI_32BIT_API 0x2000 /* generate 32 bit inodes. */
574 #define LL_SBI_64BIT_HASH 0x4000 /* support 64-bits dir hash/offset */
575 #define LL_SBI_AGL_ENABLED 0x8000 /* enable agl */
576 #define LL_SBI_VERBOSE 0x10000 /* verbose mount/umount */
577 #define LL_SBI_LAYOUT_LOCK 0x20000 /* layout lock support */
578 #define LL_SBI_USER_FID2PATH 0x40000 /* allow fid2path by unprivileged users */
579 #define LL_SBI_XATTR_CACHE 0x80000 /* support for xattr cache */
580 #define LL_SBI_NOROOTSQUASH 0x100000 /* do not apply root squash */
581 #define LL_SBI_ALWAYS_PING 0x200000 /* always ping even if server
583 #define LL_SBI_FAST_READ 0x400000 /* fast read support */
584 #define LL_SBI_FILE_SECCTX 0x800000 /* set file security context at create */
585 /* LL_SBI_PIO 0x1000000 parallel IO support, introduced in
587 #define LL_SBI_TINY_WRITE 0x2000000 /* tiny write support */
588 #define LL_SBI_FILE_HEAT 0x4000000 /* file heat support */
589 #define LL_SBI_TEST_DUMMY_ENCRYPTION 0x8000000 /* test dummy encryption */
590 #define LL_SBI_ENCRYPT 0x10000000 /* client side encryption */
591 #define LL_SBI_FLAGS { \
619 "test_dummy_encryption", \
623 /* This is embedded into llite super-blocks to keep track of connect
624 * flags (capabilities) supported by all imports given mount is
626 struct lustre_client_ocd {
627 /* This is conjunction of connect_flags across all imports
628 * (LOVs) this mount is connected to. This field is updated by
629 * cl_ocd_update() under ->lco_lock. */
631 struct mutex lco_lock;
632 struct obd_export *lco_md_exp;
633 struct obd_export *lco_dt_exp;
637 /* this protects pglist and ra_info. It isn't safe to
638 * grab from interrupt contexts */
640 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
641 spinlock_t ll_process_lock; /* ll_rw_process_info */
642 struct obd_uuid ll_sb_uuid;
643 struct obd_export *ll_md_exp;
644 struct obd_export *ll_dt_exp;
645 struct obd_device *ll_md_obd;
646 struct obd_device *ll_dt_obd;
647 struct dentry *ll_debugfs_entry;
648 struct lu_fid ll_root_fid; /* root object fid */
651 unsigned int ll_xattr_cache_enabled:1,
652 ll_xattr_cache_set:1, /* already set to 0/1 */
653 ll_client_common_fill_super_succeeded:1,
656 struct lustre_client_ocd ll_lco;
658 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
660 /* Used to track "unstable" pages on a client, and maintain a
661 * LRU list of clean pages. An "unstable" page is defined as
662 * any page which is sent to a server as part of a bulk request,
663 * but is uncommitted to stable storage. */
664 struct cl_client_cache *ll_cache;
666 struct lprocfs_stats *ll_ra_stats;
668 struct ll_ra_info ll_ra_info;
669 unsigned int ll_namelen;
670 struct file_operations *ll_fop;
672 struct lu_site *ll_site;
673 struct cl_device *ll_cl;
675 struct ll_rw_extents_info ll_rw_extents_info;
676 int ll_extent_process_count;
677 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
678 unsigned int ll_offset_process_count;
679 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
680 unsigned int ll_rw_offset_entry_count;
681 int ll_stats_track_id;
682 enum stats_track_type ll_stats_track_type;
685 /* metadata stat-ahead */
686 unsigned int ll_sa_running_max;/* max concurrent
687 * statahead instances */
688 unsigned int ll_sa_max; /* max statahead RPCs */
689 atomic_t ll_sa_total; /* statahead thread started
691 atomic_t ll_sa_wrong; /* statahead thread stopped for
693 atomic_t ll_sa_running; /* running statahead thread
695 atomic_t ll_agl_total; /* AGL thread started count */
697 dev_t ll_sdev_orig; /* save s_dev before assign for
700 struct root_squash_info ll_squash;
703 /* st_blksize returned by stat(2), when non-zero */
704 unsigned int ll_stat_blksize;
706 /* maximum relative age of cached statfs results */
707 unsigned int ll_statfs_max_age;
709 struct kset ll_kset; /* sysfs object */
710 struct completion ll_kobj_unregister;
713 unsigned int ll_heat_decay_weight;
714 unsigned int ll_heat_period_second;
716 /* filesystem fsname */
717 char ll_fsname[LUSTRE_MAXFSNAME + 1];
719 /* Persistent Client Cache */
720 struct pcc_super ll_pcc_super;
723 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
724 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
726 * per file-descriptor read-ahead data.
728 struct ll_readahead_state {
730 /* End byte that read(2) try to read. */
731 loff_t ras_last_read_end_bytes;
733 * number of bytes read after last read-ahead window reset. As window
734 * is reset on each seek, this is effectively a number of consecutive
735 * accesses. Maybe ->ras_accessed_in_window is better name.
737 * XXX nikita: window is also reset (by ras_update()) when Lustre
738 * believes that memory pressure evicts read-ahead pages. In that
739 * case, it probably doesn't make sense to expand window to
740 * PTLRPC_MAX_BRW_PAGES on the third access.
742 loff_t ras_consecutive_bytes;
744 * number of read requests after the last read-ahead window reset
745 * As window is reset on each seek, this is effectively the number
746 * on consecutive read request and is used to trigger read-ahead.
748 unsigned long ras_consecutive_requests;
750 * Parameters of current read-ahead window. Handled by
751 * ras_update(). On the initial access to the file or after a seek,
752 * window is reset to 0. After 3 consecutive accesses, window is
753 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
754 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
756 pgoff_t ras_window_start_idx;
757 pgoff_t ras_window_pages;
759 * Optimal RPC size in pages.
760 * It decides how many pages will be sent for each read-ahead.
762 unsigned long ras_rpc_pages;
764 * Where next read-ahead should start at. This lies within read-ahead
765 * window. Read-ahead window is read in pieces rather than at once
766 * because: 1. lustre limits total number of pages under read-ahead by
767 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
768 * not covered by DLM lock.
770 pgoff_t ras_next_readahead_idx;
772 * Total number of ll_file_read requests issued, reads originating
773 * due to mmap are not counted in this total. This value is used to
774 * trigger full file read-ahead after multiple reads to a small file.
776 unsigned long ras_requests;
778 * The following 3 items are used for detecting the stride I/O
780 * In stride I/O mode,
781 * ...............|-----data-----|****gap*****|--------|******|....
782 * offset |-stride_bytes-|-stride_gap-|
783 * ras_stride_offset = offset;
784 * ras_stride_length = stride_bytes + stride_gap;
785 * ras_stride_bytes = stride_bytes;
786 * Note: all these three items are counted by bytes.
788 loff_t ras_stride_offset;
789 loff_t ras_stride_length;
790 loff_t ras_stride_bytes;
792 * number of consecutive stride request count, and it is similar as
793 * ras_consecutive_requests, but used for stride I/O mode.
794 * Note: only more than 2 consecutive stride request are detected,
795 * stride read-ahead will be enable
797 unsigned long ras_consecutive_stride_requests;
798 /* index of the last page that async readahead starts */
799 pgoff_t ras_async_last_readpage_idx;
800 /* whether we should increase readahead window */
801 bool ras_need_increase_window;
802 /* whether ra miss check should be skipped */
803 bool ras_no_miss_check;
806 struct ll_readahead_work {
807 /** File to readahead */
808 struct file *lrw_file;
809 pgoff_t lrw_start_idx;
812 /* async worker to handler read */
813 struct work_struct lrw_readahead_work;
814 char lrw_jobid[LUSTRE_JOBID_SIZE];
817 extern struct kmem_cache *ll_file_data_slab;
818 struct lustre_handle;
819 struct ll_file_data {
820 struct ll_readahead_state fd_ras;
821 struct ll_grouplock fd_grouplock;
825 /* openhandle if lease exists for this file.
826 * Borrow lli->lli_och_mutex to protect assignment */
827 struct obd_client_handle *fd_lease_och;
828 struct obd_client_handle *fd_och;
829 struct file *fd_file;
830 /* Indicate whether need to report failure when close.
831 * true: failure is known, not report again.
832 * false: unknown failure, should report. */
833 bool fd_write_failed;
834 bool ll_lock_no_expand;
835 rwlock_t fd_lock; /* protect lcc list */
836 struct list_head fd_lccs; /* list of ll_cl_context */
837 /* Used by mirrored file to lead IOs to a specific mirror, usually
838 * for mirror resync. 0 means default. */
839 __u32 fd_designated_mirror;
840 /* The layout version when resync starts. Resync I/O should carry this
841 * layout version for verification to OST objects */
842 __u32 fd_layout_version;
843 struct pcc_file fd_pcc_file;
846 void llite_tunables_unregister(void);
847 int llite_tunables_register(void);
849 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
851 return &lli->lli_vfs_inode;
854 __u32 ll_i2suppgid(struct inode *i);
855 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
857 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
859 #if BITS_PER_LONG == 32
861 #elif defined(CONFIG_COMPAT)
862 if (unlikely(sbi->ll_flags & LL_SBI_32BIT_API))
865 # ifdef CONFIG_X86_X32
866 /* in_compat_syscall() returns true when called from a kthread
867 * and CONFIG_X86_X32 is enabled, which is wrong. So check
868 * whether the caller comes from a syscall (ie. not a kthread)
869 * before calling in_compat_syscall(). */
870 if (current->flags & PF_KTHREAD)
874 return unlikely(in_compat_syscall());
876 return unlikely(sbi->ll_flags & LL_SBI_32BIT_API);
880 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
882 return !!(sbi->ll_flags & LL_SBI_FAST_READ);
885 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
887 return !!(sbi->ll_flags & LL_SBI_TINY_WRITE);
890 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
892 return !!(sbi->ll_flags & LL_SBI_FILE_HEAT);
895 void ll_ras_enter(struct file *f, loff_t pos, size_t count);
897 /* llite/lcommon_misc.c */
898 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
899 enum obd_notify_event ev, void *owner);
900 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
901 struct ll_grouplock *lg);
902 void cl_put_grouplock(struct ll_grouplock *lg);
904 /* llite/lproc_llite.c */
905 int ll_debugfs_register_super(struct super_block *sb, const char *name);
906 void ll_debugfs_unregister_super(struct super_block *sb);
907 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
911 LPROC_LL_WRITE_BYTES,
938 LPROC_LL_GETXATTR_HITS,
940 LPROC_LL_REMOVEXATTR,
943 LPROC_LL_FILE_OPCODES
947 enum get_default_layout_type {
948 GET_DEFAULT_LAYOUT_ROOT = 1,
951 struct ll_dir_chain {
954 static inline void ll_dir_chain_init(struct ll_dir_chain *chain)
958 static inline void ll_dir_chain_fini(struct ll_dir_chain *chain)
962 extern const struct file_operations ll_dir_operations;
963 extern const struct inode_operations ll_dir_inode_operations;
964 #ifdef HAVE_DIR_CONTEXT
965 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
966 struct dir_context *ctx);
968 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
969 void *cookie, filldir_t filldir);
971 int ll_get_mdt_idx(struct inode *inode);
972 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
973 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
974 __u64 offset, struct ll_dir_chain *chain);
975 void ll_release_page(struct inode *inode, struct page *page, bool remove);
978 extern const struct inode_operations ll_special_inode_operations;
980 struct inode *ll_iget(struct super_block *sb, ino_t hash,
981 struct lustre_md *lic);
982 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
983 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
984 void *data, int flag);
985 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
986 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
987 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
990 int ll_writepage(struct page *page, struct writeback_control *wbc);
991 int ll_writepages(struct address_space *, struct writeback_control *wbc);
992 int ll_readpage(struct file *file, struct page *page);
993 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
994 struct cl_page *page, struct file *file);
995 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
996 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
999 void ll_cl_add(struct file *file, const struct lu_env *env, struct cl_io *io,
1000 enum lcc_type type);
1001 void ll_cl_remove(struct file *file, const struct lu_env *env);
1002 struct ll_cl_context *ll_cl_find(struct file *file);
1004 extern const struct address_space_operations ll_aops;
1007 extern struct file_operations ll_file_operations;
1008 extern struct file_operations ll_file_operations_flock;
1009 extern struct file_operations ll_file_operations_noflock;
1010 extern struct inode_operations ll_file_inode_operations;
1011 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
1012 enum ldlm_mode l_req_mode);
1013 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1014 struct lustre_handle *lockh, __u64 flags,
1015 enum ldlm_mode mode);
1017 int ll_file_open(struct inode *inode, struct file *file);
1018 int ll_file_release(struct inode *inode, struct file *file);
1019 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1020 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1021 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1022 struct ll_file_data *file, loff_t pos,
1023 size_t count, int rw);
1024 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
1025 int ll_getattr(const struct path *path, struct kstat *stat,
1026 u32 request_mask, unsigned int flags);
1028 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1030 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1031 unsigned int flags);
1032 struct posix_acl *ll_get_acl(struct inode *inode, int type);
1033 #ifdef HAVE_IOP_SET_ACL
1034 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1035 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1036 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1037 #define ll_set_acl NULL
1038 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1042 static inline int ll_xflags_to_inode_flags(int xflags)
1044 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1045 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1046 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1047 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1050 static inline int ll_inode_flags_to_xflags(int flags)
1052 return ((flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1053 ((flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1054 ((flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1055 ((flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1058 int ll_migrate(struct inode *parent, struct file *file,
1059 struct lmv_user_md *lum, const char *name);
1060 int ll_get_fid_by_name(struct inode *parent, const char *name,
1061 int namelen, struct lu_fid *fid, struct inode **inode);
1062 int ll_inode_permission(struct inode *inode, int mask);
1063 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa);
1064 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1066 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1069 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1070 __u64 flags, struct lov_user_md *lum,
1072 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1073 struct lov_mds_md **lmm, int *lmm_size,
1074 struct ptlrpc_request **request);
1075 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1077 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1078 int *lmm_size, struct ptlrpc_request **request,
1079 struct ptlrpc_request **root_request, u64 valid);
1080 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1081 struct ptlrpc_request **request, u64 valid);
1082 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1083 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1084 int ll_fid2path(struct inode *inode, void __user *arg);
1085 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1086 int ll_hsm_release(struct inode *inode);
1087 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1088 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1090 /* llite/dcache.c */
1092 int ll_d_init(struct dentry *de);
1093 extern const struct dentry_operations ll_d_ops;
1094 void ll_intent_drop_lock(struct lookup_intent *);
1095 void ll_intent_release(struct lookup_intent *);
1096 void ll_invalidate_aliases(struct inode *);
1097 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1098 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1099 struct lookup_intent *it, struct dentry *de);
1101 /* llite/llite_lib.c */
1102 extern struct super_operations lustre_super_operations;
1104 void ll_lli_init(struct ll_inode_info *lli);
1105 int ll_fill_super(struct super_block *sb);
1106 void ll_put_super(struct super_block *sb);
1107 void ll_kill_super(struct super_block *sb);
1108 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1109 void ll_dir_clear_lsm_md(struct inode *inode);
1110 void ll_clear_inode(struct inode *inode);
1111 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1112 enum op_xvalid xvalid, bool hsm_import);
1113 int ll_setattr(struct dentry *de, struct iattr *attr);
1114 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1115 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1117 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1118 void ll_update_inode_flags(struct inode *inode, int ext_flags);
1119 int ll_read_inode2(struct inode *inode, void *opaque);
1120 void ll_delete_inode(struct inode *inode);
1121 int ll_iocontrol(struct inode *inode, struct file *file,
1122 unsigned int cmd, unsigned long arg);
1123 int ll_flush_ctx(struct inode *inode);
1124 void ll_umount_begin(struct super_block *sb);
1125 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1126 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1127 void ll_dirty_page_discard_warn(struct page *page, int ioret);
1128 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1129 struct super_block *, struct lookup_intent *);
1130 int ll_obd_statfs(struct inode *inode, void __user *arg);
1131 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1132 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1133 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1135 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1136 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1137 struct inode *i1, struct inode *i2,
1138 const char *name, size_t namelen,
1139 __u32 mode, enum md_op_code opc,
1141 void ll_finish_md_op_data(struct md_op_data *op_data);
1142 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
1143 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1144 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1145 struct lov_user_md **kbuf);
1146 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req);
1148 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req,
1149 struct lookup_intent *it);
1151 /* Compute expected user md size when passing in a md from user space */
1152 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1154 switch (lum->lmm_magic) {
1155 case LOV_USER_MAGIC_V1:
1156 return sizeof(struct lov_user_md_v1);
1157 case LOV_USER_MAGIC_V3:
1158 return sizeof(struct lov_user_md_v3);
1159 case LOV_USER_MAGIC_SPECIFIC:
1160 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1163 return lov_user_md_size(lum->lmm_stripe_count,
1164 LOV_USER_MAGIC_SPECIFIC);
1165 case LOV_USER_MAGIC_COMP_V1:
1166 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1167 case LOV_USER_MAGIC_FOREIGN:
1168 return foreign_size(lum);
1174 /* llite/llite_nfs.c */
1175 extern struct export_operations lustre_export_operations;
1176 __u32 get_uuid2int(const char *name, int len);
1177 struct inode *search_inode_for_lustre(struct super_block *sb,
1178 const struct lu_fid *fid);
1179 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1181 /* llite/symlink.c */
1182 extern struct inode_operations ll_fast_symlink_inode_operations;
1185 * IO arguments for various VFS I/O interfaces.
1187 struct vvp_io_args {
1188 /** normal/sendfile/splice */
1189 enum vvp_io_subtype via_io_subtype;
1193 struct kiocb *via_iocb;
1194 struct iov_iter *via_iter;
1197 struct pipe_inode_info *via_pipe;
1198 unsigned int via_flags;
1208 struct ll_cl_context {
1209 struct list_head lcc_list;
1211 const struct lu_env *lcc_env;
1212 struct cl_io *lcc_io;
1213 struct cl_page *lcc_page;
1214 enum lcc_type lcc_type;
1217 struct ll_thread_info {
1218 struct vvp_io_args lti_args;
1219 struct ra_io_arg lti_ria;
1220 struct ll_cl_context lti_io_ctx;
1223 extern struct lu_context_key ll_thread_key;
1225 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1227 struct ll_thread_info *lti;
1229 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1230 LASSERT(lti != NULL);
1235 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env,
1236 enum vvp_io_subtype type)
1238 struct vvp_io_args *via = &ll_env_info(env)->lti_args;
1240 via->via_io_subtype = type;
1245 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1246 struct vvp_io_args *args);
1248 /* llite/llite_mmap.c */
1250 int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last);
1251 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1252 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1253 unsigned long addr, size_t count);
1254 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1257 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1259 /* don't need an addref as the sb_info should be holding one */
1260 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1262 return ll_s2sbi(sb)->ll_dt_exp;
1265 /* don't need an addref as the sb_info should be holding one */
1266 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1268 return ll_s2sbi(sb)->ll_md_exp;
1271 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1273 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1279 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1280 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1282 return ll_s2sbi(inode->i_sb);
1285 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1287 return ll_s2dtexp(inode->i_sb);
1290 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1292 return ll_s2mdexp(inode->i_sb);
1295 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1299 LASSERT(inode != NULL);
1300 fid = &ll_i2info(inode)->lli_fid;
1305 static inline bool ll_dir_striped(struct inode *inode)
1308 return S_ISDIR(inode->i_mode) &&
1309 lmv_dir_striped(ll_i2info(inode)->lli_lsm_md);
1312 static inline loff_t ll_file_maxbytes(struct inode *inode)
1314 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1317 return MAX_LFS_FILESIZE;
1319 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1323 extern const struct xattr_handler *ll_xattr_handlers[];
1325 #define XATTR_USER_T 1
1326 #define XATTR_TRUSTED_T 2
1327 #define XATTR_SECURITY_T 3
1328 #define XATTR_ACL_ACCESS_T 4
1329 #define XATTR_ACL_DEFAULT_T 5
1330 #define XATTR_LUSTRE_T 6
1331 #define XATTR_OTHER_T 7
1333 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1334 int ll_xattr_list(struct inode *inode, const char *name, int type,
1335 void *buffer, size_t size, u64 valid);
1336 const struct xattr_handler *get_xattr_type(const char *name);
1339 * Common IO arguments for various VFS I/O interfaces.
1341 int cl_sb_init(struct super_block *sb);
1342 int cl_sb_fini(struct super_block *sb);
1344 enum ras_update_flags {
1348 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1349 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1353 #define LL_SA_RPC_MIN 2
1354 #define LL_SA_RPC_DEF 32
1355 #define LL_SA_RPC_MAX 512
1357 /* XXX: If want to support more concurrent statahead instances,
1358 * please consider to decentralize the RPC lists attached
1359 * on related import, such as imp_{sending,delayed}_list.
1361 #define LL_SA_RUNNING_MAX 256
1362 #define LL_SA_RUNNING_DEF 16
1364 #define LL_SA_CACHE_BIT 5
1365 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1366 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1368 /* per inode struct, for dir only */
1369 struct ll_statahead_info {
1370 struct dentry *sai_dentry;
1371 atomic_t sai_refcount; /* when access this struct, hold
1373 unsigned int sai_max; /* max ahead of lookup */
1374 __u64 sai_sent; /* stat requests sent count */
1375 __u64 sai_replied; /* stat requests which received
1377 __u64 sai_index; /* index of statahead entry */
1378 __u64 sai_index_wait; /* index of entry which is the
1379 * caller is waiting for */
1380 __u64 sai_hit; /* hit count */
1381 __u64 sai_miss; /* miss count:
1382 * for "ls -al" case, includes
1383 * hidden dentry miss;
1384 * for "ls -l" case, it does not
1385 * include hidden dentry miss.
1386 * "sai_miss_hidden" is used for
1389 unsigned int sai_consecutive_miss; /* consecutive miss */
1390 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1391 * is not a hidden one */
1392 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1394 unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
1396 sai_agl_valid:1,/* AGL is valid for the dir */
1397 sai_in_readpage:1;/* statahead is in readdir()*/
1398 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1399 struct task_struct *sai_task; /* stat-ahead thread */
1400 struct task_struct *sai_agl_task; /* AGL thread */
1401 struct list_head sai_interim_entries; /* entries which got async
1402 * stat reply, but not
1404 struct list_head sai_entries; /* completed entries */
1405 struct list_head sai_agls; /* AGLs to be sent */
1406 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1407 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1408 atomic_t sai_cache_count; /* entry count in cache */
1411 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1413 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1414 void ll_authorize_statahead(struct inode *dir, void *key);
1415 void ll_deauthorize_statahead(struct inode *dir, void *key);
1418 blkcnt_t dirty_cnt(struct inode *inode);
1420 int cl_glimpse_size0(struct inode *inode, int agl);
1421 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1422 struct inode *inode, struct cl_object *clob, int agl);
1424 static inline int cl_glimpse_size(struct inode *inode)
1426 return cl_glimpse_size0(inode, 0);
1429 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1430 * part of statahead */
1431 static inline int cl_agl(struct inode *inode)
1433 return cl_glimpse_size0(inode, 1);
1436 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1438 int cl_io_get(struct inode *inode, struct lu_env **envout,
1439 struct cl_io **ioout, __u16 *refcheck);
1441 static inline int ll_glimpse_size(struct inode *inode)
1443 struct ll_inode_info *lli = ll_i2info(inode);
1446 down_read(&lli->lli_glimpse_sem);
1447 rc = cl_glimpse_size(inode);
1448 lli->lli_glimpse_time = ktime_get();
1449 up_read(&lli->lli_glimpse_sem);
1453 /* dentry may statahead when statahead is enabled and current process has opened
1454 * parent directory, and this dentry hasn't accessed statahead cache before */
1456 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1458 struct ll_inode_info *lli;
1459 struct ll_dentry_data *ldd;
1461 if (ll_i2sbi(dir)->ll_sa_max == 0)
1464 lli = ll_i2info(dir);
1466 /* statahead is not allowed for this dir, there may be three causes:
1467 * 1. dir is not opened.
1468 * 2. statahead hit ratio is too low.
1469 * 3. previous stat started statahead thread failed. */
1470 if (!lli->lli_sa_enabled)
1473 /* not the same process, don't statahead */
1474 if (lli->lli_opendir_pid != current->pid)
1478 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1479 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1480 * For patchless client, lookup intent is not accurate, which may
1481 * misguide statahead. For example:
1482 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1483 * have the same intent -- IT_GETATTR, while one dentry should access
1484 * statahead cache once, otherwise statahead windows is messed up.
1485 * The solution is as following:
1486 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1487 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1488 * bypass interacting with statahead cache by checking
1489 * 'lld_sa_generation == lli->lli_sa_generation'.
1491 ldd = ll_d2d(dentry);
1492 if (ldd != NULL && lli->lli_sa_generation &&
1493 ldd->lld_sa_generation == lli->lli_sa_generation)
1499 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1500 enum cl_fsync_mode mode, int ignore_layout);
1502 static inline int ll_file_nolock(const struct file *file)
1504 struct ll_file_data *fd = file->private_data;
1505 struct inode *inode = file_inode((struct file *)file);
1507 LASSERT(fd != NULL);
1508 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1509 (ll_i2sbi(inode)->ll_flags & LL_SBI_NOLCK));
1512 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1513 struct lookup_intent *it, __u64 *bits)
1515 if (!it->it_lock_set) {
1516 struct lustre_handle handle;
1518 /* If this inode is a remote object, it will get two
1519 * separate locks in different namespaces, Master MDT,
1520 * where the name entry is, will grant LOOKUP lock,
1521 * remote MDT, where the object is, will grant
1522 * UPDATE|PERM lock. The inode will be attched to both
1523 * LOOKUP and PERM locks, so revoking either locks will
1524 * case the dcache being cleared */
1525 if (it->it_remote_lock_mode) {
1526 handle.cookie = it->it_remote_lock_handle;
1527 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1528 "(%p) for remote lock %#llx\n",
1529 PFID(ll_inode2fid(inode)), inode,
1531 md_set_lock_data(exp, &handle, inode, NULL);
1534 handle.cookie = it->it_lock_handle;
1536 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1537 " for lock %#llx\n",
1538 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1540 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1541 it->it_lock_set = 1;
1545 *bits = it->it_lock_bits;
1548 static inline int d_lustre_invalid(const struct dentry *dentry)
1550 struct ll_dentry_data *lld = ll_d2d(dentry);
1552 return (lld == NULL) || lld->lld_invalid;
1555 static inline void __d_lustre_invalidate(struct dentry *dentry)
1557 struct ll_dentry_data *lld = ll_d2d(dentry);
1560 lld->lld_invalid = 1;
1564 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1565 * ll_md_blocking_ast), unhash this dentry, and let dcache to reclaim it later;
1566 * else dput() of the last refcount will unhash this dentry and kill it.
1568 static inline void d_lustre_invalidate(struct dentry *dentry, int nested)
1570 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1572 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1574 spin_lock_nested(&dentry->d_lock,
1575 nested ? DENTRY_D_LOCK_NESTED : DENTRY_D_LOCK_NORMAL);
1576 __d_lustre_invalidate(dentry);
1578 * We should be careful about dentries created by d_obtain_alias().
1579 * These dentries are not put in the dentry tree, instead they are
1580 * linked to sb->s_anon through dentry->d_hash.
1581 * shrink_dcache_for_umount() shrinks the tree and sb->s_anon list.
1582 * If we unhashed such a dentry, unmount would not be able to find
1583 * it and busy inodes would be reported.
1585 if (ll_d_count(dentry) == 0 && !(dentry->d_flags & DCACHE_DISCONNECTED))
1587 spin_unlock(&dentry->d_lock);
1590 static inline void d_lustre_revalidate(struct dentry *dentry)
1592 spin_lock(&dentry->d_lock);
1593 LASSERT(ll_d2d(dentry) != NULL);
1594 ll_d2d(dentry)->lld_invalid = 0;
1595 spin_unlock(&dentry->d_lock);
1598 static inline dev_t ll_compat_encode_dev(dev_t dev)
1600 /* The compat_sys_*stat*() syscalls will fail unless the
1601 * device majors and minors are both less than 256. Note that
1602 * the value returned here will be passed through
1603 * old_encode_dev() in cp_compat_stat(). And so we are not
1604 * trying to return a valid compat (u16) device number, just
1605 * one that will pass the old_valid_dev() check. */
1607 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1610 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1611 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1612 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1613 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1614 struct lu_extent *ext);
1616 int ll_xattr_init(void);
1617 void ll_xattr_fini(void);
1619 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1620 struct cl_page *page, enum cl_req_type crt);
1622 int ll_getparent(struct file *file, struct getparent __user *arg);
1625 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1626 enum op_xvalid xvalid, unsigned int attr_flags);
1628 extern struct lu_env *cl_inode_fini_env;
1629 extern __u16 cl_inode_fini_refcheck;
1631 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1632 void cl_inode_fini(struct inode *inode);
1634 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1635 u32 cl_fid_build_gen(const struct lu_fid *fid);
1637 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1639 return &ll_i2sbi(inode)->ll_pcc_super;
1642 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1644 return ll_i2pccs(ll_info2i(lli));
1647 #ifdef HAVE_LUSTRE_CRYPTO
1649 extern const struct llcrypt_operations lustre_cryptops;
1652 #endif /* LLITE_INTERNAL_H */