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 HAVE_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;
188 struct mutex lli_setattr_mutex;
190 struct rw_semaphore lli_glimpse_sem;
191 ktime_t lli_glimpse_time;
192 struct list_head lli_agl_list;
195 /* for writepage() only to communicate to fsync */
198 /* protect the file heat fields */
199 spinlock_t lli_heat_lock;
200 __u32 lli_heat_flags;
201 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
204 * Whenever a process try to read/write the file, the
205 * jobid of the process will be saved here, and it'll
206 * be packed into the write PRC when flush later.
208 * So the read/write statistics for jobid will not be
209 * accurate if the file is shared by different jobs.
211 char lli_jobid[LUSTRE_JOBID_SIZE];
213 struct mutex lli_pcc_lock;
214 enum lu_pcc_state_flags lli_pcc_state;
216 * @lli_pcc_generation saves the gobal PCC generation
217 * when the file was successfully attached into PCC.
218 * The flags of the PCC dataset are saved in
220 * The gobal PCC generation will be increased when add
221 * or delete a PCC backend, or change the configuration
222 * parameters for PCC.
223 * If @lli_pcc_generation is same as the gobal PCC
224 * generation, we can use the saved flags of the PCC
225 * dataset to determine whether need to try auto attach
228 __u64 lli_pcc_generation;
229 enum pcc_dataset_flags lli_pcc_dsflags;
230 struct pcc_inode *lli_pcc_inode;
232 struct mutex lli_group_mutex;
233 __u64 lli_group_users;
234 unsigned long lli_group_gid;
236 __u64 lli_attr_valid;
238 __u64 lli_lazyblocks;
242 /* XXX: For following frequent used members, although they maybe special
243 * used for non-directory object, it is some time-wasting to check
244 * whether the object is directory or not before using them. On the
245 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
246 * the "ll_inode_info" size even if moving those members into u.f.
247 * So keep them out side.
249 * In the future, if more members are added only for directory,
250 * some of the following members can be moved into u.f.
252 struct cl_object *lli_clob;
254 /* mutex to request for layout lock exclusively. */
255 struct mutex lli_layout_mutex;
256 /* Layout version, protected by lli_layout_lock */
257 __u32 lli_layout_gen;
258 spinlock_t lli_layout_lock;
260 __u32 lli_projid; /* project id */
262 struct rw_semaphore lli_xattrs_list_rwsem;
263 struct mutex lli_xattrs_enq_lock;
264 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
267 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
269 atomic_set(&sem->ll_trunc_readers, 0);
270 atomic_set(&sem->ll_trunc_waiters, 0);
273 /* This version of down read ignores waiting writers, meaning if the semaphore
274 * is already held for read, this down_read will 'join' that reader and also
275 * take the semaphore.
277 * This lets us avoid an unusual deadlock.
279 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
280 * in Lustre, in order to exclude truncates. Some of these paths then need to
281 * take the mmap_sem, while still holding the trunc_sem. The problem is that
282 * page faults hold the mmap_sem when calling in to Lustre, and then must also
283 * take the trunc_sem to exclude truncate.
285 * This means the locking order for trunc_sem and mmap_sem is sometimes AB,
286 * sometimes BA. This is almost OK because in both cases, we take the trunc
287 * sem for read, so it doesn't block.
289 * However, if a write mode user (truncate, a setattr op) arrives in the
290 * middle of this, the second reader on the truncate_sem will wait behind that
293 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
294 * to the mode in which they take the semaphore):
295 * reader (holding mmap_sem, needs truncate_sem)
297 * reader (holding truncate sem, waiting for mmap_sem)
299 * And so the readers deadlock.
301 * The solution is this modified semaphore, where this down_read ignores
302 * waiting write operations, and all waiters are woken up at once, so readers
303 * using down_read_nowait cannot get stuck behind waiting writers, regardless
304 * of the order they arrived in.
306 * down_read_nowait is only used in the page fault case, where we already hold
307 * the mmap_sem. This is because otherwise repeated read and write operations
308 * (which take the truncate sem) could prevent a truncate from ever starting.
309 * This could still happen with page faults, but without an even more complex
310 * mechanism, this is unavoidable.
314 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
316 wait_var_event(&sem->ll_trunc_readers,
317 atomic_inc_unless_negative(&sem->ll_trunc_readers));
320 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
322 wait_var_event(&sem->ll_trunc_readers,
323 atomic_read(&sem->ll_trunc_waiters) == 0 &&
324 atomic_inc_unless_negative(&sem->ll_trunc_readers));
327 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
329 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
330 atomic_read(&sem->ll_trunc_waiters))
331 wake_up_var(&sem->ll_trunc_readers);
334 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
336 atomic_inc(&sem->ll_trunc_waiters);
337 wait_var_event(&sem->ll_trunc_readers,
338 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
339 atomic_dec(&sem->ll_trunc_waiters);
342 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
344 atomic_set(&sem->ll_trunc_readers, 0);
345 wake_up_var(&sem->ll_trunc_readers);
348 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
349 static inline void lli_clear_acl(struct ll_inode_info *lli)
351 if (lli->lli_posix_acl) {
352 posix_acl_release(lli->lli_posix_acl);
353 lli->lli_posix_acl = NULL;
357 static inline void lli_replace_acl(struct ll_inode_info *lli,
358 struct lustre_md *md)
360 spin_lock(&lli->lli_lock);
361 if (lli->lli_posix_acl)
362 posix_acl_release(lli->lli_posix_acl);
363 lli->lli_posix_acl = md->posix_acl;
364 spin_unlock(&lli->lli_lock);
367 static inline void lli_clear_acl(struct ll_inode_info *lli)
371 static inline void lli_replace_acl(struct ll_inode_info *lli,
372 struct lustre_md *md)
377 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
381 spin_lock(&lli->lli_layout_lock);
382 gen = lli->lli_layout_gen;
383 spin_unlock(&lli->lli_layout_lock);
388 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
390 spin_lock(&lli->lli_layout_lock);
391 lli->lli_layout_gen = gen;
392 spin_unlock(&lli->lli_layout_lock);
396 /* File data is modified. */
397 LLIF_DATA_MODIFIED = 0,
398 /* File is being restored */
399 LLIF_FILE_RESTORING = 1,
400 /* Xattr cache is attached to the file */
401 LLIF_XATTR_CACHE = 2,
402 /* Project inherit */
403 LLIF_PROJECT_INHERIT = 3,
404 /* update atime from MDS even if it's older than local inode atime. */
405 LLIF_UPDATE_ATIME = 4,
409 static inline void ll_file_set_flag(struct ll_inode_info *lli,
410 enum ll_file_flags flag)
412 set_bit(flag, &lli->lli_flags);
415 static inline void ll_file_clear_flag(struct ll_inode_info *lli,
416 enum ll_file_flags flag)
418 clear_bit(flag, &lli->lli_flags);
421 static inline bool ll_file_test_flag(struct ll_inode_info *lli,
422 enum ll_file_flags flag)
424 return test_bit(flag, &lli->lli_flags);
427 static inline bool ll_file_test_and_clear_flag(struct ll_inode_info *lli,
428 enum ll_file_flags flag)
430 return test_and_clear_bit(flag, &lli->lli_flags);
433 int ll_xattr_cache_destroy(struct inode *inode);
435 int ll_xattr_cache_get(struct inode *inode,
441 int ll_xattr_cache_insert(struct inode *inode,
446 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
448 #ifdef CONFIG_SECURITY
449 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
450 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
456 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
458 #ifdef CONFIG_SECURITY
459 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
463 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
464 const char **secctx_name, void **secctx,
466 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
469 int ll_listsecurity(struct inode *inode, char *secctx_name,
470 size_t secctx_name_size);
472 static inline bool obd_connect_has_enc(struct obd_connect_data *data)
474 #ifdef HAVE_LUSTRE_CRYPTO
475 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
476 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT;
482 static inline void obd_connect_set_enc(struct obd_connect_data *data)
484 #ifdef HAVE_LUSTRE_CRYPTO
485 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT;
490 * Locking to guarantee consistency of non-atomic updates to long long i_size,
491 * consistency between file size and KMS.
493 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
496 void ll_inode_size_lock(struct inode *inode);
497 void ll_inode_size_unlock(struct inode *inode);
499 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
501 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
504 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
506 return ll_i2info(inode)->lli_pcc_inode;
509 /* default to use at least 16M for fast read if possible */
510 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
512 /* default readahead on a given system. */
513 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(64UL)
515 /* default read-ahead full files smaller than limit on the second read */
516 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
521 RA_STAT_DISTANT_READPAGE,
522 RA_STAT_MISS_IN_WINDOW,
523 RA_STAT_FAILED_GRAB_PAGE,
524 RA_STAT_FAILED_MATCH,
529 RA_STAT_MAX_IN_FLIGHT,
530 RA_STAT_WRONG_GRAB_PAGE,
531 RA_STAT_FAILED_REACH_END,
533 RA_STAT_FAILED_FAST_READ,
538 atomic_t ra_cur_pages;
539 unsigned long ra_max_pages;
540 unsigned long ra_max_pages_per_file;
541 unsigned long ra_max_read_ahead_whole_pages;
542 struct workqueue_struct *ll_readahead_wq;
544 * Max number of active works could be triggered
545 * for async readahead.
547 unsigned int ra_async_max_active;
548 /* how many async readahead triggered in flight */
549 atomic_t ra_async_inflight;
550 /* Threshold to control when to trigger async readahead */
551 unsigned long ra_async_pages_per_file_threshold;
554 /* ra_io_arg will be filled in the beginning of ll_readahead with
555 * ras_lock, then the following ll_read_ahead_pages will read RA
556 * pages according to this arg, all the items in this structure are
557 * counted by page index.
560 pgoff_t ria_start_idx; /* start offset of read-ahead*/
561 pgoff_t ria_end_idx; /* end offset of read-ahead*/
562 unsigned long ria_reserved; /* reserved pages for read-ahead */
563 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
564 bool ria_eof; /* reach end of file */
565 /* If stride read pattern is detected, ria_stoff is the byte offset
566 * where stride read is started. Note: for normal read-ahead, the
567 * value here is meaningless, and also it will not be accessed*/
569 /* ria_length and ria_bytes are the length and pages length in the
570 * stride I/O mode. And they will also be used to check whether
571 * it is stride I/O read-ahead in the read-ahead pages*/
576 /* LL_HIST_MAX=32 causes an overflow */
577 #define LL_HIST_MAX 28
578 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
579 #define LL_PROCESS_HIST_MAX 10
580 struct per_process_info {
582 struct obd_histogram pp_r_hist;
583 struct obd_histogram pp_w_hist;
586 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
587 struct ll_rw_extents_info {
588 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
591 #define LL_OFFSET_HIST_MAX 100
592 struct ll_rw_process_info {
595 loff_t rw_range_start;
597 loff_t rw_last_file_pos;
599 size_t rw_smallest_extent;
600 size_t rw_largest_extent;
601 struct ll_file_data *rw_last_file;
604 enum stats_track_type {
605 STATS_TRACK_ALL = 0, /* track all processes */
606 STATS_TRACK_PID, /* track process with this pid */
607 STATS_TRACK_PPID, /* track processes with this ppid */
608 STATS_TRACK_GID, /* track processes with this gid */
612 /* flags for sbi->ll_flags */
613 #define LL_SBI_NOLCK 0x01 /* DLM locking disabled (directio-only) */
614 #define LL_SBI_CHECKSUM 0x02 /* checksum each page as it's written */
615 #define LL_SBI_FLOCK 0x04
616 #define LL_SBI_USER_XATTR 0x08 /* support user xattr */
617 #define LL_SBI_ACL 0x10 /* support ACL */
618 /* LL_SBI_RMT_CLIENT 0x40 remote client */
619 #define LL_SBI_MDS_CAPA 0x80 /* support mds capa, obsolete */
620 #define LL_SBI_OSS_CAPA 0x100 /* support oss capa, obsolete */
621 #define LL_SBI_LOCALFLOCK 0x200 /* Local flocks support by kernel */
622 #define LL_SBI_LRU_RESIZE 0x400 /* lru resize support */
623 #define LL_SBI_LAZYSTATFS 0x800 /* lazystatfs mount option */
624 /* LL_SBI_SOM_PREVIEW 0x1000 SOM preview mount option, obsolete */
625 #define LL_SBI_32BIT_API 0x2000 /* generate 32 bit inodes. */
626 #define LL_SBI_64BIT_HASH 0x4000 /* support 64-bits dir hash/offset */
627 #define LL_SBI_AGL_ENABLED 0x8000 /* enable agl */
628 #define LL_SBI_VERBOSE 0x10000 /* verbose mount/umount */
629 #define LL_SBI_LAYOUT_LOCK 0x20000 /* layout lock support */
630 #define LL_SBI_USER_FID2PATH 0x40000 /* allow fid2path by unprivileged users */
631 #define LL_SBI_XATTR_CACHE 0x80000 /* support for xattr cache */
632 #define LL_SBI_NOROOTSQUASH 0x100000 /* do not apply root squash */
633 #define LL_SBI_ALWAYS_PING 0x200000 /* always ping even if server
635 #define LL_SBI_FAST_READ 0x400000 /* fast read support */
636 #define LL_SBI_FILE_SECCTX 0x800000 /* set file security context at create */
637 /* LL_SBI_PIO 0x1000000 parallel IO support, introduced in
639 #define LL_SBI_TINY_WRITE 0x2000000 /* tiny write support */
640 #define LL_SBI_FILE_HEAT 0x4000000 /* file heat support */
641 #define LL_SBI_TEST_DUMMY_ENCRYPTION 0x8000000 /* test dummy encryption */
642 #define LL_SBI_ENCRYPT 0x10000000 /* client side encryption */
643 #define LL_SBI_FLAGS { \
671 "test_dummy_encryption", \
675 /* This is embedded into llite super-blocks to keep track of connect
676 * flags (capabilities) supported by all imports given mount is
678 struct lustre_client_ocd {
679 /* This is conjunction of connect_flags across all imports
680 * (LOVs) this mount is connected to. This field is updated by
681 * cl_ocd_update() under ->lco_lock. */
683 struct mutex lco_lock;
684 struct obd_export *lco_md_exp;
685 struct obd_export *lco_dt_exp;
689 /* this protects pglist and ra_info. It isn't safe to
690 * grab from interrupt contexts */
692 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
693 spinlock_t ll_process_lock; /* ll_rw_process_info */
694 struct obd_uuid ll_sb_uuid;
695 struct obd_export *ll_md_exp;
696 struct obd_export *ll_dt_exp;
697 struct obd_device *ll_md_obd;
698 struct obd_device *ll_dt_obd;
699 struct dentry *ll_debugfs_entry;
700 struct lu_fid ll_root_fid; /* root object fid */
703 unsigned int ll_xattr_cache_enabled:1,
704 ll_xattr_cache_set:1, /* already set to 0/1 */
705 ll_client_common_fill_super_succeeded:1,
708 struct lustre_client_ocd ll_lco;
710 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
712 /* Used to track "unstable" pages on a client, and maintain a
713 * LRU list of clean pages. An "unstable" page is defined as
714 * any page which is sent to a server as part of a bulk request,
715 * but is uncommitted to stable storage. */
716 struct cl_client_cache *ll_cache;
718 struct lprocfs_stats *ll_ra_stats;
720 struct ll_ra_info ll_ra_info;
721 unsigned int ll_namelen;
722 struct file_operations *ll_fop;
724 struct lu_site *ll_site;
725 struct cl_device *ll_cl;
727 struct ll_rw_extents_info ll_rw_extents_info;
728 int ll_extent_process_count;
729 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
730 unsigned int ll_offset_process_count;
731 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
732 unsigned int ll_rw_offset_entry_count;
733 int ll_stats_track_id;
734 enum stats_track_type ll_stats_track_type;
737 /* metadata stat-ahead */
738 unsigned int ll_sa_running_max;/* max concurrent
739 * statahead instances */
740 unsigned int ll_sa_max; /* max statahead RPCs */
741 atomic_t ll_sa_total; /* statahead thread started
743 atomic_t ll_sa_wrong; /* statahead thread stopped for
745 atomic_t ll_sa_running; /* running statahead thread
747 atomic_t ll_agl_total; /* AGL thread started count */
749 dev_t ll_sdev_orig; /* save s_dev before assign for
752 struct root_squash_info ll_squash;
755 /* st_blksize returned by stat(2), when non-zero */
756 unsigned int ll_stat_blksize;
758 /* maximum relative age of cached statfs results */
759 unsigned int ll_statfs_max_age;
761 struct kset ll_kset; /* sysfs object */
762 struct completion ll_kobj_unregister;
765 unsigned int ll_heat_decay_weight;
766 unsigned int ll_heat_period_second;
768 /* filesystem fsname */
769 char ll_fsname[LUSTRE_MAXFSNAME + 1];
771 /* Persistent Client Cache */
772 struct pcc_super ll_pcc_super;
775 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
776 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
778 * per file-descriptor read-ahead data.
780 struct ll_readahead_state {
782 /* End byte that read(2) try to read. */
783 loff_t ras_last_read_end_bytes;
785 * number of bytes read after last read-ahead window reset. As window
786 * is reset on each seek, this is effectively a number of consecutive
787 * accesses. Maybe ->ras_accessed_in_window is better name.
789 * XXX nikita: window is also reset (by ras_update()) when Lustre
790 * believes that memory pressure evicts read-ahead pages. In that
791 * case, it probably doesn't make sense to expand window to
792 * PTLRPC_MAX_BRW_PAGES on the third access.
794 loff_t ras_consecutive_bytes;
796 * number of read requests after the last read-ahead window reset
797 * As window is reset on each seek, this is effectively the number
798 * on consecutive read request and is used to trigger read-ahead.
800 unsigned long ras_consecutive_requests;
802 * Parameters of current read-ahead window. Handled by
803 * ras_update(). On the initial access to the file or after a seek,
804 * window is reset to 0. After 3 consecutive accesses, window is
805 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
806 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
808 pgoff_t ras_window_start_idx;
809 pgoff_t ras_window_pages;
811 * Optimal RPC size in pages.
812 * It decides how many pages will be sent for each read-ahead.
814 unsigned long ras_rpc_pages;
816 * Where next read-ahead should start at. This lies within read-ahead
817 * window. Read-ahead window is read in pieces rather than at once
818 * because: 1. lustre limits total number of pages under read-ahead by
819 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
820 * not covered by DLM lock.
822 pgoff_t ras_next_readahead_idx;
824 * Total number of ll_file_read requests issued, reads originating
825 * due to mmap are not counted in this total. This value is used to
826 * trigger full file read-ahead after multiple reads to a small file.
828 unsigned long ras_requests;
830 * The following 3 items are used for detecting the stride I/O
832 * In stride I/O mode,
833 * ...............|-----data-----|****gap*****|--------|******|....
834 * offset |-stride_bytes-|-stride_gap-|
835 * ras_stride_offset = offset;
836 * ras_stride_length = stride_bytes + stride_gap;
837 * ras_stride_bytes = stride_bytes;
838 * Note: all these three items are counted by bytes.
840 loff_t ras_stride_offset;
841 loff_t ras_stride_length;
842 loff_t ras_stride_bytes;
844 * number of consecutive stride request count, and it is similar as
845 * ras_consecutive_requests, but used for stride I/O mode.
846 * Note: only more than 2 consecutive stride request are detected,
847 * stride read-ahead will be enable
849 unsigned long ras_consecutive_stride_requests;
850 /* index of the last page that async readahead starts */
851 pgoff_t ras_async_last_readpage_idx;
852 /* whether we should increase readahead window */
853 bool ras_need_increase_window;
854 /* whether ra miss check should be skipped */
855 bool ras_no_miss_check;
858 struct ll_readahead_work {
859 /** File to readahead */
860 struct file *lrw_file;
861 pgoff_t lrw_start_idx;
864 /* async worker to handler read */
865 struct work_struct lrw_readahead_work;
866 char lrw_jobid[LUSTRE_JOBID_SIZE];
869 extern struct kmem_cache *ll_file_data_slab;
870 struct lustre_handle;
871 struct ll_file_data {
872 struct ll_readahead_state fd_ras;
873 struct ll_grouplock fd_grouplock;
877 /* openhandle if lease exists for this file.
878 * Borrow lli->lli_och_mutex to protect assignment */
879 struct obd_client_handle *fd_lease_och;
880 struct obd_client_handle *fd_och;
881 struct file *fd_file;
882 /* Indicate whether need to report failure when close.
883 * true: failure is known, not report again.
884 * false: unknown failure, should report. */
885 bool fd_write_failed;
886 bool ll_lock_no_expand;
887 rwlock_t fd_lock; /* protect lcc list */
888 struct list_head fd_lccs; /* list of ll_cl_context */
889 /* Used by mirrored file to lead IOs to a specific mirror, usually
890 * for mirror resync. 0 means default. */
891 __u32 fd_designated_mirror;
892 /* The layout version when resync starts. Resync I/O should carry this
893 * layout version for verification to OST objects */
894 __u32 fd_layout_version;
895 struct pcc_file fd_pcc_file;
898 void llite_tunables_unregister(void);
899 int llite_tunables_register(void);
901 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
903 return &lli->lli_vfs_inode;
906 __u32 ll_i2suppgid(struct inode *i);
907 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
909 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
911 #if BITS_PER_LONG == 32
913 #elif defined(CONFIG_COMPAT)
914 if (unlikely(sbi->ll_flags & LL_SBI_32BIT_API))
917 # ifdef CONFIG_X86_X32
918 /* in_compat_syscall() returns true when called from a kthread
919 * and CONFIG_X86_X32 is enabled, which is wrong. So check
920 * whether the caller comes from a syscall (ie. not a kthread)
921 * before calling in_compat_syscall(). */
922 if (current->flags & PF_KTHREAD)
926 return unlikely(in_compat_syscall());
928 return unlikely(sbi->ll_flags & LL_SBI_32BIT_API);
932 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
934 return !!(sbi->ll_flags & LL_SBI_FAST_READ);
937 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
939 return !!(sbi->ll_flags & LL_SBI_TINY_WRITE);
942 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
944 return !!(sbi->ll_flags & LL_SBI_FILE_HEAT);
947 void ll_ras_enter(struct file *f, loff_t pos, size_t count);
949 /* llite/lcommon_misc.c */
950 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
951 enum obd_notify_event ev, void *owner);
952 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
953 struct ll_grouplock *lg);
954 void cl_put_grouplock(struct ll_grouplock *lg);
956 /* llite/lproc_llite.c */
957 int ll_debugfs_register_super(struct super_block *sb, const char *name);
958 void ll_debugfs_unregister_super(struct super_block *sb);
959 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
963 LPROC_LL_WRITE_BYTES,
990 LPROC_LL_GETXATTR_HITS,
992 LPROC_LL_REMOVEXATTR,
995 LPROC_LL_FILE_OPCODES
999 enum get_default_layout_type {
1000 GET_DEFAULT_LAYOUT_ROOT = 1,
1003 struct ll_dir_chain {
1006 static inline void ll_dir_chain_init(struct ll_dir_chain *chain)
1010 static inline void ll_dir_chain_fini(struct ll_dir_chain *chain)
1014 extern const struct file_operations ll_dir_operations;
1015 extern const struct inode_operations ll_dir_inode_operations;
1016 #ifdef HAVE_DIR_CONTEXT
1017 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1018 struct dir_context *ctx);
1020 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1021 void *cookie, filldir_t filldir);
1023 int ll_get_mdt_idx(struct inode *inode);
1024 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
1025 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
1026 __u64 offset, struct ll_dir_chain *chain);
1027 void ll_release_page(struct inode *inode, struct page *page, bool remove);
1030 extern const struct inode_operations ll_special_inode_operations;
1032 struct inode *ll_iget(struct super_block *sb, ino_t hash,
1033 struct lustre_md *lic);
1034 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
1035 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1036 void *data, int flag);
1037 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
1038 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
1039 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
1042 int ll_writepage(struct page *page, struct writeback_control *wbc);
1043 int ll_writepages(struct address_space *, struct writeback_control *wbc);
1044 int ll_readpage(struct file *file, struct page *page);
1045 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
1046 struct cl_page *page, struct file *file);
1047 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
1048 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
1051 void ll_cl_add(struct file *file, const struct lu_env *env, struct cl_io *io,
1052 enum lcc_type type);
1053 void ll_cl_remove(struct file *file, const struct lu_env *env);
1054 struct ll_cl_context *ll_cl_find(struct file *file);
1056 extern const struct address_space_operations ll_aops;
1059 extern struct file_operations ll_file_operations;
1060 extern struct file_operations ll_file_operations_flock;
1061 extern struct file_operations ll_file_operations_noflock;
1062 extern struct inode_operations ll_file_inode_operations;
1063 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
1064 enum ldlm_mode l_req_mode);
1065 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1066 struct lustre_handle *lockh, __u64 flags,
1067 enum ldlm_mode mode);
1069 int ll_file_open(struct inode *inode, struct file *file);
1070 int ll_file_release(struct inode *inode, struct file *file);
1071 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1072 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1073 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1074 struct ll_file_data *file, loff_t pos,
1075 size_t count, int rw);
1076 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
1077 int ll_getattr(const struct path *path, struct kstat *stat,
1078 u32 request_mask, unsigned int flags);
1080 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1082 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1083 unsigned int flags);
1084 struct posix_acl *ll_get_acl(struct inode *inode, int type);
1085 #ifdef HAVE_IOP_SET_ACL
1086 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1087 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1088 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1089 #define ll_set_acl NULL
1090 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1094 static inline int ll_xflags_to_inode_flags(int xflags)
1096 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1097 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1098 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1099 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1102 static inline int ll_inode_flags_to_xflags(int flags)
1104 return ((flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1105 ((flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1106 ((flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1107 ((flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1110 int ll_migrate(struct inode *parent, struct file *file,
1111 struct lmv_user_md *lum, const char *name);
1112 int ll_get_fid_by_name(struct inode *parent, const char *name,
1113 int namelen, struct lu_fid *fid, struct inode **inode);
1114 int ll_inode_permission(struct inode *inode, int mask);
1115 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa);
1116 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1118 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1121 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1122 __u64 flags, struct lov_user_md *lum,
1124 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1125 struct lov_mds_md **lmm, int *lmm_size,
1126 struct ptlrpc_request **request);
1127 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1129 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1130 int *lmm_size, struct ptlrpc_request **request,
1131 struct ptlrpc_request **root_request, u64 valid);
1132 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1133 struct ptlrpc_request **request, u64 valid);
1134 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1135 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1136 int ll_fid2path(struct inode *inode, void __user *arg);
1137 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1138 int ll_hsm_release(struct inode *inode);
1139 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1140 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1142 /* llite/dcache.c */
1144 int ll_d_init(struct dentry *de);
1145 extern const struct dentry_operations ll_d_ops;
1146 void ll_intent_drop_lock(struct lookup_intent *);
1147 void ll_intent_release(struct lookup_intent *);
1148 void ll_prune_aliases(struct inode *inode);
1149 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1150 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1151 struct lookup_intent *it, struct dentry *de);
1153 /* llite/llite_lib.c */
1154 extern struct super_operations lustre_super_operations;
1156 void ll_lli_init(struct ll_inode_info *lli);
1157 int ll_fill_super(struct super_block *sb);
1158 void ll_put_super(struct super_block *sb);
1159 void ll_kill_super(struct super_block *sb);
1160 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1161 void ll_dir_clear_lsm_md(struct inode *inode);
1162 void ll_clear_inode(struct inode *inode);
1163 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1164 enum op_xvalid xvalid, bool hsm_import);
1165 int ll_setattr(struct dentry *de, struct iattr *attr);
1166 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1167 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1169 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1170 void ll_update_inode_flags(struct inode *inode, int ext_flags);
1171 int ll_read_inode2(struct inode *inode, void *opaque);
1172 void ll_delete_inode(struct inode *inode);
1173 int ll_iocontrol(struct inode *inode, struct file *file,
1174 unsigned int cmd, unsigned long arg);
1175 int ll_flush_ctx(struct inode *inode);
1176 void ll_umount_begin(struct super_block *sb);
1177 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1178 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1179 void ll_dirty_page_discard_warn(struct page *page, int ioret);
1180 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1181 struct super_block *, struct lookup_intent *);
1182 int ll_obd_statfs(struct inode *inode, void __user *arg);
1183 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1184 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1185 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1187 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1188 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1189 struct inode *i1, struct inode *i2,
1190 const char *name, size_t namelen,
1191 __u32 mode, enum md_op_code opc,
1193 void ll_finish_md_op_data(struct md_op_data *op_data);
1194 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
1195 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1196 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1197 struct lov_user_md **kbuf);
1198 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req);
1200 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req);
1202 /* Compute expected user md size when passing in a md from user space */
1203 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1205 switch (lum->lmm_magic) {
1206 case LOV_USER_MAGIC_V1:
1207 return sizeof(struct lov_user_md_v1);
1208 case LOV_USER_MAGIC_V3:
1209 return sizeof(struct lov_user_md_v3);
1210 case LOV_USER_MAGIC_SPECIFIC:
1211 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1214 return lov_user_md_size(lum->lmm_stripe_count,
1215 LOV_USER_MAGIC_SPECIFIC);
1216 case LOV_USER_MAGIC_COMP_V1:
1217 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1218 case LOV_USER_MAGIC_FOREIGN:
1219 return foreign_size(lum);
1225 /* llite/llite_nfs.c */
1226 extern struct export_operations lustre_export_operations;
1227 __u32 get_uuid2int(const char *name, int len);
1228 struct inode *search_inode_for_lustre(struct super_block *sb,
1229 const struct lu_fid *fid);
1230 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1232 /* llite/symlink.c */
1233 extern struct inode_operations ll_fast_symlink_inode_operations;
1236 * IO arguments for various VFS I/O interfaces.
1238 struct vvp_io_args {
1239 /** normal/sendfile/splice */
1242 struct kiocb *via_iocb;
1243 struct iov_iter *via_iter;
1253 struct ll_cl_context {
1254 struct list_head lcc_list;
1256 const struct lu_env *lcc_env;
1257 struct cl_io *lcc_io;
1258 struct cl_page *lcc_page;
1259 enum lcc_type lcc_type;
1262 struct ll_thread_info {
1263 struct vvp_io_args lti_args;
1264 struct ra_io_arg lti_ria;
1265 struct ll_cl_context lti_io_ctx;
1268 extern struct lu_context_key ll_thread_key;
1270 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1272 struct ll_thread_info *lti;
1274 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1275 LASSERT(lti != NULL);
1280 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env)
1282 return &ll_env_info(env)->lti_args;
1285 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1286 struct vvp_io_args *args);
1288 /* llite/llite_mmap.c */
1290 int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last);
1291 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1292 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1293 unsigned long addr, size_t count);
1294 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1297 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1299 /* don't need an addref as the sb_info should be holding one */
1300 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1302 return ll_s2sbi(sb)->ll_dt_exp;
1305 /* don't need an addref as the sb_info should be holding one */
1306 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1308 return ll_s2sbi(sb)->ll_md_exp;
1311 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1313 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1319 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1320 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1322 return ll_s2sbi(inode->i_sb);
1325 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1327 return ll_s2dtexp(inode->i_sb);
1330 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1332 return ll_s2mdexp(inode->i_sb);
1335 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1339 LASSERT(inode != NULL);
1340 fid = &ll_i2info(inode)->lli_fid;
1345 static inline bool ll_dir_striped(struct inode *inode)
1348 return S_ISDIR(inode->i_mode) &&
1349 lmv_dir_striped(ll_i2info(inode)->lli_lsm_md);
1352 static inline loff_t ll_file_maxbytes(struct inode *inode)
1354 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1357 return MAX_LFS_FILESIZE;
1359 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1363 extern const struct xattr_handler *ll_xattr_handlers[];
1365 #define XATTR_USER_T 1
1366 #define XATTR_TRUSTED_T 2
1367 #define XATTR_SECURITY_T 3
1368 #define XATTR_ACL_ACCESS_T 4
1369 #define XATTR_ACL_DEFAULT_T 5
1370 #define XATTR_LUSTRE_T 6
1371 #define XATTR_OTHER_T 7
1373 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1374 int ll_xattr_list(struct inode *inode, const char *name, int type,
1375 void *buffer, size_t size, u64 valid);
1376 const struct xattr_handler *get_xattr_type(const char *name);
1379 * Common IO arguments for various VFS I/O interfaces.
1381 int cl_sb_init(struct super_block *sb);
1382 int cl_sb_fini(struct super_block *sb);
1384 enum ras_update_flags {
1388 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1389 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1393 #define LL_SA_RPC_MIN 2
1394 #define LL_SA_RPC_DEF 32
1395 #define LL_SA_RPC_MAX 512
1397 /* XXX: If want to support more concurrent statahead instances,
1398 * please consider to decentralize the RPC lists attached
1399 * on related import, such as imp_{sending,delayed}_list.
1401 #define LL_SA_RUNNING_MAX 256
1402 #define LL_SA_RUNNING_DEF 16
1404 #define LL_SA_CACHE_BIT 5
1405 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1406 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1408 /* per inode struct, for dir only */
1409 struct ll_statahead_info {
1410 struct dentry *sai_dentry;
1411 atomic_t sai_refcount; /* when access this struct, hold
1413 unsigned int sai_max; /* max ahead of lookup */
1414 __u64 sai_sent; /* stat requests sent count */
1415 __u64 sai_replied; /* stat requests which received
1417 __u64 sai_index; /* index of statahead entry */
1418 __u64 sai_index_wait; /* index of entry which is the
1419 * caller is waiting for */
1420 __u64 sai_hit; /* hit count */
1421 __u64 sai_miss; /* miss count:
1422 * for "ls -al" case, includes
1423 * hidden dentry miss;
1424 * for "ls -l" case, it does not
1425 * include hidden dentry miss.
1426 * "sai_miss_hidden" is used for
1429 unsigned int sai_consecutive_miss; /* consecutive miss */
1430 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1431 * is not a hidden one */
1432 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1434 unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
1436 sai_agl_valid:1,/* AGL is valid for the dir */
1437 sai_in_readpage:1;/* statahead is in readdir()*/
1438 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1439 struct task_struct *sai_task; /* stat-ahead thread */
1440 struct task_struct *sai_agl_task; /* AGL thread */
1441 struct list_head sai_interim_entries; /* entries which got async
1442 * stat reply, but not
1444 struct list_head sai_entries; /* completed entries */
1445 struct list_head sai_agls; /* AGLs to be sent */
1446 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1447 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1448 atomic_t sai_cache_count; /* entry count in cache */
1451 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1453 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1454 void ll_authorize_statahead(struct inode *dir, void *key);
1455 void ll_deauthorize_statahead(struct inode *dir, void *key);
1458 blkcnt_t dirty_cnt(struct inode *inode);
1460 int cl_glimpse_size0(struct inode *inode, int agl);
1461 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1462 struct inode *inode, struct cl_object *clob, int agl);
1464 static inline int cl_glimpse_size(struct inode *inode)
1466 return cl_glimpse_size0(inode, 0);
1469 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1470 * part of statahead */
1471 static inline int cl_agl(struct inode *inode)
1473 return cl_glimpse_size0(inode, 1);
1476 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1478 int cl_io_get(struct inode *inode, struct lu_env **envout,
1479 struct cl_io **ioout, __u16 *refcheck);
1481 static inline int ll_glimpse_size(struct inode *inode)
1483 struct ll_inode_info *lli = ll_i2info(inode);
1486 down_read(&lli->lli_glimpse_sem);
1487 rc = cl_glimpse_size(inode);
1488 lli->lli_glimpse_time = ktime_get();
1489 up_read(&lli->lli_glimpse_sem);
1493 /* dentry may statahead when statahead is enabled and current process has opened
1494 * parent directory, and this dentry hasn't accessed statahead cache before */
1496 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1498 struct ll_inode_info *lli;
1499 struct ll_dentry_data *ldd;
1501 if (ll_i2sbi(dir)->ll_sa_max == 0)
1504 lli = ll_i2info(dir);
1506 /* statahead is not allowed for this dir, there may be three causes:
1507 * 1. dir is not opened.
1508 * 2. statahead hit ratio is too low.
1509 * 3. previous stat started statahead thread failed. */
1510 if (!lli->lli_sa_enabled)
1513 /* not the same process, don't statahead */
1514 if (lli->lli_opendir_pid != current->pid)
1518 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1519 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1520 * For patchless client, lookup intent is not accurate, which may
1521 * misguide statahead. For example:
1522 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1523 * have the same intent -- IT_GETATTR, while one dentry should access
1524 * statahead cache once, otherwise statahead windows is messed up.
1525 * The solution is as following:
1526 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1527 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1528 * bypass interacting with statahead cache by checking
1529 * 'lld_sa_generation == lli->lli_sa_generation'.
1531 ldd = ll_d2d(dentry);
1532 if (ldd != NULL && lli->lli_sa_generation &&
1533 ldd->lld_sa_generation == lli->lli_sa_generation)
1539 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1540 enum cl_fsync_mode mode, int ignore_layout);
1542 static inline int ll_file_nolock(const struct file *file)
1544 struct ll_file_data *fd = file->private_data;
1545 struct inode *inode = file_inode((struct file *)file);
1547 LASSERT(fd != NULL);
1548 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1549 (ll_i2sbi(inode)->ll_flags & LL_SBI_NOLCK));
1552 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1553 struct lookup_intent *it, __u64 *bits)
1555 if (!it->it_lock_set) {
1556 struct lustre_handle handle;
1558 /* If this inode is a remote object, it will get two
1559 * separate locks in different namespaces, Master MDT,
1560 * where the name entry is, will grant LOOKUP lock,
1561 * remote MDT, where the object is, will grant
1562 * UPDATE|PERM lock. The inode will be attched to both
1563 * LOOKUP and PERM locks, so revoking either locks will
1564 * case the dcache being cleared */
1565 if (it->it_remote_lock_mode) {
1566 handle.cookie = it->it_remote_lock_handle;
1567 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1568 "(%p) for remote lock %#llx\n",
1569 PFID(ll_inode2fid(inode)), inode,
1571 md_set_lock_data(exp, &handle, inode, NULL);
1574 handle.cookie = it->it_lock_handle;
1576 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1577 " for lock %#llx\n",
1578 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1580 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1581 it->it_lock_set = 1;
1585 *bits = it->it_lock_bits;
1588 static inline int d_lustre_invalid(const struct dentry *dentry)
1590 struct ll_dentry_data *lld = ll_d2d(dentry);
1592 return (lld == NULL) || lld->lld_invalid;
1595 static inline void __d_lustre_invalidate(struct dentry *dentry)
1597 struct ll_dentry_data *lld = ll_d2d(dentry);
1600 lld->lld_invalid = 1;
1604 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1605 * ll_md_blocking_ast), it will be pruned by ll_prune_aliases() and
1606 * ll_prune_negative_children(); otherwise dput() of the last refcount will
1607 * unhash this dentry and kill it.
1609 static inline void d_lustre_invalidate(struct dentry *dentry)
1611 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1613 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1615 spin_lock(&dentry->d_lock);
1616 __d_lustre_invalidate(dentry);
1617 spin_unlock(&dentry->d_lock);
1620 static inline void d_lustre_revalidate(struct dentry *dentry)
1622 spin_lock(&dentry->d_lock);
1623 LASSERT(ll_d2d(dentry) != NULL);
1624 ll_d2d(dentry)->lld_invalid = 0;
1625 spin_unlock(&dentry->d_lock);
1628 static inline dev_t ll_compat_encode_dev(dev_t dev)
1630 /* The compat_sys_*stat*() syscalls will fail unless the
1631 * device majors and minors are both less than 256. Note that
1632 * the value returned here will be passed through
1633 * old_encode_dev() in cp_compat_stat(). And so we are not
1634 * trying to return a valid compat (u16) device number, just
1635 * one that will pass the old_valid_dev() check. */
1637 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1640 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1641 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1642 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1643 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1644 struct lu_extent *ext);
1646 int ll_xattr_init(void);
1647 void ll_xattr_fini(void);
1649 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1650 struct cl_page *page, enum cl_req_type crt);
1652 int ll_getparent(struct file *file, struct getparent __user *arg);
1655 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1656 enum op_xvalid xvalid, unsigned int attr_flags);
1658 extern struct lu_env *cl_inode_fini_env;
1659 extern __u16 cl_inode_fini_refcheck;
1661 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1662 void cl_inode_fini(struct inode *inode);
1664 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1665 u32 cl_fid_build_gen(const struct lu_fid *fid);
1667 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1669 return &ll_i2sbi(inode)->ll_pcc_super;
1672 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1674 return ll_i2pccs(ll_info2i(lli));
1677 #ifdef HAVE_LUSTRE_CRYPTO
1679 extern const struct llcrypt_operations lustre_cryptops;
1682 #endif /* LLITE_INTERNAL_H */