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 #ifndef VM_FAULT_RETRY
51 #include <linux/mm_types.h>
54 #include "vvp_internal.h"
55 #include "range_lock.h"
62 #ifndef VM_FAULT_RETRY
63 #define VM_FAULT_RETRY 0
66 /* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
67 * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
68 #ifndef LOOKUP_CONTINUE
69 #define LOOKUP_CONTINUE LOOKUP_PARENT
72 /** Only used on client-side for indicating the tail of dir hash/offset. */
73 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
74 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
76 /* 4UL * 1024 * 1024 */
77 #define LL_MAX_BLKSIZE_BITS 22
79 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
81 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
83 struct ll_dentry_data {
84 unsigned int lld_sa_generation;
85 unsigned int lld_invalid:1;
86 unsigned int lld_nfs_dentry:1;
87 struct rcu_head lld_rcu_head;
90 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
92 #define LLI_INODE_MAGIC 0x111d0de5
93 #define LLI_INODE_DEAD 0xdeadd00d
95 struct ll_getname_data {
96 #ifdef HAVE_DIR_CONTEXT
97 struct dir_context ctx;
99 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
100 struct lu_fid lgd_fid; /* target fid we are looking for */
101 int lgd_found; /* inode matched? */
104 struct ll_grouplock {
105 struct lu_env *lg_env;
107 struct cl_lock *lg_lock;
108 unsigned long lg_gid;
111 /* See comment on trunc_sem_down_read_nowait */
112 struct ll_trunc_sem {
113 /* when positive, this is a count of readers, when -1, it indicates
114 * the semaphore is held for write, and 0 is unlocked
116 atomic_t ll_trunc_readers;
117 /* this tracks a count of waiting writers */
118 atomic_t ll_trunc_waiters;
121 struct ll_inode_info {
122 __u32 lli_inode_magic;
125 volatile unsigned long lli_flags;
126 struct posix_acl *lli_posix_acl;
128 /* identifying fields for both metadata and data stacks. */
129 struct lu_fid lli_fid;
130 /* master inode fid for stripe directory */
131 struct lu_fid lli_pfid;
133 /* We need all three because every inode may be opened in different
135 struct obd_client_handle *lli_mds_read_och;
136 struct obd_client_handle *lli_mds_write_och;
137 struct obd_client_handle *lli_mds_exec_och;
138 __u64 lli_open_fd_read_count;
139 __u64 lli_open_fd_write_count;
140 __u64 lli_open_fd_exec_count;
141 /* Protects access to och pointers and their usage counters */
142 struct mutex lli_och_mutex;
144 struct inode lli_vfs_inode;
146 /* the most recent timestamps obtained from mds */
151 spinlock_t lli_agl_lock;
153 /* Try to make the d::member and f::member are aligned. Before using
154 * these members, make clear whether it is directory or not. */
158 /* metadata statahead */
159 /* since parent-child threads can share the same @file
160 * struct, "opendir_key" is the token when dir close for
161 * case of parent exit before child -- it is me should
162 * cleanup the dir readahead. */
163 void *lli_opendir_key;
164 struct ll_statahead_info *lli_sai;
165 /* protect statahead stuff. */
166 spinlock_t lli_sa_lock;
167 /* "opendir_pid" is the token when lookup/revalid
168 * -- I am the owner of dir statahead. */
169 pid_t lli_opendir_pid;
170 /* stat will try to access statahead entries or start
171 * statahead if this flag is set, and this flag will be
172 * set upon dir open, and cleared when dir is closed,
173 * statahead hit ratio is too low, or start statahead
175 unsigned int lli_sa_enabled:1;
176 /* generation for statahead */
177 unsigned int lli_sa_generation;
178 /* rw lock protects lli_lsm_md */
179 struct rw_semaphore lli_lsm_sem;
180 /* directory stripe information */
181 struct lmv_stripe_md *lli_lsm_md;
182 /* directory default LMV */
183 struct lmv_stripe_md *lli_default_lsm_md;
186 /* for non-directory */
188 struct mutex lli_size_mutex;
189 char *lli_symlink_name;
190 struct ll_trunc_sem lli_trunc_sem;
191 struct range_lock_tree lli_write_tree;
192 struct mutex lli_setattr_mutex;
194 struct rw_semaphore lli_glimpse_sem;
195 ktime_t lli_glimpse_time;
196 struct list_head lli_agl_list;
199 /* for writepage() only to communicate to fsync */
202 /* protect the file heat fields */
203 spinlock_t lli_heat_lock;
204 __u32 lli_heat_flags;
205 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
208 * Whenever a process try to read/write the file, the
209 * jobid of the process will be saved here, and it'll
210 * be packed into the write PRC when flush later.
212 * So the read/write statistics for jobid will not be
213 * accurate if the file is shared by different jobs.
215 char lli_jobid[LUSTRE_JOBID_SIZE];
217 struct mutex lli_pcc_lock;
218 enum lu_pcc_state_flags lli_pcc_state;
220 * @lli_pcc_generation saves the gobal PCC generation
221 * when the file was successfully attached into PCC.
222 * The flags of the PCC dataset are saved in
224 * The gobal PCC generation will be increased when add
225 * or delete a PCC backend, or change the configuration
226 * parameters for PCC.
227 * If @lli_pcc_generation is same as the gobal PCC
228 * generation, we can use the saved flags of the PCC
229 * dataset to determine whether need to try auto attach
232 __u64 lli_pcc_generation;
233 enum pcc_dataset_flags lli_pcc_dsflags;
234 struct pcc_inode *lli_pcc_inode;
236 struct mutex lli_group_mutex;
237 __u64 lli_group_users;
238 unsigned long lli_group_gid;
240 __u64 lli_attr_valid;
242 __u64 lli_lazyblocks;
246 /* XXX: For following frequent used members, although they maybe special
247 * used for non-directory object, it is some time-wasting to check
248 * whether the object is directory or not before using them. On the
249 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
250 * the "ll_inode_info" size even if moving those members into u.f.
251 * So keep them out side.
253 * In the future, if more members are added only for directory,
254 * some of the following members can be moved into u.f.
256 struct cl_object *lli_clob;
258 /* mutex to request for layout lock exclusively. */
259 struct mutex lli_layout_mutex;
260 /* Layout version, protected by lli_layout_lock */
261 __u32 lli_layout_gen;
262 spinlock_t lli_layout_lock;
264 __u32 lli_projid; /* project id */
266 struct rw_semaphore lli_xattrs_list_rwsem;
267 struct mutex lli_xattrs_enq_lock;
268 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
271 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
273 atomic_set(&sem->ll_trunc_readers, 0);
274 atomic_set(&sem->ll_trunc_waiters, 0);
277 /* This version of down read ignores waiting writers, meaning if the semaphore
278 * is already held for read, this down_read will 'join' that reader and also
279 * take the semaphore.
281 * This lets us avoid an unusual deadlock.
283 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
284 * in Lustre, in order to exclude truncates. Some of these paths then need to
285 * take the mmap_sem, while still holding the trunc_sem. The problem is that
286 * page faults hold the mmap_sem when calling in to Lustre, and then must also
287 * take the trunc_sem to exclude truncate.
289 * This means the locking order for trunc_sem and mmap_sem is sometimes AB,
290 * sometimes BA. This is almost OK because in both cases, we take the trunc
291 * sem for read, so it doesn't block.
293 * However, if a write mode user (truncate, a setattr op) arrives in the
294 * middle of this, the second reader on the truncate_sem will wait behind that
297 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
298 * to the mode in which they take the semaphore):
299 * reader (holding mmap_sem, needs truncate_sem)
301 * reader (holding truncate sem, waiting for mmap_sem)
303 * And so the readers deadlock.
305 * The solution is this modified semaphore, where this down_read ignores
306 * waiting write operations, and all waiters are woken up at once, so readers
307 * using down_read_nowait cannot get stuck behind waiting writers, regardless
308 * of the order they arrived in.
310 * down_read_nowait is only used in the page fault case, where we already hold
311 * the mmap_sem. This is because otherwise repeated read and write operations
312 * (which take the truncate sem) could prevent a truncate from ever starting.
313 * This could still happen with page faults, but without an even more complex
314 * mechanism, this is unavoidable.
318 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
320 wait_var_event(&sem->ll_trunc_readers,
321 atomic_inc_unless_negative(&sem->ll_trunc_readers));
324 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
326 wait_var_event(&sem->ll_trunc_readers,
327 atomic_read(&sem->ll_trunc_waiters) == 0 &&
328 atomic_inc_unless_negative(&sem->ll_trunc_readers));
331 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
333 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
334 atomic_read(&sem->ll_trunc_waiters))
335 wake_up_var(&sem->ll_trunc_readers);
338 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
340 atomic_inc(&sem->ll_trunc_waiters);
341 wait_var_event(&sem->ll_trunc_readers,
342 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
343 atomic_dec(&sem->ll_trunc_waiters);
346 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
348 atomic_set(&sem->ll_trunc_readers, 0);
349 wake_up_var(&sem->ll_trunc_readers);
352 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
356 spin_lock(&lli->lli_layout_lock);
357 gen = lli->lli_layout_gen;
358 spin_unlock(&lli->lli_layout_lock);
363 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
365 spin_lock(&lli->lli_layout_lock);
366 lli->lli_layout_gen = gen;
367 spin_unlock(&lli->lli_layout_lock);
371 /* File data is modified. */
372 LLIF_DATA_MODIFIED = 0,
373 /* File is being restored */
374 LLIF_FILE_RESTORING = 1,
375 /* Xattr cache is attached to the file */
376 LLIF_XATTR_CACHE = 2,
377 /* Project inherit */
378 LLIF_PROJECT_INHERIT = 3,
379 /* update atime from MDS even if it's older than local inode atime. */
380 LLIF_UPDATE_ATIME = 4,
384 static inline void ll_file_set_flag(struct ll_inode_info *lli,
385 enum ll_file_flags flag)
387 set_bit(flag, &lli->lli_flags);
390 static inline void ll_file_clear_flag(struct ll_inode_info *lli,
391 enum ll_file_flags flag)
393 clear_bit(flag, &lli->lli_flags);
396 static inline bool ll_file_test_flag(struct ll_inode_info *lli,
397 enum ll_file_flags flag)
399 return test_bit(flag, &lli->lli_flags);
402 static inline bool ll_file_test_and_clear_flag(struct ll_inode_info *lli,
403 enum ll_file_flags flag)
405 return test_and_clear_bit(flag, &lli->lli_flags);
408 int ll_xattr_cache_destroy(struct inode *inode);
410 int ll_xattr_cache_get(struct inode *inode,
416 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
418 #ifdef CONFIG_SECURITY
419 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
420 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
426 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
428 #ifdef CONFIG_SECURITY
429 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
433 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
434 const char **secctx_name, void **secctx,
436 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
439 int ll_listsecurity(struct inode *inode, char *secctx_name,
440 size_t secctx_name_size);
442 static inline bool obd_connect_has_enc(struct obd_connect_data *data)
444 #ifdef HAVE_LUSTRE_CRYPTO
445 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
446 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT;
452 static inline void obd_connect_set_enc(struct obd_connect_data *data)
454 #ifdef HAVE_LUSTRE_CRYPTO
455 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT;
460 * Locking to guarantee consistency of non-atomic updates to long long i_size,
461 * consistency between file size and KMS.
463 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
466 void ll_inode_size_lock(struct inode *inode);
467 void ll_inode_size_unlock(struct inode *inode);
469 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
471 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
474 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
476 return ll_i2info(inode)->lli_pcc_inode;
479 /* default to use at least 16M for fast read if possible */
480 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
482 /* default readahead on a given system. */
483 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(64UL)
485 /* default read-ahead full files smaller than limit on the second read */
486 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
491 RA_STAT_DISTANT_READPAGE,
492 RA_STAT_MISS_IN_WINDOW,
493 RA_STAT_FAILED_GRAB_PAGE,
494 RA_STAT_FAILED_MATCH,
499 RA_STAT_MAX_IN_FLIGHT,
500 RA_STAT_WRONG_GRAB_PAGE,
501 RA_STAT_FAILED_REACH_END,
503 RA_STAT_FAILED_FAST_READ,
508 atomic_t ra_cur_pages;
509 unsigned long ra_max_pages;
510 unsigned long ra_max_pages_per_file;
511 unsigned long ra_max_read_ahead_whole_pages;
512 struct workqueue_struct *ll_readahead_wq;
514 * Max number of active works could be triggered
515 * for async readahead.
517 unsigned int ra_async_max_active;
518 /* how many async readahead triggered in flight */
519 atomic_t ra_async_inflight;
520 /* Threshold to control when to trigger async readahead */
521 unsigned long ra_async_pages_per_file_threshold;
524 /* ra_io_arg will be filled in the beginning of ll_readahead with
525 * ras_lock, then the following ll_read_ahead_pages will read RA
526 * pages according to this arg, all the items in this structure are
527 * counted by page index.
530 pgoff_t ria_start_idx; /* start offset of read-ahead*/
531 pgoff_t ria_end_idx; /* end offset of read-ahead*/
532 unsigned long ria_reserved; /* reserved pages for read-ahead */
533 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
534 bool ria_eof; /* reach end of file */
535 /* If stride read pattern is detected, ria_stoff is the byte offset
536 * where stride read is started. Note: for normal read-ahead, the
537 * value here is meaningless, and also it will not be accessed*/
539 /* ria_length and ria_bytes are the length and pages length in the
540 * stride I/O mode. And they will also be used to check whether
541 * it is stride I/O read-ahead in the read-ahead pages*/
546 /* LL_HIST_MAX=32 causes an overflow */
547 #define LL_HIST_MAX 28
548 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
549 #define LL_PROCESS_HIST_MAX 10
550 struct per_process_info {
552 struct obd_histogram pp_r_hist;
553 struct obd_histogram pp_w_hist;
556 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
557 struct ll_rw_extents_info {
558 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
561 #define LL_OFFSET_HIST_MAX 100
562 struct ll_rw_process_info {
565 loff_t rw_range_start;
567 loff_t rw_last_file_pos;
569 size_t rw_smallest_extent;
570 size_t rw_largest_extent;
571 struct ll_file_data *rw_last_file;
574 enum stats_track_type {
575 STATS_TRACK_ALL = 0, /* track all processes */
576 STATS_TRACK_PID, /* track process with this pid */
577 STATS_TRACK_PPID, /* track processes with this ppid */
578 STATS_TRACK_GID, /* track processes with this gid */
582 /* flags for sbi->ll_flags */
583 #define LL_SBI_NOLCK 0x01 /* DLM locking disabled (directio-only) */
584 #define LL_SBI_CHECKSUM 0x02 /* checksum each page as it's written */
585 #define LL_SBI_FLOCK 0x04
586 #define LL_SBI_USER_XATTR 0x08 /* support user xattr */
587 #define LL_SBI_ACL 0x10 /* support ACL */
588 /* LL_SBI_RMT_CLIENT 0x40 remote client */
589 #define LL_SBI_MDS_CAPA 0x80 /* support mds capa, obsolete */
590 #define LL_SBI_OSS_CAPA 0x100 /* support oss capa, obsolete */
591 #define LL_SBI_LOCALFLOCK 0x200 /* Local flocks support by kernel */
592 #define LL_SBI_LRU_RESIZE 0x400 /* lru resize support */
593 #define LL_SBI_LAZYSTATFS 0x800 /* lazystatfs mount option */
594 /* LL_SBI_SOM_PREVIEW 0x1000 SOM preview mount option, obsolete */
595 #define LL_SBI_32BIT_API 0x2000 /* generate 32 bit inodes. */
596 #define LL_SBI_64BIT_HASH 0x4000 /* support 64-bits dir hash/offset */
597 #define LL_SBI_AGL_ENABLED 0x8000 /* enable agl */
598 #define LL_SBI_VERBOSE 0x10000 /* verbose mount/umount */
599 #define LL_SBI_LAYOUT_LOCK 0x20000 /* layout lock support */
600 #define LL_SBI_USER_FID2PATH 0x40000 /* allow fid2path by unprivileged users */
601 #define LL_SBI_XATTR_CACHE 0x80000 /* support for xattr cache */
602 #define LL_SBI_NOROOTSQUASH 0x100000 /* do not apply root squash */
603 #define LL_SBI_ALWAYS_PING 0x200000 /* always ping even if server
605 #define LL_SBI_FAST_READ 0x400000 /* fast read support */
606 #define LL_SBI_FILE_SECCTX 0x800000 /* set file security context at create */
607 /* LL_SBI_PIO 0x1000000 parallel IO support, introduced in
609 #define LL_SBI_TINY_WRITE 0x2000000 /* tiny write support */
610 #define LL_SBI_FILE_HEAT 0x4000000 /* file heat support */
611 #define LL_SBI_TEST_DUMMY_ENCRYPTION 0x8000000 /* test dummy encryption */
612 #define LL_SBI_ENCRYPT 0x10000000 /* client side encryption */
613 #define LL_SBI_FLAGS { \
641 "test_dummy_encryption", \
645 /* This is embedded into llite super-blocks to keep track of connect
646 * flags (capabilities) supported by all imports given mount is
648 struct lustre_client_ocd {
649 /* This is conjunction of connect_flags across all imports
650 * (LOVs) this mount is connected to. This field is updated by
651 * cl_ocd_update() under ->lco_lock. */
653 struct mutex lco_lock;
654 struct obd_export *lco_md_exp;
655 struct obd_export *lco_dt_exp;
659 /* this protects pglist and ra_info. It isn't safe to
660 * grab from interrupt contexts */
662 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
663 spinlock_t ll_process_lock; /* ll_rw_process_info */
664 struct obd_uuid ll_sb_uuid;
665 struct obd_export *ll_md_exp;
666 struct obd_export *ll_dt_exp;
667 struct obd_device *ll_md_obd;
668 struct obd_device *ll_dt_obd;
669 struct dentry *ll_debugfs_entry;
670 struct lu_fid ll_root_fid; /* root object fid */
673 unsigned int ll_xattr_cache_enabled:1,
674 ll_xattr_cache_set:1, /* already set to 0/1 */
675 ll_client_common_fill_super_succeeded:1,
678 struct lustre_client_ocd ll_lco;
680 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
682 /* Used to track "unstable" pages on a client, and maintain a
683 * LRU list of clean pages. An "unstable" page is defined as
684 * any page which is sent to a server as part of a bulk request,
685 * but is uncommitted to stable storage. */
686 struct cl_client_cache *ll_cache;
688 struct lprocfs_stats *ll_ra_stats;
690 struct ll_ra_info ll_ra_info;
691 unsigned int ll_namelen;
692 struct file_operations *ll_fop;
694 struct lu_site *ll_site;
695 struct cl_device *ll_cl;
697 struct ll_rw_extents_info ll_rw_extents_info;
698 int ll_extent_process_count;
699 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
700 unsigned int ll_offset_process_count;
701 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
702 unsigned int ll_rw_offset_entry_count;
703 int ll_stats_track_id;
704 enum stats_track_type ll_stats_track_type;
707 /* metadata stat-ahead */
708 unsigned int ll_sa_running_max;/* max concurrent
709 * statahead instances */
710 unsigned int ll_sa_max; /* max statahead RPCs */
711 atomic_t ll_sa_total; /* statahead thread started
713 atomic_t ll_sa_wrong; /* statahead thread stopped for
715 atomic_t ll_sa_running; /* running statahead thread
717 atomic_t ll_agl_total; /* AGL thread started count */
719 dev_t ll_sdev_orig; /* save s_dev before assign for
722 struct root_squash_info ll_squash;
725 /* st_blksize returned by stat(2), when non-zero */
726 unsigned int ll_stat_blksize;
728 /* maximum relative age of cached statfs results */
729 unsigned int ll_statfs_max_age;
731 struct kset ll_kset; /* sysfs object */
732 struct completion ll_kobj_unregister;
735 unsigned int ll_heat_decay_weight;
736 unsigned int ll_heat_period_second;
738 /* filesystem fsname */
739 char ll_fsname[LUSTRE_MAXFSNAME + 1];
741 /* Persistent Client Cache */
742 struct pcc_super ll_pcc_super;
745 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
746 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
748 * per file-descriptor read-ahead data.
750 struct ll_readahead_state {
752 /* End byte that read(2) try to read. */
753 loff_t ras_last_read_end_bytes;
755 * number of bytes read after last read-ahead window reset. As window
756 * is reset on each seek, this is effectively a number of consecutive
757 * accesses. Maybe ->ras_accessed_in_window is better name.
759 * XXX nikita: window is also reset (by ras_update()) when Lustre
760 * believes that memory pressure evicts read-ahead pages. In that
761 * case, it probably doesn't make sense to expand window to
762 * PTLRPC_MAX_BRW_PAGES on the third access.
764 loff_t ras_consecutive_bytes;
766 * number of read requests after the last read-ahead window reset
767 * As window is reset on each seek, this is effectively the number
768 * on consecutive read request and is used to trigger read-ahead.
770 unsigned long ras_consecutive_requests;
772 * Parameters of current read-ahead window. Handled by
773 * ras_update(). On the initial access to the file or after a seek,
774 * window is reset to 0. After 3 consecutive accesses, window is
775 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
776 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
778 pgoff_t ras_window_start_idx;
779 pgoff_t ras_window_pages;
781 * Optimal RPC size in pages.
782 * It decides how many pages will be sent for each read-ahead.
784 unsigned long ras_rpc_pages;
786 * Where next read-ahead should start at. This lies within read-ahead
787 * window. Read-ahead window is read in pieces rather than at once
788 * because: 1. lustre limits total number of pages under read-ahead by
789 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
790 * not covered by DLM lock.
792 pgoff_t ras_next_readahead_idx;
794 * Total number of ll_file_read requests issued, reads originating
795 * due to mmap are not counted in this total. This value is used to
796 * trigger full file read-ahead after multiple reads to a small file.
798 unsigned long ras_requests;
800 * The following 3 items are used for detecting the stride I/O
802 * In stride I/O mode,
803 * ...............|-----data-----|****gap*****|--------|******|....
804 * offset |-stride_bytes-|-stride_gap-|
805 * ras_stride_offset = offset;
806 * ras_stride_length = stride_bytes + stride_gap;
807 * ras_stride_bytes = stride_bytes;
808 * Note: all these three items are counted by bytes.
810 loff_t ras_stride_offset;
811 loff_t ras_stride_length;
812 loff_t ras_stride_bytes;
814 * number of consecutive stride request count, and it is similar as
815 * ras_consecutive_requests, but used for stride I/O mode.
816 * Note: only more than 2 consecutive stride request are detected,
817 * stride read-ahead will be enable
819 unsigned long ras_consecutive_stride_requests;
820 /* index of the last page that async readahead starts */
821 pgoff_t ras_async_last_readpage_idx;
822 /* whether we should increase readahead window */
823 bool ras_need_increase_window;
824 /* whether ra miss check should be skipped */
825 bool ras_no_miss_check;
828 struct ll_readahead_work {
829 /** File to readahead */
830 struct file *lrw_file;
831 pgoff_t lrw_start_idx;
834 /* async worker to handler read */
835 struct work_struct lrw_readahead_work;
836 char lrw_jobid[LUSTRE_JOBID_SIZE];
839 extern struct kmem_cache *ll_file_data_slab;
840 struct lustre_handle;
841 struct ll_file_data {
842 struct ll_readahead_state fd_ras;
843 struct ll_grouplock fd_grouplock;
847 /* openhandle if lease exists for this file.
848 * Borrow lli->lli_och_mutex to protect assignment */
849 struct obd_client_handle *fd_lease_och;
850 struct obd_client_handle *fd_och;
851 struct file *fd_file;
852 /* Indicate whether need to report failure when close.
853 * true: failure is known, not report again.
854 * false: unknown failure, should report. */
855 bool fd_write_failed;
856 bool ll_lock_no_expand;
857 rwlock_t fd_lock; /* protect lcc list */
858 struct list_head fd_lccs; /* list of ll_cl_context */
859 /* Used by mirrored file to lead IOs to a specific mirror, usually
860 * for mirror resync. 0 means default. */
861 __u32 fd_designated_mirror;
862 /* The layout version when resync starts. Resync I/O should carry this
863 * layout version for verification to OST objects */
864 __u32 fd_layout_version;
865 struct pcc_file fd_pcc_file;
868 void llite_tunables_unregister(void);
869 int llite_tunables_register(void);
871 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
873 return &lli->lli_vfs_inode;
876 __u32 ll_i2suppgid(struct inode *i);
877 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
879 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
881 #if BITS_PER_LONG == 32
883 #elif defined(CONFIG_COMPAT)
884 if (unlikely(sbi->ll_flags & LL_SBI_32BIT_API))
887 # ifdef CONFIG_X86_X32
888 /* in_compat_syscall() returns true when called from a kthread
889 * and CONFIG_X86_X32 is enabled, which is wrong. So check
890 * whether the caller comes from a syscall (ie. not a kthread)
891 * before calling in_compat_syscall(). */
892 if (current->flags & PF_KTHREAD)
896 return unlikely(in_compat_syscall());
898 return unlikely(sbi->ll_flags & LL_SBI_32BIT_API);
902 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
904 return !!(sbi->ll_flags & LL_SBI_FAST_READ);
907 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
909 return !!(sbi->ll_flags & LL_SBI_TINY_WRITE);
912 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
914 return !!(sbi->ll_flags & LL_SBI_FILE_HEAT);
917 void ll_ras_enter(struct file *f, loff_t pos, size_t count);
919 /* llite/lcommon_misc.c */
920 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
921 enum obd_notify_event ev, void *owner);
922 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
923 struct ll_grouplock *lg);
924 void cl_put_grouplock(struct ll_grouplock *lg);
926 /* llite/lproc_llite.c */
927 int ll_debugfs_register_super(struct super_block *sb, const char *name);
928 void ll_debugfs_unregister_super(struct super_block *sb);
929 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
933 LPROC_LL_WRITE_BYTES,
960 LPROC_LL_GETXATTR_HITS,
962 LPROC_LL_REMOVEXATTR,
965 LPROC_LL_FILE_OPCODES
969 enum get_default_layout_type {
970 GET_DEFAULT_LAYOUT_ROOT = 1,
973 struct ll_dir_chain {
976 static inline void ll_dir_chain_init(struct ll_dir_chain *chain)
980 static inline void ll_dir_chain_fini(struct ll_dir_chain *chain)
984 extern const struct file_operations ll_dir_operations;
985 extern const struct inode_operations ll_dir_inode_operations;
986 #ifdef HAVE_DIR_CONTEXT
987 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
988 struct dir_context *ctx);
990 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
991 void *cookie, filldir_t filldir);
993 int ll_get_mdt_idx(struct inode *inode);
994 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
995 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
996 __u64 offset, struct ll_dir_chain *chain);
997 void ll_release_page(struct inode *inode, struct page *page, bool remove);
1000 extern const struct inode_operations ll_special_inode_operations;
1002 struct inode *ll_iget(struct super_block *sb, ino_t hash,
1003 struct lustre_md *lic);
1004 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
1005 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1006 void *data, int flag);
1007 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
1008 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
1009 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
1012 int ll_writepage(struct page *page, struct writeback_control *wbc);
1013 int ll_writepages(struct address_space *, struct writeback_control *wbc);
1014 int ll_readpage(struct file *file, struct page *page);
1015 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
1016 struct cl_page *page, struct file *file);
1017 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
1018 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
1021 void ll_cl_add(struct file *file, const struct lu_env *env, struct cl_io *io,
1022 enum lcc_type type);
1023 void ll_cl_remove(struct file *file, const struct lu_env *env);
1024 struct ll_cl_context *ll_cl_find(struct file *file);
1026 extern const struct address_space_operations ll_aops;
1029 extern struct file_operations ll_file_operations;
1030 extern struct file_operations ll_file_operations_flock;
1031 extern struct file_operations ll_file_operations_noflock;
1032 extern struct inode_operations ll_file_inode_operations;
1033 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
1034 enum ldlm_mode l_req_mode);
1035 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1036 struct lustre_handle *lockh, __u64 flags,
1037 enum ldlm_mode mode);
1039 int ll_file_open(struct inode *inode, struct file *file);
1040 int ll_file_release(struct inode *inode, struct file *file);
1041 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1042 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1043 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1044 struct ll_file_data *file, loff_t pos,
1045 size_t count, int rw);
1046 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
1047 int ll_getattr(const struct path *path, struct kstat *stat,
1048 u32 request_mask, unsigned int flags);
1050 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1052 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1053 unsigned int flags);
1054 struct posix_acl *ll_get_acl(struct inode *inode, int type);
1055 #ifdef HAVE_IOP_SET_ACL
1056 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1057 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1058 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1059 #define ll_set_acl NULL
1060 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1064 static inline int ll_xflags_to_inode_flags(int xflags)
1066 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1067 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1068 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1069 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1072 static inline int ll_inode_flags_to_xflags(int flags)
1074 return ((flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1075 ((flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1076 ((flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1077 ((flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1080 int ll_migrate(struct inode *parent, struct file *file,
1081 struct lmv_user_md *lum, const char *name);
1082 int ll_get_fid_by_name(struct inode *parent, const char *name,
1083 int namelen, struct lu_fid *fid, struct inode **inode);
1084 int ll_inode_permission(struct inode *inode, int mask);
1085 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa);
1086 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1088 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1091 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1092 __u64 flags, struct lov_user_md *lum,
1094 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1095 struct lov_mds_md **lmm, int *lmm_size,
1096 struct ptlrpc_request **request);
1097 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1099 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1100 int *lmm_size, struct ptlrpc_request **request,
1101 struct ptlrpc_request **root_request, u64 valid);
1102 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1103 struct ptlrpc_request **request, u64 valid);
1104 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1105 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1106 int ll_fid2path(struct inode *inode, void __user *arg);
1107 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1108 int ll_hsm_release(struct inode *inode);
1109 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1110 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1112 /* llite/dcache.c */
1114 int ll_d_init(struct dentry *de);
1115 extern const struct dentry_operations ll_d_ops;
1116 void ll_intent_drop_lock(struct lookup_intent *);
1117 void ll_intent_release(struct lookup_intent *);
1118 void ll_invalidate_aliases(struct inode *);
1119 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1120 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1121 struct lookup_intent *it, struct dentry *de);
1123 /* llite/llite_lib.c */
1124 extern struct super_operations lustre_super_operations;
1126 void ll_lli_init(struct ll_inode_info *lli);
1127 int ll_fill_super(struct super_block *sb);
1128 void ll_put_super(struct super_block *sb);
1129 void ll_kill_super(struct super_block *sb);
1130 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1131 void ll_dir_clear_lsm_md(struct inode *inode);
1132 void ll_clear_inode(struct inode *inode);
1133 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1134 enum op_xvalid xvalid, bool hsm_import);
1135 int ll_setattr(struct dentry *de, struct iattr *attr);
1136 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1137 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1139 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1140 void ll_update_inode_flags(struct inode *inode, int ext_flags);
1141 int ll_read_inode2(struct inode *inode, void *opaque);
1142 void ll_delete_inode(struct inode *inode);
1143 int ll_iocontrol(struct inode *inode, struct file *file,
1144 unsigned int cmd, unsigned long arg);
1145 int ll_flush_ctx(struct inode *inode);
1146 void ll_umount_begin(struct super_block *sb);
1147 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1148 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1149 void ll_dirty_page_discard_warn(struct page *page, int ioret);
1150 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1151 struct super_block *, struct lookup_intent *);
1152 int ll_obd_statfs(struct inode *inode, void __user *arg);
1153 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1154 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1155 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1157 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1158 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1159 struct inode *i1, struct inode *i2,
1160 const char *name, size_t namelen,
1161 __u32 mode, enum md_op_code opc,
1163 void ll_finish_md_op_data(struct md_op_data *op_data);
1164 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
1165 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1166 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1167 struct lov_user_md **kbuf);
1168 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req);
1170 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req,
1171 struct lookup_intent *it);
1173 /* Compute expected user md size when passing in a md from user space */
1174 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1176 switch (lum->lmm_magic) {
1177 case LOV_USER_MAGIC_V1:
1178 return sizeof(struct lov_user_md_v1);
1179 case LOV_USER_MAGIC_V3:
1180 return sizeof(struct lov_user_md_v3);
1181 case LOV_USER_MAGIC_SPECIFIC:
1182 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1185 return lov_user_md_size(lum->lmm_stripe_count,
1186 LOV_USER_MAGIC_SPECIFIC);
1187 case LOV_USER_MAGIC_COMP_V1:
1188 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1189 case LOV_USER_MAGIC_FOREIGN:
1190 return foreign_size(lum);
1196 /* llite/llite_nfs.c */
1197 extern struct export_operations lustre_export_operations;
1198 __u32 get_uuid2int(const char *name, int len);
1199 struct inode *search_inode_for_lustre(struct super_block *sb,
1200 const struct lu_fid *fid);
1201 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1203 /* llite/symlink.c */
1204 extern struct inode_operations ll_fast_symlink_inode_operations;
1207 * IO arguments for various VFS I/O interfaces.
1209 struct vvp_io_args {
1210 /** normal/sendfile/splice */
1211 enum vvp_io_subtype via_io_subtype;
1215 struct kiocb *via_iocb;
1216 struct iov_iter *via_iter;
1219 struct pipe_inode_info *via_pipe;
1220 unsigned int via_flags;
1230 struct ll_cl_context {
1231 struct list_head lcc_list;
1233 const struct lu_env *lcc_env;
1234 struct cl_io *lcc_io;
1235 struct cl_page *lcc_page;
1236 enum lcc_type lcc_type;
1239 struct ll_thread_info {
1240 struct vvp_io_args lti_args;
1241 struct ra_io_arg lti_ria;
1242 struct ll_cl_context lti_io_ctx;
1245 extern struct lu_context_key ll_thread_key;
1247 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1249 struct ll_thread_info *lti;
1251 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1252 LASSERT(lti != NULL);
1257 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env,
1258 enum vvp_io_subtype type)
1260 struct vvp_io_args *via = &ll_env_info(env)->lti_args;
1262 via->via_io_subtype = type;
1267 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1268 struct vvp_io_args *args);
1270 /* llite/llite_mmap.c */
1272 int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last);
1273 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1274 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1275 unsigned long addr, size_t count);
1276 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1279 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1281 /* don't need an addref as the sb_info should be holding one */
1282 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1284 return ll_s2sbi(sb)->ll_dt_exp;
1287 /* don't need an addref as the sb_info should be holding one */
1288 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1290 return ll_s2sbi(sb)->ll_md_exp;
1293 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1295 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1301 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1302 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1304 return ll_s2sbi(inode->i_sb);
1307 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1309 return ll_s2dtexp(inode->i_sb);
1312 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1314 return ll_s2mdexp(inode->i_sb);
1317 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1321 LASSERT(inode != NULL);
1322 fid = &ll_i2info(inode)->lli_fid;
1327 static inline bool ll_dir_striped(struct inode *inode)
1330 return S_ISDIR(inode->i_mode) &&
1331 lmv_dir_striped(ll_i2info(inode)->lli_lsm_md);
1334 static inline loff_t ll_file_maxbytes(struct inode *inode)
1336 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1339 return MAX_LFS_FILESIZE;
1341 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1345 extern const struct xattr_handler *ll_xattr_handlers[];
1347 #define XATTR_USER_T 1
1348 #define XATTR_TRUSTED_T 2
1349 #define XATTR_SECURITY_T 3
1350 #define XATTR_ACL_ACCESS_T 4
1351 #define XATTR_ACL_DEFAULT_T 5
1352 #define XATTR_LUSTRE_T 6
1353 #define XATTR_OTHER_T 7
1355 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1356 int ll_xattr_list(struct inode *inode, const char *name, int type,
1357 void *buffer, size_t size, u64 valid);
1358 const struct xattr_handler *get_xattr_type(const char *name);
1361 * Common IO arguments for various VFS I/O interfaces.
1363 int cl_sb_init(struct super_block *sb);
1364 int cl_sb_fini(struct super_block *sb);
1366 enum ras_update_flags {
1370 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1371 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1375 #define LL_SA_RPC_MIN 2
1376 #define LL_SA_RPC_DEF 32
1377 #define LL_SA_RPC_MAX 512
1379 /* XXX: If want to support more concurrent statahead instances,
1380 * please consider to decentralize the RPC lists attached
1381 * on related import, such as imp_{sending,delayed}_list.
1383 #define LL_SA_RUNNING_MAX 256
1384 #define LL_SA_RUNNING_DEF 16
1386 #define LL_SA_CACHE_BIT 5
1387 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1388 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1390 /* per inode struct, for dir only */
1391 struct ll_statahead_info {
1392 struct dentry *sai_dentry;
1393 atomic_t sai_refcount; /* when access this struct, hold
1395 unsigned int sai_max; /* max ahead of lookup */
1396 __u64 sai_sent; /* stat requests sent count */
1397 __u64 sai_replied; /* stat requests which received
1399 __u64 sai_index; /* index of statahead entry */
1400 __u64 sai_index_wait; /* index of entry which is the
1401 * caller is waiting for */
1402 __u64 sai_hit; /* hit count */
1403 __u64 sai_miss; /* miss count:
1404 * for "ls -al" case, includes
1405 * hidden dentry miss;
1406 * for "ls -l" case, it does not
1407 * include hidden dentry miss.
1408 * "sai_miss_hidden" is used for
1411 unsigned int sai_consecutive_miss; /* consecutive miss */
1412 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1413 * is not a hidden one */
1414 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1416 unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
1418 sai_agl_valid:1,/* AGL is valid for the dir */
1419 sai_in_readpage:1;/* statahead is in readdir()*/
1420 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1421 struct task_struct *sai_task; /* stat-ahead thread */
1422 struct task_struct *sai_agl_task; /* AGL thread */
1423 struct list_head sai_interim_entries; /* entries which got async
1424 * stat reply, but not
1426 struct list_head sai_entries; /* completed entries */
1427 struct list_head sai_agls; /* AGLs to be sent */
1428 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1429 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1430 atomic_t sai_cache_count; /* entry count in cache */
1433 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1435 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1436 void ll_authorize_statahead(struct inode *dir, void *key);
1437 void ll_deauthorize_statahead(struct inode *dir, void *key);
1440 blkcnt_t dirty_cnt(struct inode *inode);
1442 int cl_glimpse_size0(struct inode *inode, int agl);
1443 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1444 struct inode *inode, struct cl_object *clob, int agl);
1446 static inline int cl_glimpse_size(struct inode *inode)
1448 return cl_glimpse_size0(inode, 0);
1451 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1452 * part of statahead */
1453 static inline int cl_agl(struct inode *inode)
1455 return cl_glimpse_size0(inode, 1);
1458 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1460 int cl_io_get(struct inode *inode, struct lu_env **envout,
1461 struct cl_io **ioout, __u16 *refcheck);
1463 static inline int ll_glimpse_size(struct inode *inode)
1465 struct ll_inode_info *lli = ll_i2info(inode);
1468 down_read(&lli->lli_glimpse_sem);
1469 rc = cl_glimpse_size(inode);
1470 lli->lli_glimpse_time = ktime_get();
1471 up_read(&lli->lli_glimpse_sem);
1475 /* dentry may statahead when statahead is enabled and current process has opened
1476 * parent directory, and this dentry hasn't accessed statahead cache before */
1478 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1480 struct ll_inode_info *lli;
1481 struct ll_dentry_data *ldd;
1483 if (ll_i2sbi(dir)->ll_sa_max == 0)
1486 lli = ll_i2info(dir);
1488 /* statahead is not allowed for this dir, there may be three causes:
1489 * 1. dir is not opened.
1490 * 2. statahead hit ratio is too low.
1491 * 3. previous stat started statahead thread failed. */
1492 if (!lli->lli_sa_enabled)
1495 /* not the same process, don't statahead */
1496 if (lli->lli_opendir_pid != current->pid)
1500 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1501 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1502 * For patchless client, lookup intent is not accurate, which may
1503 * misguide statahead. For example:
1504 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1505 * have the same intent -- IT_GETATTR, while one dentry should access
1506 * statahead cache once, otherwise statahead windows is messed up.
1507 * The solution is as following:
1508 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1509 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1510 * bypass interacting with statahead cache by checking
1511 * 'lld_sa_generation == lli->lli_sa_generation'.
1513 ldd = ll_d2d(dentry);
1514 if (ldd != NULL && lli->lli_sa_generation &&
1515 ldd->lld_sa_generation == lli->lli_sa_generation)
1521 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1522 enum cl_fsync_mode mode, int ignore_layout);
1524 static inline int ll_file_nolock(const struct file *file)
1526 struct ll_file_data *fd = file->private_data;
1527 struct inode *inode = file_inode((struct file *)file);
1529 LASSERT(fd != NULL);
1530 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1531 (ll_i2sbi(inode)->ll_flags & LL_SBI_NOLCK));
1534 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1535 struct lookup_intent *it, __u64 *bits)
1537 if (!it->it_lock_set) {
1538 struct lustre_handle handle;
1540 /* If this inode is a remote object, it will get two
1541 * separate locks in different namespaces, Master MDT,
1542 * where the name entry is, will grant LOOKUP lock,
1543 * remote MDT, where the object is, will grant
1544 * UPDATE|PERM lock. The inode will be attched to both
1545 * LOOKUP and PERM locks, so revoking either locks will
1546 * case the dcache being cleared */
1547 if (it->it_remote_lock_mode) {
1548 handle.cookie = it->it_remote_lock_handle;
1549 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1550 "(%p) for remote lock %#llx\n",
1551 PFID(ll_inode2fid(inode)), inode,
1553 md_set_lock_data(exp, &handle, inode, NULL);
1556 handle.cookie = it->it_lock_handle;
1558 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1559 " for lock %#llx\n",
1560 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1562 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1563 it->it_lock_set = 1;
1567 *bits = it->it_lock_bits;
1570 static inline int d_lustre_invalid(const struct dentry *dentry)
1572 struct ll_dentry_data *lld = ll_d2d(dentry);
1574 return (lld == NULL) || lld->lld_invalid;
1577 static inline void __d_lustre_invalidate(struct dentry *dentry)
1579 struct ll_dentry_data *lld = ll_d2d(dentry);
1582 lld->lld_invalid = 1;
1586 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1587 * ll_md_blocking_ast), unhash this dentry, and let dcache to reclaim it later;
1588 * else dput() of the last refcount will unhash this dentry and kill it.
1590 static inline void d_lustre_invalidate(struct dentry *dentry, int nested)
1592 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1594 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1596 spin_lock_nested(&dentry->d_lock,
1597 nested ? DENTRY_D_LOCK_NESTED : DENTRY_D_LOCK_NORMAL);
1598 __d_lustre_invalidate(dentry);
1600 * We should be careful about dentries created by d_obtain_alias().
1601 * These dentries are not put in the dentry tree, instead they are
1602 * linked to sb->s_anon through dentry->d_hash.
1603 * shrink_dcache_for_umount() shrinks the tree and sb->s_anon list.
1604 * If we unhashed such a dentry, unmount would not be able to find
1605 * it and busy inodes would be reported.
1607 if (ll_d_count(dentry) == 0 && !(dentry->d_flags & DCACHE_DISCONNECTED))
1609 spin_unlock(&dentry->d_lock);
1612 static inline void d_lustre_revalidate(struct dentry *dentry)
1614 spin_lock(&dentry->d_lock);
1615 LASSERT(ll_d2d(dentry) != NULL);
1616 ll_d2d(dentry)->lld_invalid = 0;
1617 spin_unlock(&dentry->d_lock);
1620 static inline dev_t ll_compat_encode_dev(dev_t dev)
1622 /* The compat_sys_*stat*() syscalls will fail unless the
1623 * device majors and minors are both less than 256. Note that
1624 * the value returned here will be passed through
1625 * old_encode_dev() in cp_compat_stat(). And so we are not
1626 * trying to return a valid compat (u16) device number, just
1627 * one that will pass the old_valid_dev() check. */
1629 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1632 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1633 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1634 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1635 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1636 struct lu_extent *ext);
1638 int ll_xattr_init(void);
1639 void ll_xattr_fini(void);
1641 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1642 struct cl_page *page, enum cl_req_type crt);
1644 int ll_getparent(struct file *file, struct getparent __user *arg);
1647 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1648 enum op_xvalid xvalid, unsigned int attr_flags);
1650 extern struct lu_env *cl_inode_fini_env;
1651 extern __u16 cl_inode_fini_refcheck;
1653 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1654 void cl_inode_fini(struct inode *inode);
1656 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1657 u32 cl_fid_build_gen(const struct lu_fid *fid);
1659 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1661 return &ll_i2sbi(inode)->ll_pcc_super;
1664 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1666 return ll_i2pccs(ll_info2i(lli));
1669 #ifdef HAVE_LUSTRE_CRYPTO
1671 extern const struct llcrypt_operations lustre_cryptops;
1674 #endif /* LLITE_INTERNAL_H */