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/
32 #ifndef LLITE_INTERNAL_H
33 #define LLITE_INTERNAL_H
35 #include <lustre_disk.h> /* for s2sbi */
36 #include <lustre_linkea.h>
38 /* for struct cl_lock_descr and struct cl_io */
39 #include <cl_object.h>
40 #include <lustre_lmv.h>
41 #include <lustre_mdc.h>
42 #include <lustre_intent.h>
43 #include <linux/compat.h>
44 #include <linux/aio.h>
45 #include <lustre_compat.h>
46 #include <lustre_crypto.h>
47 #include <range_lock.h>
49 #include "vvp_internal.h"
51 #include "foreign_symlink.h"
57 #ifndef HAVE_VM_FAULT_RETRY
58 #define VM_FAULT_RETRY 0
61 /* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
62 * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
63 #ifndef LOOKUP_CONTINUE
64 #define LOOKUP_CONTINUE LOOKUP_PARENT
67 /** Only used on client-side for indicating the tail of dir hash/offset. */
68 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
69 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
71 /* 4UL * 1024 * 1024 */
72 #define LL_MAX_BLKSIZE_BITS 22
74 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
76 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
78 struct ll_dentry_data {
79 unsigned int lld_sa_generation;
80 unsigned int lld_invalid:1;
81 unsigned int lld_nfs_dentry:1;
82 struct rcu_head lld_rcu_head;
85 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
87 #define LLI_INODE_MAGIC 0x111d0de5
88 #define LLI_INODE_DEAD 0xdeadd00d
90 struct ll_getname_data {
91 #ifdef HAVE_DIR_CONTEXT
92 struct dir_context ctx;
94 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
95 struct lu_fid lgd_fid; /* target fid we are looking for */
96 int lgd_found; /* inode matched? */
100 struct lu_env *lg_env;
102 struct cl_lock *lg_lock;
103 unsigned long lg_gid;
106 /* See comment on trunc_sem_down_read_nowait */
107 struct ll_trunc_sem {
108 /* when positive, this is a count of readers, when -1, it indicates
109 * the semaphore is held for write, and 0 is unlocked
111 atomic_t ll_trunc_readers;
112 /* this tracks a count of waiting writers */
113 atomic_t ll_trunc_waiters;
116 struct ll_inode_info {
117 __u32 lli_inode_magic;
120 volatile unsigned long lli_flags;
121 struct posix_acl *lli_posix_acl;
123 /* identifying fields for both metadata and data stacks. */
124 struct lu_fid lli_fid;
125 /* master inode fid for stripe directory */
126 struct lu_fid lli_pfid;
128 /* We need all three because every inode may be opened in different
130 struct obd_client_handle *lli_mds_read_och;
131 struct obd_client_handle *lli_mds_write_och;
132 struct obd_client_handle *lli_mds_exec_och;
133 __u64 lli_open_fd_read_count;
134 __u64 lli_open_fd_write_count;
135 __u64 lli_open_fd_exec_count;
137 /* Number of times this inode was opened */
138 u64 lli_open_fd_count;
139 /* When last close was performed on this inode */
140 ktime_t lli_close_fd_time;
142 /* Protects access to och pointers and their usage counters */
143 struct mutex lli_och_mutex;
145 struct inode lli_vfs_inode;
147 /* the most recent timestamps obtained from mds */
152 spinlock_t lli_agl_lock;
154 /* Try to make the d::member and f::member are aligned. Before using
155 * these members, make clear whether it is directory or not. */
159 /* metadata statahead */
160 /* since parent-child threads can share the same @file
161 * struct, "opendir_key" is the token when dir close for
162 * case of parent exit before child -- it is me should
163 * cleanup the dir readahead. */
164 void *lli_opendir_key;
165 struct ll_statahead_info *lli_sai;
166 /* protect statahead stuff. */
167 spinlock_t lli_sa_lock;
168 /* "opendir_pid" is the token when lookup/revalid
169 * -- I am the owner of dir statahead. */
170 pid_t lli_opendir_pid;
171 /* directory depth to ROOT */
172 unsigned short lli_depth;
173 /* stat will try to access statahead entries or start
174 * statahead if this flag is set, and this flag will be
175 * set upon dir open, and cleared when dir is closed,
176 * statahead hit ratio is too low, or start statahead
178 unsigned short lli_sa_enabled:1;
179 /* generation for statahead */
180 unsigned int lli_sa_generation;
181 /* rw lock protects lli_lsm_md */
182 struct rw_semaphore lli_lsm_sem;
183 /* directory stripe information */
184 struct lmv_stripe_md *lli_lsm_md;
185 /* directory default LMV */
186 struct lmv_stripe_md *lli_default_lsm_md;
189 /* for non-directory */
191 struct mutex lli_size_mutex;
192 char *lli_symlink_name;
193 struct ll_trunc_sem lli_trunc_sem;
194 struct range_lock_tree lli_write_tree;
195 struct mutex lli_setattr_mutex;
197 struct rw_semaphore lli_glimpse_sem;
198 ktime_t lli_glimpse_time;
199 struct list_head lli_agl_list;
202 /* for writepage() only to communicate to fsync */
205 /* protect the file heat fields */
206 spinlock_t lli_heat_lock;
207 __u32 lli_heat_flags;
208 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
211 * Whenever a process try to read/write the file, the
212 * jobid of the process will be saved here, and it'll
213 * be packed into the write PRC when flush later.
215 * So the read/write statistics for jobid will not be
216 * accurate if the file is shared by different jobs.
218 char lli_jobid[LUSTRE_JOBID_SIZE];
220 struct mutex lli_pcc_lock;
221 enum lu_pcc_state_flags lli_pcc_state;
223 * @lli_pcc_generation saves the gobal PCC generation
224 * when the file was successfully attached into PCC.
225 * The flags of the PCC dataset are saved in
227 * The gobal PCC generation will be increased when add
228 * or delete a PCC backend, or change the configuration
229 * parameters for PCC.
230 * If @lli_pcc_generation is same as the gobal PCC
231 * generation, we can use the saved flags of the PCC
232 * dataset to determine whether need to try auto attach
235 __u64 lli_pcc_generation;
236 enum pcc_dataset_flags lli_pcc_dsflags;
237 struct pcc_inode *lli_pcc_inode;
239 struct mutex lli_group_mutex;
240 __u64 lli_group_users;
241 unsigned long lli_group_gid;
243 __u64 lli_attr_valid;
245 __u64 lli_lazyblocks;
249 /* XXX: For following frequent used members, although they maybe special
250 * used for non-directory object, it is some time-wasting to check
251 * whether the object is directory or not before using them. On the
252 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
253 * the "ll_inode_info" size even if moving those members into u.f.
254 * So keep them out side.
256 * In the future, if more members are added only for directory,
257 * some of the following members can be moved into u.f.
259 struct cl_object *lli_clob;
261 /* mutex to request for layout lock exclusively. */
262 struct mutex lli_layout_mutex;
263 /* Layout version, protected by lli_layout_lock */
264 __u32 lli_layout_gen;
265 spinlock_t lli_layout_lock;
267 __u32 lli_projid; /* project id */
269 struct rw_semaphore lli_xattrs_list_rwsem;
270 struct mutex lli_xattrs_enq_lock;
271 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
272 struct list_head lli_lccs; /* list of ll_cl_context */
275 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
277 atomic_set(&sem->ll_trunc_readers, 0);
278 atomic_set(&sem->ll_trunc_waiters, 0);
281 /* This version of down read ignores waiting writers, meaning if the semaphore
282 * is already held for read, this down_read will 'join' that reader and also
283 * take the semaphore.
285 * This lets us avoid an unusual deadlock.
287 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
288 * in Lustre, in order to exclude truncates. Some of these paths then need to
289 * take the mmap_lock, while still holding the trunc_sem. The problem is that
290 * page faults hold the mmap_lock when calling in to Lustre, and then must also
291 * take the trunc_sem to exclude truncate.
293 * This means the locking order for trunc_sem and mmap_lock is sometimes AB,
294 * sometimes BA. This is almost OK because in both cases, we take the trunc
295 * sem for read, so it doesn't block.
297 * However, if a write mode user (truncate, a setattr op) arrives in the
298 * middle of this, the second reader on the truncate_sem will wait behind that
301 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
302 * to the mode in which they take the semaphore):
303 * reader (holding mmap_lock, needs truncate_sem)
305 * reader (holding truncate sem, waiting for mmap_lock)
307 * And so the readers deadlock.
309 * The solution is this modified semaphore, where this down_read ignores
310 * waiting write operations, and all waiters are woken up at once, so readers
311 * using down_read_nowait cannot get stuck behind waiting writers, regardless
312 * of the order they arrived in.
314 * down_read_nowait is only used in the page fault case, where we already hold
315 * the mmap_lock. This is because otherwise repeated read and write operations
316 * (which take the truncate sem) could prevent a truncate from ever starting.
317 * This could still happen with page faults, but without an even more complex
318 * mechanism, this is unavoidable.
322 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
324 wait_var_event(&sem->ll_trunc_readers,
325 atomic_inc_unless_negative(&sem->ll_trunc_readers));
328 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
330 wait_var_event(&sem->ll_trunc_readers,
331 atomic_read(&sem->ll_trunc_waiters) == 0 &&
332 atomic_inc_unless_negative(&sem->ll_trunc_readers));
335 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
337 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
338 atomic_read(&sem->ll_trunc_waiters))
339 wake_up_var(&sem->ll_trunc_readers);
342 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
344 atomic_inc(&sem->ll_trunc_waiters);
345 wait_var_event(&sem->ll_trunc_readers,
346 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
347 atomic_dec(&sem->ll_trunc_waiters);
350 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
352 atomic_set(&sem->ll_trunc_readers, 0);
353 wake_up_var(&sem->ll_trunc_readers);
356 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
357 static inline void lli_clear_acl(struct ll_inode_info *lli)
359 if (lli->lli_posix_acl) {
360 posix_acl_release(lli->lli_posix_acl);
361 lli->lli_posix_acl = NULL;
365 static inline void lli_replace_acl(struct ll_inode_info *lli,
366 struct lustre_md *md)
368 write_lock(&lli->lli_lock);
369 if (lli->lli_posix_acl)
370 posix_acl_release(lli->lli_posix_acl);
371 lli->lli_posix_acl = md->posix_acl;
372 write_unlock(&lli->lli_lock);
375 static inline void lli_clear_acl(struct ll_inode_info *lli)
379 static inline void lli_replace_acl(struct ll_inode_info *lli,
380 struct lustre_md *md)
385 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
389 spin_lock(&lli->lli_layout_lock);
390 gen = lli->lli_layout_gen;
391 spin_unlock(&lli->lli_layout_lock);
396 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
398 spin_lock(&lli->lli_layout_lock);
399 lli->lli_layout_gen = gen;
400 spin_unlock(&lli->lli_layout_lock);
404 /* File data is modified. */
405 LLIF_DATA_MODIFIED = 0,
406 /* File is being restored */
407 LLIF_FILE_RESTORING = 1,
408 /* Xattr cache is attached to the file */
409 LLIF_XATTR_CACHE = 2,
410 /* Project inherit */
411 LLIF_PROJECT_INHERIT = 3,
412 /* update atime from MDS even if it's older than local inode atime. */
413 LLIF_UPDATE_ATIME = 4,
414 /* foreign file/dir can be unlinked unconditionnaly */
415 LLIF_FOREIGN_REMOVABLE = 5,
416 /* setting encryption context in progress */
417 LLIF_SET_ENC_CTX = 6,
421 int ll_xattr_cache_destroy(struct inode *inode);
423 int ll_xattr_cache_get(struct inode *inode,
429 int ll_xattr_cache_insert(struct inode *inode,
434 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
436 #ifdef CONFIG_SECURITY
437 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
438 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
444 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
446 #ifdef CONFIG_SECURITY
447 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
451 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
452 const char **secctx_name, void **secctx,
454 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
457 int ll_listsecurity(struct inode *inode, char *secctx_name,
458 size_t secctx_name_size);
460 static inline bool obd_connect_has_enc(struct obd_connect_data *data)
462 #ifdef HAVE_LUSTRE_CRYPTO
463 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
464 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT;
470 static inline void obd_connect_set_enc(struct obd_connect_data *data)
472 #ifdef HAVE_LUSTRE_CRYPTO
473 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT;
478 * Locking to guarantee consistency of non-atomic updates to long long i_size,
479 * consistency between file size and KMS.
481 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
484 void ll_inode_size_lock(struct inode *inode);
485 void ll_inode_size_unlock(struct inode *inode);
487 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
489 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
492 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
494 return ll_i2info(inode)->lli_pcc_inode;
497 /* default to use at least 16M for fast read if possible */
498 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
500 /* default read-ahead on a given client mountpoint. */
501 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(1024UL)
503 /* default read-ahead for a single file descriptor */
504 #define SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX MiB_TO_PAGES(256UL)
506 /* default read-ahead full files smaller than limit on the second read */
507 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
509 /* default range pages */
510 #define SBI_DEFAULT_RA_RANGE_PAGES MiB_TO_PAGES(1ULL)
512 /* Min range pages */
513 #define RA_MIN_MMAP_RANGE_PAGES 16UL
518 RA_STAT_DISTANT_READPAGE,
519 RA_STAT_MISS_IN_WINDOW,
520 RA_STAT_FAILED_GRAB_PAGE,
521 RA_STAT_FAILED_MATCH,
526 RA_STAT_MAX_IN_FLIGHT,
527 RA_STAT_WRONG_GRAB_PAGE,
528 RA_STAT_FAILED_REACH_END,
530 RA_STAT_FAILED_FAST_READ,
531 RA_STAT_MMAP_RANGE_READ,
536 atomic_t ra_cur_pages;
537 unsigned long ra_max_pages;
538 unsigned long ra_max_pages_per_file;
539 unsigned long ra_range_pages;
540 unsigned long ra_max_read_ahead_whole_pages;
541 struct workqueue_struct *ll_readahead_wq;
543 * Max number of active works could be triggered
544 * for async readahead.
546 unsigned int ra_async_max_active;
547 /* how many async readahead triggered in flight */
548 atomic_t ra_async_inflight;
549 /* Threshold to control when to trigger async readahead */
550 unsigned long ra_async_pages_per_file_threshold;
553 /* ra_io_arg will be filled in the beginning of ll_readahead with
554 * ras_lock, then the following ll_read_ahead_pages will read RA
555 * pages according to this arg, all the items in this structure are
556 * counted by page index.
559 pgoff_t ria_start_idx; /* start offset of read-ahead*/
560 pgoff_t ria_end_idx; /* end offset of read-ahead*/
561 unsigned long ria_reserved; /* reserved pages for read-ahead */
562 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
563 bool ria_eof; /* reach end of file */
564 /* If stride read pattern is detected, ria_stoff is the byte offset
565 * where stride read is started. Note: for normal read-ahead, the
566 * value here is meaningless, and also it will not be accessed*/
568 /* ria_length and ria_bytes are the length and pages length in the
569 * stride I/O mode. And they will also be used to check whether
570 * it is stride I/O read-ahead in the read-ahead pages*/
575 /* LL_HIST_MAX=32 causes an overflow */
576 #define LL_HIST_MAX 28
577 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
578 #define LL_PROCESS_HIST_MAX 10
579 struct per_process_info {
581 struct obd_histogram pp_r_hist;
582 struct obd_histogram pp_w_hist;
585 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
586 struct ll_rw_extents_info {
587 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
590 #define LL_OFFSET_HIST_MAX 100
591 struct ll_rw_process_info {
594 loff_t rw_range_start;
596 loff_t rw_last_file_pos;
598 size_t rw_smallest_extent;
599 size_t rw_largest_extent;
600 struct ll_file_data *rw_last_file;
603 enum stats_track_type {
604 STATS_TRACK_ALL = 0, /* track all processes */
605 STATS_TRACK_PID, /* track process with this pid */
606 STATS_TRACK_PPID, /* track processes with this ppid */
607 STATS_TRACK_GID, /* track processes with this gid */
611 /* flags for sbi->ll_flags */
612 #define LL_SBI_NOLCK 0x01 /* DLM locking disabled (directio-only) */
613 #define LL_SBI_CHECKSUM 0x02 /* checksum each page as it's written */
614 #define LL_SBI_FLOCK 0x04
615 #define LL_SBI_USER_XATTR 0x08 /* support user xattr */
616 #define LL_SBI_ACL 0x10 /* support ACL */
617 /* LL_SBI_RMT_CLIENT 0x40 remote client */
618 #define LL_SBI_MDS_CAPA 0x80 /* support mds capa, obsolete */
619 #define LL_SBI_OSS_CAPA 0x100 /* support oss capa, obsolete */
620 #define LL_SBI_LOCALFLOCK 0x200 /* Local flocks support by kernel */
621 #define LL_SBI_LRU_RESIZE 0x400 /* lru resize support */
622 #define LL_SBI_LAZYSTATFS 0x800 /* lazystatfs mount option */
623 /* LL_SBI_SOM_PREVIEW 0x1000 SOM preview mount option, obsolete */
624 #define LL_SBI_32BIT_API 0x2000 /* generate 32 bit inodes. */
625 #define LL_SBI_64BIT_HASH 0x4000 /* support 64-bits dir hash/offset */
626 #define LL_SBI_AGL_ENABLED 0x8000 /* enable agl */
627 #define LL_SBI_VERBOSE 0x10000 /* verbose mount/umount */
628 #define LL_SBI_LAYOUT_LOCK 0x20000 /* layout lock support */
629 #define LL_SBI_USER_FID2PATH 0x40000 /* allow fid2path by unprivileged users */
630 #define LL_SBI_XATTR_CACHE 0x80000 /* support for xattr cache */
631 #define LL_SBI_NOROOTSQUASH 0x100000 /* do not apply root squash */
632 #define LL_SBI_ALWAYS_PING 0x200000 /* always ping even if server
634 #define LL_SBI_FAST_READ 0x400000 /* fast read support */
635 #define LL_SBI_FILE_SECCTX 0x800000 /* set file security context at create */
636 /* LL_SBI_PIO 0x1000000 parallel IO support, introduced in
638 #define LL_SBI_TINY_WRITE 0x2000000 /* tiny write support */
639 #define LL_SBI_FILE_HEAT 0x4000000 /* file heat support */
640 #define LL_SBI_TEST_DUMMY_ENCRYPTION 0x8000000 /* test dummy encryption */
641 #define LL_SBI_ENCRYPT 0x10000000 /* client side encryption */
642 #define LL_SBI_FOREIGN_SYMLINK 0x20000000 /* foreign fake-symlink support */
643 /* foreign fake-symlink upcall registered */
644 #define LL_SBI_FOREIGN_SYMLINK_UPCALL 0x40000000
645 #define LL_SBI_PARALLEL_DIO 0x80000000 /* parallel (async) submission of
647 #define LL_SBI_FLAGS { \
675 "test_dummy_encryption", \
678 "foreign_symlink_upcall", \
682 /* This is embedded into llite super-blocks to keep track of connect
683 * flags (capabilities) supported by all imports given mount is
685 struct lustre_client_ocd {
686 /* This is conjunction of connect_flags across all imports
687 * (LOVs) this mount is connected to. This field is updated by
688 * cl_ocd_update() under ->lco_lock. */
690 struct mutex lco_lock;
691 struct obd_export *lco_md_exp;
692 struct obd_export *lco_dt_exp;
696 /* this protects pglist and ra_info. It isn't safe to
697 * grab from interrupt contexts */
699 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
700 spinlock_t ll_process_lock; /* ll_rw_process_info */
701 struct obd_uuid ll_sb_uuid;
702 struct obd_export *ll_md_exp;
703 struct obd_export *ll_dt_exp;
704 struct obd_device *ll_md_obd;
705 struct obd_device *ll_dt_obd;
706 struct dentry *ll_debugfs_entry;
707 struct lu_fid ll_root_fid; /* root object fid */
710 unsigned int ll_xattr_cache_enabled:1,
711 ll_xattr_cache_set:1, /* already set to 0/1 */
712 ll_client_common_fill_super_succeeded:1,
715 struct lustre_client_ocd ll_lco;
717 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
719 /* Used to track "unstable" pages on a client, and maintain a
720 * LRU list of clean pages. An "unstable" page is defined as
721 * any page which is sent to a server as part of a bulk request,
722 * but is uncommitted to stable storage. */
723 struct cl_client_cache *ll_cache;
725 struct lprocfs_stats *ll_ra_stats;
727 struct ll_ra_info ll_ra_info;
728 unsigned int ll_namelen;
729 const struct file_operations *ll_fop;
731 struct lu_site *ll_site;
732 struct cl_device *ll_cl;
734 struct ll_rw_extents_info ll_rw_extents_info;
735 int ll_extent_process_count;
736 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
737 unsigned int ll_offset_process_count;
738 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
739 unsigned int ll_rw_offset_entry_count;
740 int ll_stats_track_id;
741 enum stats_track_type ll_stats_track_type;
744 /* metadata stat-ahead */
745 unsigned int ll_sa_running_max;/* max concurrent
746 * statahead instances */
747 unsigned int ll_sa_max; /* max statahead RPCs */
748 atomic_t ll_sa_total; /* statahead thread started
750 atomic_t ll_sa_wrong; /* statahead thread stopped for
752 atomic_t ll_sa_running; /* running statahead thread
754 atomic_t ll_agl_total; /* AGL thread started count */
756 dev_t ll_sdev_orig; /* save s_dev before assign for
759 struct root_squash_info ll_squash;
762 /* st_blksize returned by stat(2), when non-zero */
763 unsigned int ll_stat_blksize;
765 /* maximum relative age of cached statfs results */
766 unsigned int ll_statfs_max_age;
768 struct kset ll_kset; /* sysfs object */
769 struct completion ll_kobj_unregister;
772 unsigned int ll_heat_decay_weight;
773 unsigned int ll_heat_period_second;
775 /* Opens of the same inode before we start requesting open lock */
776 u32 ll_oc_thrsh_count;
778 /* Time in ms between last inode close and next open to be considered
779 * instant back to back and would trigger an open lock request
783 /* Time in ms after last file close that we no longer count prior opens*/
786 /* filesystem fsname */
787 char ll_fsname[LUSTRE_MAXFSNAME + 1];
789 /* Persistent Client Cache */
790 struct pcc_super ll_pcc_super;
792 /* to protect vs updates in all following foreign symlink fields */
793 struct rw_semaphore ll_foreign_symlink_sem;
794 /* foreign symlink path prefix */
795 char *ll_foreign_symlink_prefix;
796 /* full prefix size including leading '\0' */
797 size_t ll_foreign_symlink_prefix_size;
798 /* foreign symlink path upcall */
799 char *ll_foreign_symlink_upcall;
800 /* foreign symlink path upcall infos */
801 struct ll_foreign_symlink_upcall_item *ll_foreign_symlink_upcall_items;
802 /* foreign symlink path upcall nb infos */
803 unsigned int ll_foreign_symlink_upcall_nb_items;
806 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
807 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
809 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT (5)
810 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MS (100) /* 0.1 second */
811 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS (60000) /* 1 minute */
814 * per file-descriptor read-ahead data.
816 struct ll_readahead_state {
818 /* End byte that read(2) try to read. */
819 loff_t ras_last_read_end_bytes;
821 * number of bytes read after last read-ahead window reset. As window
822 * is reset on each seek, this is effectively a number of consecutive
823 * accesses. Maybe ->ras_accessed_in_window is better name.
825 * XXX nikita: window is also reset (by ras_update()) when Lustre
826 * believes that memory pressure evicts read-ahead pages. In that
827 * case, it probably doesn't make sense to expand window to
828 * PTLRPC_MAX_BRW_PAGES on the third access.
830 loff_t ras_consecutive_bytes;
832 * number of read requests after the last read-ahead window reset
833 * As window is reset on each seek, this is effectively the number
834 * on consecutive read request and is used to trigger read-ahead.
836 unsigned long ras_consecutive_requests;
838 * Parameters of current read-ahead window. Handled by
839 * ras_update(). On the initial access to the file or after a seek,
840 * window is reset to 0. After 3 consecutive accesses, window is
841 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
842 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
844 pgoff_t ras_window_start_idx;
845 pgoff_t ras_window_pages;
847 /* Page index where min range read starts */
848 pgoff_t ras_range_min_start_idx;
849 /* Page index where mmap range read ends */
850 pgoff_t ras_range_max_end_idx;
851 /* number of mmap pages where last time detected */
852 pgoff_t ras_last_range_pages;
853 /* number of mmap range requests */
854 pgoff_t ras_range_requests;
857 * Optimal RPC size in pages.
858 * It decides how many pages will be sent for each read-ahead.
860 unsigned long ras_rpc_pages;
862 * Where next read-ahead should start at. This lies within read-ahead
863 * window. Read-ahead window is read in pieces rather than at once
864 * because: 1. lustre limits total number of pages under read-ahead by
865 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
866 * not covered by DLM lock.
868 pgoff_t ras_next_readahead_idx;
870 * Total number of ll_file_read requests issued, reads originating
871 * due to mmap are not counted in this total. This value is used to
872 * trigger full file read-ahead after multiple reads to a small file.
874 unsigned long ras_requests;
876 * The following 3 items are used for detecting the stride I/O
878 * In stride I/O mode,
879 * ...............|-----data-----|****gap*****|--------|******|....
880 * offset |-stride_bytes-|-stride_gap-|
881 * ras_stride_offset = offset;
882 * ras_stride_length = stride_bytes + stride_gap;
883 * ras_stride_bytes = stride_bytes;
884 * Note: all these three items are counted by bytes.
886 loff_t ras_stride_offset;
887 loff_t ras_stride_length;
888 loff_t ras_stride_bytes;
890 * number of consecutive stride request count, and it is similar as
891 * ras_consecutive_requests, but used for stride I/O mode.
892 * Note: only more than 2 consecutive stride request are detected,
893 * stride read-ahead will be enable
895 unsigned long ras_consecutive_stride_requests;
896 /* index of the last page that async readahead starts */
897 pgoff_t ras_async_last_readpage_idx;
898 /* whether we should increase readahead window */
899 bool ras_need_increase_window;
900 /* whether ra miss check should be skipped */
901 bool ras_no_miss_check;
904 struct ll_readahead_work {
905 /** File to readahead */
906 struct file *lrw_file;
907 pgoff_t lrw_start_idx;
910 /* async worker to handler read */
911 struct work_struct lrw_readahead_work;
912 char lrw_jobid[LUSTRE_JOBID_SIZE];
915 extern struct kmem_cache *ll_file_data_slab;
916 struct lustre_handle;
917 struct ll_file_data {
918 struct ll_readahead_state fd_ras;
919 struct ll_grouplock fd_grouplock;
923 /* openhandle if lease exists for this file.
924 * Borrow lli->lli_och_mutex to protect assignment */
925 struct obd_client_handle *fd_lease_och;
926 struct obd_client_handle *fd_och;
927 struct file *fd_file;
928 /* Indicate whether need to report failure when close.
929 * true: failure is known, not report again.
930 * false: unknown failure, should report. */
931 bool fd_write_failed;
932 bool ll_lock_no_expand;
933 /* Used by mirrored file to lead IOs to a specific mirror, usually
934 * for mirror resync. 0 means default. */
935 __u32 fd_designated_mirror;
936 /* The layout version when resync starts. Resync I/O should carry this
937 * layout version for verification to OST objects */
938 __u32 fd_layout_version;
939 struct pcc_file fd_pcc_file;
942 void llite_tunables_unregister(void);
943 int llite_tunables_register(void);
945 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
947 return &lli->lli_vfs_inode;
950 __u32 ll_i2suppgid(struct inode *i);
951 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
953 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
955 #if BITS_PER_LONG == 32
957 #elif defined(CONFIG_COMPAT)
958 if (unlikely(sbi->ll_flags & LL_SBI_32BIT_API))
961 # ifdef CONFIG_X86_X32
962 /* in_compat_syscall() returns true when called from a kthread
963 * and CONFIG_X86_X32 is enabled, which is wrong. So check
964 * whether the caller comes from a syscall (ie. not a kthread)
965 * before calling in_compat_syscall(). */
966 if (current->flags & PF_KTHREAD)
970 return unlikely(in_compat_syscall());
972 return unlikely(sbi->ll_flags & LL_SBI_32BIT_API);
976 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
978 return !!(sbi->ll_flags & LL_SBI_FAST_READ);
981 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
983 return !!(sbi->ll_flags & LL_SBI_TINY_WRITE);
986 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
988 return !!(sbi->ll_flags & LL_SBI_FILE_HEAT);
991 static inline bool ll_sbi_has_foreign_symlink(struct ll_sb_info *sbi)
993 return !!(sbi->ll_flags & LL_SBI_FOREIGN_SYMLINK);
996 static inline bool ll_sbi_has_parallel_dio(struct ll_sb_info *sbi)
998 return !!(sbi->ll_flags & LL_SBI_PARALLEL_DIO);
1001 void ll_ras_enter(struct file *f, loff_t pos, size_t count);
1003 /* llite/lcommon_misc.c */
1004 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
1005 enum obd_notify_event ev, void *owner);
1006 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
1007 struct ll_grouplock *lg);
1008 void cl_put_grouplock(struct ll_grouplock *lg);
1010 /* llite/lproc_llite.c */
1011 int ll_debugfs_register_super(struct super_block *sb, const char *name);
1012 void ll_debugfs_unregister_super(struct super_block *sb);
1013 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
1016 LPROC_LL_READ_BYTES,
1017 LPROC_LL_WRITE_BYTES,
1044 LPROC_LL_GETXATTR_HITS,
1046 LPROC_LL_REMOVEXATTR,
1047 LPROC_LL_INODE_PERM,
1049 LPROC_LL_INODE_OCOUNT,
1050 LPROC_LL_INODE_OPCLTM,
1051 LPROC_LL_FILE_OPCODES
1055 enum get_default_layout_type {
1056 GET_DEFAULT_LAYOUT_ROOT = 1,
1059 extern const struct file_operations ll_dir_operations;
1060 extern const struct inode_operations ll_dir_inode_operations;
1061 #ifdef HAVE_DIR_CONTEXT
1062 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1063 struct dir_context *ctx);
1065 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1066 void *cookie, filldir_t filldir);
1068 int ll_get_mdt_idx(struct inode *inode);
1069 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
1070 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
1072 void ll_release_page(struct inode *inode, struct page *page, bool remove);
1073 int quotactl_ioctl(struct super_block *sb, struct if_quotactl *qctl);
1076 extern const struct inode_operations ll_special_inode_operations;
1078 struct inode *ll_iget(struct super_block *sb, ino_t hash,
1079 struct lustre_md *lic);
1080 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
1081 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1082 void *data, int flag);
1083 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
1084 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
1085 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
1088 int ll_writepage(struct page *page, struct writeback_control *wbc);
1089 int ll_writepages(struct address_space *, struct writeback_control *wbc);
1090 int ll_readpage(struct file *file, struct page *page);
1091 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
1092 struct cl_page *page, struct file *file);
1093 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
1094 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
1097 void ll_cl_add(struct inode *inode, const struct lu_env *env, struct cl_io *io,
1098 enum lcc_type type);
1099 void ll_cl_remove(struct inode *inode, const struct lu_env *env);
1100 struct ll_cl_context *ll_cl_find(struct inode *inode);
1102 extern const struct address_space_operations ll_aops;
1105 extern const struct inode_operations ll_file_inode_operations;
1106 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi);
1107 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
1108 enum ldlm_mode l_req_mode);
1109 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1110 struct lustre_handle *lockh, __u64 flags,
1111 enum ldlm_mode mode);
1113 int ll_file_open(struct inode *inode, struct file *file);
1114 int ll_file_release(struct inode *inode, struct file *file);
1115 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1116 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1117 void ll_track_file_opens(struct inode *inode);
1118 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1119 struct ll_file_data *file, loff_t pos,
1120 size_t count, int rw);
1121 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
1122 int ll_getattr(const struct path *path, struct kstat *stat,
1123 u32 request_mask, unsigned int flags);
1125 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1127 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1128 unsigned int flags, bool foreign);
1129 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1130 struct posix_acl *ll_get_acl(struct inode *inode, int type);
1131 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1132 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1133 #define ll_get_acl NULL
1134 #define ll_set_acl NULL
1135 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1137 static inline int ll_xflags_to_inode_flags(int xflags)
1139 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1140 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1141 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1142 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1145 static inline int ll_inode_flags_to_xflags(int inode_flags)
1147 return ((inode_flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1148 ((inode_flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1149 ((inode_flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1150 ((inode_flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1153 int ll_migrate(struct inode *parent, struct file *file,
1154 struct lmv_user_md *lum, const char *name);
1155 int ll_get_fid_by_name(struct inode *parent, const char *name,
1156 int namelen, struct lu_fid *fid, struct inode **inode);
1157 int ll_inode_permission(struct inode *inode, int mask);
1158 int ll_ioctl_check_project(struct inode *inode, __u32 xflags, __u32 projid);
1159 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1161 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1163 int ll_ioctl_project(struct file *file, unsigned int cmd,
1166 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1167 __u64 flags, struct lov_user_md *lum,
1169 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1170 struct lov_mds_md **lmm, int *lmm_size,
1171 struct ptlrpc_request **request);
1172 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1174 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1175 int *lmm_size, struct ptlrpc_request **request,
1176 struct ptlrpc_request **root_request, u64 valid);
1177 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1178 struct ptlrpc_request **request, u64 valid);
1179 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1180 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1181 int ll_fid2path(struct inode *inode, void __user *arg);
1182 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1183 int ll_hsm_release(struct inode *inode);
1184 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1185 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1187 /* llite/dcache.c */
1189 extern const struct dentry_operations ll_d_ops;
1191 bool ll_d_setup(struct dentry *de, bool do_put);
1193 static inline bool lld_is_init(struct dentry *dentry)
1195 return ll_d2d(dentry);
1198 #define ll_d_setup(de, do_put) (true)
1199 #define lld_is_init(dentry) (true)
1202 void ll_intent_drop_lock(struct lookup_intent *);
1203 void ll_intent_release(struct lookup_intent *);
1204 void ll_prune_aliases(struct inode *inode);
1205 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1206 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1207 struct lookup_intent *it, struct dentry *de);
1209 /* llite/llite_lib.c */
1210 extern const struct super_operations lustre_super_operations;
1212 void ll_lli_init(struct ll_inode_info *lli);
1213 int ll_fill_super(struct super_block *sb);
1214 void ll_put_super(struct super_block *sb);
1215 void ll_kill_super(struct super_block *sb);
1216 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1217 void ll_dir_clear_lsm_md(struct inode *inode);
1218 void ll_clear_inode(struct inode *inode);
1219 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1220 enum op_xvalid xvalid, bool hsm_import);
1221 int ll_setattr(struct dentry *de, struct iattr *attr);
1222 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1223 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1225 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1226 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags);
1227 void ll_update_dir_depth(struct inode *dir, struct inode *inode);
1228 int ll_read_inode2(struct inode *inode, void *opaque);
1229 void ll_truncate_inode_pages_final(struct inode *inode);
1230 void ll_delete_inode(struct inode *inode);
1231 int ll_iocontrol(struct inode *inode, struct file *file,
1232 unsigned int cmd, unsigned long arg);
1233 int ll_flush_ctx(struct inode *inode);
1234 void ll_umount_begin(struct super_block *sb);
1235 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1236 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1237 void ll_dirty_page_discard_warn(struct page *page, int ioret);
1238 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
1239 struct super_block *sb, struct lookup_intent *it);
1240 int ll_obd_statfs(struct inode *inode, void __user *arg);
1241 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1242 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1243 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1245 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1246 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1247 struct inode *i1, struct inode *i2,
1248 const char *name, size_t namelen,
1249 __u32 mode, enum md_op_code opc,
1251 void ll_finish_md_op_data(struct md_op_data *op_data);
1252 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
1253 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1254 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1255 struct lov_user_md **kbuf);
1256 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill);
1258 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req);
1260 /* Compute expected user md size when passing in a md from user space */
1261 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1263 switch (lum->lmm_magic) {
1264 case LOV_USER_MAGIC_V1:
1265 return sizeof(struct lov_user_md_v1);
1266 case LOV_USER_MAGIC_V3:
1267 return sizeof(struct lov_user_md_v3);
1268 case LOV_USER_MAGIC_SPECIFIC:
1269 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1272 return lov_user_md_size(lum->lmm_stripe_count,
1273 LOV_USER_MAGIC_SPECIFIC);
1274 case LOV_USER_MAGIC_COMP_V1:
1275 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1276 case LOV_USER_MAGIC_FOREIGN:
1277 return foreign_size(lum);
1283 /* llite/llite_nfs.c */
1284 extern const struct export_operations lustre_export_operations;
1285 __u32 get_uuid2int(const char *name, int len);
1286 struct inode *search_inode_for_lustre(struct super_block *sb,
1287 const struct lu_fid *fid);
1288 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1290 /* llite/symlink.c */
1291 extern const struct inode_operations ll_fast_symlink_inode_operations;
1294 * IO arguments for various VFS I/O interfaces.
1296 struct vvp_io_args {
1297 /** normal/sendfile/splice */
1300 struct kiocb *via_iocb;
1301 struct iov_iter *via_iter;
1311 struct ll_cl_context {
1312 struct list_head lcc_list;
1314 const struct lu_env *lcc_env;
1315 struct cl_io *lcc_io;
1316 struct cl_page *lcc_page;
1317 enum lcc_type lcc_type;
1320 struct ll_thread_info {
1321 struct vvp_io_args lti_args;
1322 struct ra_io_arg lti_ria;
1323 struct ll_cl_context lti_io_ctx;
1326 extern struct lu_context_key ll_thread_key;
1328 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1330 struct ll_thread_info *lti;
1332 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1333 LASSERT(lti != NULL);
1338 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env)
1340 return &ll_env_info(env)->lti_args;
1343 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1344 struct vvp_io_args *args);
1346 /* llite/llite_mmap.c */
1348 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1349 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1350 unsigned long addr, size_t count);
1351 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1354 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1356 /* don't need an addref as the sb_info should be holding one */
1357 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1359 return ll_s2sbi(sb)->ll_dt_exp;
1362 /* don't need an addref as the sb_info should be holding one */
1363 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1365 return ll_s2sbi(sb)->ll_md_exp;
1368 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1370 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1376 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1377 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1379 return ll_s2sbi(inode->i_sb);
1382 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1384 return ll_s2dtexp(inode->i_sb);
1387 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1389 return ll_s2mdexp(inode->i_sb);
1392 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1396 LASSERT(inode != NULL);
1397 fid = &ll_i2info(inode)->lli_fid;
1402 static inline bool ll_dir_striped(struct inode *inode)
1405 return S_ISDIR(inode->i_mode) &&
1406 lmv_dir_striped(ll_i2info(inode)->lli_lsm_md);
1409 static inline loff_t ll_file_maxbytes(struct inode *inode)
1411 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1414 return MAX_LFS_FILESIZE;
1416 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1420 extern const struct xattr_handler *ll_xattr_handlers[];
1422 #define XATTR_USER_T 1
1423 #define XATTR_TRUSTED_T 2
1424 #define XATTR_SECURITY_T 3
1425 #define XATTR_ACL_ACCESS_T 4
1426 #define XATTR_ACL_DEFAULT_T 5
1427 #define XATTR_LUSTRE_T 6
1428 #define XATTR_OTHER_T 7
1430 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1431 int ll_xattr_list(struct inode *inode, const char *name, int type,
1432 void *buffer, size_t size, u64 valid);
1433 const struct xattr_handler *get_xattr_type(const char *name);
1436 * Common IO arguments for various VFS I/O interfaces.
1438 int cl_sb_init(struct super_block *sb);
1439 int cl_sb_fini(struct super_block *sb);
1441 enum ras_update_flags {
1445 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1446 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1450 #define LL_SA_RPC_MIN 2
1451 #define LL_SA_RPC_DEF 32
1452 #define LL_SA_RPC_MAX 512
1454 /* XXX: If want to support more concurrent statahead instances,
1455 * please consider to decentralize the RPC lists attached
1456 * on related import, such as imp_{sending,delayed}_list.
1458 #define LL_SA_RUNNING_MAX 256
1459 #define LL_SA_RUNNING_DEF 16
1461 #define LL_SA_CACHE_BIT 5
1462 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1463 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1465 /* per inode struct, for dir only */
1466 struct ll_statahead_info {
1467 struct dentry *sai_dentry;
1468 atomic_t sai_refcount; /* when access this struct, hold
1470 unsigned int sai_max; /* max ahead of lookup */
1471 __u64 sai_sent; /* stat requests sent count */
1472 __u64 sai_replied; /* stat requests which received
1474 __u64 sai_index; /* index of statahead entry */
1475 __u64 sai_index_wait; /* index of entry which is the
1476 * caller is waiting for */
1477 __u64 sai_hit; /* hit count */
1478 __u64 sai_miss; /* miss count:
1479 * for "ls -al" case, includes
1480 * hidden dentry miss;
1481 * for "ls -l" case, it does not
1482 * include hidden dentry miss.
1483 * "sai_miss_hidden" is used for
1486 unsigned int sai_consecutive_miss; /* consecutive miss */
1487 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1488 * is not a hidden one */
1489 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1491 unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
1493 sai_in_readpage:1;/* statahead is in readdir()*/
1494 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1495 struct task_struct *sai_task; /* stat-ahead thread */
1496 struct task_struct *sai_agl_task; /* AGL thread */
1497 struct list_head sai_interim_entries; /* entries which got async
1498 * stat reply, but not
1500 struct list_head sai_entries; /* completed entries */
1501 struct list_head sai_agls; /* AGLs to be sent */
1502 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1503 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1504 atomic_t sai_cache_count; /* entry count in cache */
1507 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1509 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1510 void ll_authorize_statahead(struct inode *dir, void *key);
1511 void ll_deauthorize_statahead(struct inode *dir, void *key);
1514 blkcnt_t dirty_cnt(struct inode *inode);
1516 int cl_glimpse_size0(struct inode *inode, int agl);
1517 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1518 struct inode *inode, struct cl_object *clob, int agl);
1520 static inline int cl_glimpse_size(struct inode *inode)
1522 return cl_glimpse_size0(inode, 0);
1525 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1526 * part of statahead */
1527 static inline int cl_agl(struct inode *inode)
1529 return cl_glimpse_size0(inode, 1);
1532 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1534 int cl_io_get(struct inode *inode, struct lu_env **envout,
1535 struct cl_io **ioout, __u16 *refcheck);
1537 static inline int ll_glimpse_size(struct inode *inode)
1539 struct ll_inode_info *lli = ll_i2info(inode);
1542 down_read(&lli->lli_glimpse_sem);
1543 rc = cl_glimpse_size(inode);
1544 lli->lli_glimpse_time = ktime_get();
1545 up_read(&lli->lli_glimpse_sem);
1549 /* dentry may statahead when statahead is enabled and current process has opened
1550 * parent directory, and this dentry hasn't accessed statahead cache before */
1552 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1554 struct ll_inode_info *lli;
1555 struct ll_dentry_data *ldd;
1557 if (ll_i2sbi(dir)->ll_sa_max == 0)
1560 lli = ll_i2info(dir);
1562 /* statahead is not allowed for this dir, there may be three causes:
1563 * 1. dir is not opened.
1564 * 2. statahead hit ratio is too low.
1565 * 3. previous stat started statahead thread failed. */
1566 if (!lli->lli_sa_enabled)
1569 /* not the same process, don't statahead */
1570 if (lli->lli_opendir_pid != current->pid)
1574 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1575 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1576 * For patchless client, lookup intent is not accurate, which may
1577 * misguide statahead. For example:
1578 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1579 * have the same intent -- IT_GETATTR, while one dentry should access
1580 * statahead cache once, otherwise statahead windows is messed up.
1581 * The solution is as following:
1582 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1583 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1584 * bypass interacting with statahead cache by checking
1585 * 'lld_sa_generation == lli->lli_sa_generation'.
1587 ldd = ll_d2d(dentry);
1588 if (ldd != NULL && lli->lli_sa_generation &&
1589 ldd->lld_sa_generation == lli->lli_sa_generation)
1595 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1596 enum cl_fsync_mode mode, int ignore_layout);
1598 static inline int ll_file_nolock(const struct file *file)
1600 struct ll_file_data *fd = file->private_data;
1601 struct inode *inode = file_inode((struct file *)file);
1603 LASSERT(fd != NULL);
1604 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1605 (ll_i2sbi(inode)->ll_flags & LL_SBI_NOLCK));
1608 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1609 struct lookup_intent *it, __u64 *bits)
1611 if (!it->it_lock_set) {
1612 struct lustre_handle handle;
1614 /* If this inode is a remote object, it will get two
1615 * separate locks in different namespaces, Master MDT,
1616 * where the name entry is, will grant LOOKUP lock,
1617 * remote MDT, where the object is, will grant
1618 * UPDATE|PERM lock. The inode will be attched to both
1619 * LOOKUP and PERM locks, so revoking either locks will
1620 * case the dcache being cleared */
1621 if (it->it_remote_lock_mode) {
1622 handle.cookie = it->it_remote_lock_handle;
1623 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1624 "(%p) for remote lock %#llx\n",
1625 PFID(ll_inode2fid(inode)), inode,
1627 md_set_lock_data(exp, &handle, inode, NULL);
1630 handle.cookie = it->it_lock_handle;
1632 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1633 " for lock %#llx\n",
1634 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1636 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1637 it->it_lock_set = 1;
1641 *bits = it->it_lock_bits;
1644 static inline int d_lustre_invalid(const struct dentry *dentry)
1646 return !ll_d2d(dentry) || ll_d2d(dentry)->lld_invalid;
1650 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1651 * ll_md_blocking_ast), it will be pruned by ll_prune_aliases() and
1652 * ll_prune_negative_children(); otherwise dput() of the last refcount will
1653 * unhash this dentry and kill it.
1655 static inline void d_lustre_invalidate(struct dentry *dentry)
1657 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1659 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1661 spin_lock(&dentry->d_lock);
1662 if (lld_is_init(dentry))
1663 ll_d2d(dentry)->lld_invalid = 1;
1664 spin_unlock(&dentry->d_lock);
1667 static inline void d_lustre_revalidate(struct dentry *dentry)
1669 spin_lock(&dentry->d_lock);
1670 LASSERT(ll_d2d(dentry));
1671 ll_d2d(dentry)->lld_invalid = 0;
1672 spin_unlock(&dentry->d_lock);
1675 static inline dev_t ll_compat_encode_dev(dev_t dev)
1677 /* The compat_sys_*stat*() syscalls will fail unless the
1678 * device majors and minors are both less than 256. Note that
1679 * the value returned here will be passed through
1680 * old_encode_dev() in cp_compat_stat(). And so we are not
1681 * trying to return a valid compat (u16) device number, just
1682 * one that will pass the old_valid_dev() check. */
1684 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1687 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1688 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1689 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1690 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1691 struct lu_extent *ext);
1693 int ll_xattr_init(void);
1694 void ll_xattr_fini(void);
1696 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1697 struct cl_page *page, enum cl_req_type crt);
1699 int ll_getparent(struct file *file, struct getparent __user *arg);
1702 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1703 enum op_xvalid xvalid, unsigned int attr_flags);
1705 extern struct lu_env *cl_inode_fini_env;
1706 extern __u16 cl_inode_fini_refcheck;
1708 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1709 void cl_inode_fini(struct inode *inode);
1711 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1712 u32 cl_fid_build_gen(const struct lu_fid *fid);
1714 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1716 return &ll_i2sbi(inode)->ll_pcc_super;
1719 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1721 return ll_i2pccs(ll_info2i(lli));
1725 /* The digested form is made of a FID (16 bytes) followed by the second-to-last
1726 * ciphertext block (16 bytes), so a total length of 32 bytes.
1727 * That way, llcrypt does not compute a digested form of this digest.
1729 struct ll_digest_filename {
1730 struct lu_fid ldf_fid;
1731 char ldf_excerpt[LLCRYPT_FNAME_DIGEST_SIZE];
1734 int ll_setup_filename(struct inode *dir, const struct qstr *iname,
1735 int lookup, struct llcrypt_name *fname,
1736 struct lu_fid *fid);
1737 int ll_fname_disk_to_usr(struct inode *inode,
1738 u32 hash, u32 minor_hash,
1739 struct llcrypt_str *iname, struct llcrypt_str *oname,
1740 struct lu_fid *fid);
1741 int ll_revalidate_d_crypto(struct dentry *dentry, unsigned int flags);
1742 #ifdef HAVE_LUSTRE_CRYPTO
1743 extern const struct llcrypt_operations lustre_cryptops;
1746 /* llite/llite_foreign.c */
1747 int ll_manage_foreign(struct inode *inode, struct lustre_md *lmd);
1748 bool ll_foreign_is_openable(struct dentry *dentry, unsigned int flags);
1749 bool ll_foreign_is_removable(struct dentry *dentry, bool unset);
1751 #endif /* LLITE_INTERNAL_H */