xref: /linux-6.15/include/linux/ceph/libceph.h (revision 497888cf)
13d14c5d2SYehuda Sadeh #ifndef _FS_CEPH_LIBCEPH_H
23d14c5d2SYehuda Sadeh #define _FS_CEPH_LIBCEPH_H
33d14c5d2SYehuda Sadeh 
43d14c5d2SYehuda Sadeh #include "ceph_debug.h"
53d14c5d2SYehuda Sadeh 
63d14c5d2SYehuda Sadeh #include <asm/unaligned.h>
73d14c5d2SYehuda Sadeh #include <linux/backing-dev.h>
83d14c5d2SYehuda Sadeh #include <linux/completion.h>
93d14c5d2SYehuda Sadeh #include <linux/exportfs.h>
103d14c5d2SYehuda Sadeh #include <linux/fs.h>
113d14c5d2SYehuda Sadeh #include <linux/mempool.h>
123d14c5d2SYehuda Sadeh #include <linux/pagemap.h>
133d14c5d2SYehuda Sadeh #include <linux/wait.h>
143d14c5d2SYehuda Sadeh #include <linux/writeback.h>
153d14c5d2SYehuda Sadeh #include <linux/slab.h>
163d14c5d2SYehuda Sadeh 
173d14c5d2SYehuda Sadeh #include "types.h"
183d14c5d2SYehuda Sadeh #include "messenger.h"
193d14c5d2SYehuda Sadeh #include "msgpool.h"
203d14c5d2SYehuda Sadeh #include "mon_client.h"
213d14c5d2SYehuda Sadeh #include "osd_client.h"
223d14c5d2SYehuda Sadeh #include "ceph_fs.h"
233d14c5d2SYehuda Sadeh 
243d14c5d2SYehuda Sadeh /*
253d14c5d2SYehuda Sadeh  * Supported features
263d14c5d2SYehuda Sadeh  */
273d14c5d2SYehuda Sadeh #define CEPH_FEATURE_SUPPORTED_DEFAULT CEPH_FEATURE_NOSRCADDR
283d14c5d2SYehuda Sadeh #define CEPH_FEATURE_REQUIRED_DEFAULT  CEPH_FEATURE_NOSRCADDR
293d14c5d2SYehuda Sadeh 
303d14c5d2SYehuda Sadeh /*
313d14c5d2SYehuda Sadeh  * mount options
323d14c5d2SYehuda Sadeh  */
333d14c5d2SYehuda Sadeh #define CEPH_OPT_FSID             (1<<0)
343d14c5d2SYehuda Sadeh #define CEPH_OPT_NOSHARE          (1<<1) /* don't share client with other sbs */
353d14c5d2SYehuda Sadeh #define CEPH_OPT_MYIP             (1<<2) /* specified my ip */
363d14c5d2SYehuda Sadeh #define CEPH_OPT_NOCRC            (1<<3) /* no data crc on writes */
373d14c5d2SYehuda Sadeh 
38*497888cfSPhil Carmody #define CEPH_OPT_DEFAULT   (0)
393d14c5d2SYehuda Sadeh 
403d14c5d2SYehuda Sadeh #define ceph_set_opt(client, opt) \
413d14c5d2SYehuda Sadeh 	(client)->options->flags |= CEPH_OPT_##opt;
423d14c5d2SYehuda Sadeh #define ceph_test_opt(client, opt) \
433d14c5d2SYehuda Sadeh 	(!!((client)->options->flags & CEPH_OPT_##opt))
443d14c5d2SYehuda Sadeh 
453d14c5d2SYehuda Sadeh struct ceph_options {
463d14c5d2SYehuda Sadeh 	int flags;
473d14c5d2SYehuda Sadeh 	struct ceph_fsid fsid;
483d14c5d2SYehuda Sadeh 	struct ceph_entity_addr my_addr;
493d14c5d2SYehuda Sadeh 	int mount_timeout;
503d14c5d2SYehuda Sadeh 	int osd_idle_ttl;
513d14c5d2SYehuda Sadeh 	int osd_timeout;
523d14c5d2SYehuda Sadeh 	int osd_keepalive_timeout;
533d14c5d2SYehuda Sadeh 
543d14c5d2SYehuda Sadeh 	/*
553d14c5d2SYehuda Sadeh 	 * any type that can't be simply compared or doesn't need need
563d14c5d2SYehuda Sadeh 	 * to be compared should go beyond this point,
573d14c5d2SYehuda Sadeh 	 * ceph_compare_options() should be updated accordingly
583d14c5d2SYehuda Sadeh 	 */
593d14c5d2SYehuda Sadeh 
603d14c5d2SYehuda Sadeh 	struct ceph_entity_addr *mon_addr; /* should be the first
613d14c5d2SYehuda Sadeh 					      pointer type of args */
623d14c5d2SYehuda Sadeh 	int num_mon;
633d14c5d2SYehuda Sadeh 	char *name;
648323c3aaSTommi Virtanen 	struct ceph_crypto_key *key;
653d14c5d2SYehuda Sadeh };
663d14c5d2SYehuda Sadeh 
673d14c5d2SYehuda Sadeh /*
683d14c5d2SYehuda Sadeh  * defaults
693d14c5d2SYehuda Sadeh  */
703d14c5d2SYehuda Sadeh #define CEPH_MOUNT_TIMEOUT_DEFAULT  60
713d14c5d2SYehuda Sadeh #define CEPH_OSD_TIMEOUT_DEFAULT    60  /* seconds */
723d14c5d2SYehuda Sadeh #define CEPH_OSD_KEEPALIVE_DEFAULT  5
733d14c5d2SYehuda Sadeh #define CEPH_OSD_IDLE_TTL_DEFAULT    60
743d14c5d2SYehuda Sadeh 
753d14c5d2SYehuda Sadeh #define CEPH_MSG_MAX_FRONT_LEN	(16*1024*1024)
763d14c5d2SYehuda Sadeh #define CEPH_MSG_MAX_DATA_LEN	(16*1024*1024)
773d14c5d2SYehuda Sadeh 
783d14c5d2SYehuda Sadeh #define CEPH_AUTH_NAME_DEFAULT   "guest"
793d14c5d2SYehuda Sadeh 
803d14c5d2SYehuda Sadeh /*
813d14c5d2SYehuda Sadeh  * Delay telling the MDS we no longer want caps, in case we reopen
823d14c5d2SYehuda Sadeh  * the file.  Delay a minimum amount of time, even if we send a cap
833d14c5d2SYehuda Sadeh  * message for some other reason.  Otherwise, take the oppotunity to
843d14c5d2SYehuda Sadeh  * update the mds to avoid sending another message later.
853d14c5d2SYehuda Sadeh  */
863d14c5d2SYehuda Sadeh #define CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT      5  /* cap release delay */
873d14c5d2SYehuda Sadeh #define CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT     60  /* cap release delay */
883d14c5d2SYehuda Sadeh 
893d14c5d2SYehuda Sadeh #define CEPH_CAP_RELEASE_SAFETY_DEFAULT        (CEPH_CAPS_PER_RELEASE * 4)
903d14c5d2SYehuda Sadeh 
913d14c5d2SYehuda Sadeh /* mount state */
923d14c5d2SYehuda Sadeh enum {
933d14c5d2SYehuda Sadeh 	CEPH_MOUNT_MOUNTING,
943d14c5d2SYehuda Sadeh 	CEPH_MOUNT_MOUNTED,
953d14c5d2SYehuda Sadeh 	CEPH_MOUNT_UNMOUNTING,
963d14c5d2SYehuda Sadeh 	CEPH_MOUNT_UNMOUNTED,
973d14c5d2SYehuda Sadeh 	CEPH_MOUNT_SHUTDOWN,
983d14c5d2SYehuda Sadeh };
993d14c5d2SYehuda Sadeh 
1003d14c5d2SYehuda Sadeh /*
1013d14c5d2SYehuda Sadeh  * subtract jiffies
1023d14c5d2SYehuda Sadeh  */
1033d14c5d2SYehuda Sadeh static inline unsigned long time_sub(unsigned long a, unsigned long b)
1043d14c5d2SYehuda Sadeh {
1053d14c5d2SYehuda Sadeh 	BUG_ON(time_after(b, a));
1063d14c5d2SYehuda Sadeh 	return (long)a - (long)b;
1073d14c5d2SYehuda Sadeh }
1083d14c5d2SYehuda Sadeh 
1093d14c5d2SYehuda Sadeh struct ceph_mds_client;
1103d14c5d2SYehuda Sadeh 
1113d14c5d2SYehuda Sadeh /*
1123d14c5d2SYehuda Sadeh  * per client state
1133d14c5d2SYehuda Sadeh  *
1143d14c5d2SYehuda Sadeh  * possibly shared by multiple mount points, if they are
1153d14c5d2SYehuda Sadeh  * mounting the same ceph filesystem/cluster.
1163d14c5d2SYehuda Sadeh  */
1173d14c5d2SYehuda Sadeh struct ceph_client {
1183d14c5d2SYehuda Sadeh 	struct ceph_fsid fsid;
1193d14c5d2SYehuda Sadeh 	bool have_fsid;
1203d14c5d2SYehuda Sadeh 
1213d14c5d2SYehuda Sadeh 	void *private;
1223d14c5d2SYehuda Sadeh 
1233d14c5d2SYehuda Sadeh 	struct ceph_options *options;
1243d14c5d2SYehuda Sadeh 
1253d14c5d2SYehuda Sadeh 	struct mutex mount_mutex;      /* serialize mount attempts */
1263d14c5d2SYehuda Sadeh 	wait_queue_head_t auth_wq;
1273d14c5d2SYehuda Sadeh 	int auth_err;
1283d14c5d2SYehuda Sadeh 
1293d14c5d2SYehuda Sadeh 	int (*extra_mon_dispatch)(struct ceph_client *, struct ceph_msg *);
1303d14c5d2SYehuda Sadeh 
1313d14c5d2SYehuda Sadeh 	u32 supported_features;
1323d14c5d2SYehuda Sadeh 	u32 required_features;
1333d14c5d2SYehuda Sadeh 
1343d14c5d2SYehuda Sadeh 	struct ceph_messenger *msgr;   /* messenger instance */
1353d14c5d2SYehuda Sadeh 	struct ceph_mon_client monc;
1363d14c5d2SYehuda Sadeh 	struct ceph_osd_client osdc;
1373d14c5d2SYehuda Sadeh 
1383d14c5d2SYehuda Sadeh #ifdef CONFIG_DEBUG_FS
1393d14c5d2SYehuda Sadeh 	struct dentry *debugfs_dir;
1403d14c5d2SYehuda Sadeh 	struct dentry *debugfs_monmap;
1413d14c5d2SYehuda Sadeh 	struct dentry *debugfs_osdmap;
1423d14c5d2SYehuda Sadeh #endif
1433d14c5d2SYehuda Sadeh };
1443d14c5d2SYehuda Sadeh 
1453d14c5d2SYehuda Sadeh 
1463d14c5d2SYehuda Sadeh 
1473d14c5d2SYehuda Sadeh /*
1483d14c5d2SYehuda Sadeh  * snapshots
1493d14c5d2SYehuda Sadeh  */
1503d14c5d2SYehuda Sadeh 
1513d14c5d2SYehuda Sadeh /*
1523d14c5d2SYehuda Sadeh  * A "snap context" is the set of existing snapshots when we
1533d14c5d2SYehuda Sadeh  * write data.  It is used by the OSD to guide its COW behavior.
1543d14c5d2SYehuda Sadeh  *
1553d14c5d2SYehuda Sadeh  * The ceph_snap_context is refcounted, and attached to each dirty
1563d14c5d2SYehuda Sadeh  * page, indicating which context the dirty data belonged when it was
1573d14c5d2SYehuda Sadeh  * dirtied.
1583d14c5d2SYehuda Sadeh  */
1593d14c5d2SYehuda Sadeh struct ceph_snap_context {
1603d14c5d2SYehuda Sadeh 	atomic_t nref;
1613d14c5d2SYehuda Sadeh 	u64 seq;
1623d14c5d2SYehuda Sadeh 	int num_snaps;
1633d14c5d2SYehuda Sadeh 	u64 snaps[];
1643d14c5d2SYehuda Sadeh };
1653d14c5d2SYehuda Sadeh 
1663d14c5d2SYehuda Sadeh static inline struct ceph_snap_context *
1673d14c5d2SYehuda Sadeh ceph_get_snap_context(struct ceph_snap_context *sc)
1683d14c5d2SYehuda Sadeh {
1693d14c5d2SYehuda Sadeh 	/*
1703d14c5d2SYehuda Sadeh 	printk("get_snap_context %p %d -> %d\n", sc, atomic_read(&sc->nref),
1713d14c5d2SYehuda Sadeh 	       atomic_read(&sc->nref)+1);
1723d14c5d2SYehuda Sadeh 	*/
1733d14c5d2SYehuda Sadeh 	if (sc)
1743d14c5d2SYehuda Sadeh 		atomic_inc(&sc->nref);
1753d14c5d2SYehuda Sadeh 	return sc;
1763d14c5d2SYehuda Sadeh }
1773d14c5d2SYehuda Sadeh 
1783d14c5d2SYehuda Sadeh static inline void ceph_put_snap_context(struct ceph_snap_context *sc)
1793d14c5d2SYehuda Sadeh {
1803d14c5d2SYehuda Sadeh 	if (!sc)
1813d14c5d2SYehuda Sadeh 		return;
1823d14c5d2SYehuda Sadeh 	/*
1833d14c5d2SYehuda Sadeh 	printk("put_snap_context %p %d -> %d\n", sc, atomic_read(&sc->nref),
1843d14c5d2SYehuda Sadeh 	       atomic_read(&sc->nref)-1);
1853d14c5d2SYehuda Sadeh 	*/
1863d14c5d2SYehuda Sadeh 	if (atomic_dec_and_test(&sc->nref)) {
1873d14c5d2SYehuda Sadeh 		/*printk(" deleting snap_context %p\n", sc);*/
1883d14c5d2SYehuda Sadeh 		kfree(sc);
1893d14c5d2SYehuda Sadeh 	}
1903d14c5d2SYehuda Sadeh }
1913d14c5d2SYehuda Sadeh 
1923d14c5d2SYehuda Sadeh /*
1933d14c5d2SYehuda Sadeh  * calculate the number of pages a given length and offset map onto,
1943d14c5d2SYehuda Sadeh  * if we align the data.
1953d14c5d2SYehuda Sadeh  */
1963d14c5d2SYehuda Sadeh static inline int calc_pages_for(u64 off, u64 len)
1973d14c5d2SYehuda Sadeh {
1983d14c5d2SYehuda Sadeh 	return ((off+len+PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT) -
1993d14c5d2SYehuda Sadeh 		(off >> PAGE_CACHE_SHIFT);
2003d14c5d2SYehuda Sadeh }
2013d14c5d2SYehuda Sadeh 
2023d14c5d2SYehuda Sadeh /* ceph_common.c */
2033d14c5d2SYehuda Sadeh extern const char *ceph_msg_type_name(int type);
2043d14c5d2SYehuda Sadeh extern int ceph_check_fsid(struct ceph_client *client, struct ceph_fsid *fsid);
2053d14c5d2SYehuda Sadeh extern struct kmem_cache *ceph_inode_cachep;
2063d14c5d2SYehuda Sadeh extern struct kmem_cache *ceph_cap_cachep;
2073d14c5d2SYehuda Sadeh extern struct kmem_cache *ceph_dentry_cachep;
2083d14c5d2SYehuda Sadeh extern struct kmem_cache *ceph_file_cachep;
2093d14c5d2SYehuda Sadeh 
2103d14c5d2SYehuda Sadeh extern int ceph_parse_options(struct ceph_options **popt, char *options,
2113d14c5d2SYehuda Sadeh 			      const char *dev_name, const char *dev_name_end,
2123d14c5d2SYehuda Sadeh 			      int (*parse_extra_token)(char *c, void *private),
2133d14c5d2SYehuda Sadeh 			      void *private);
2143d14c5d2SYehuda Sadeh extern void ceph_destroy_options(struct ceph_options *opt);
2153d14c5d2SYehuda Sadeh extern int ceph_compare_options(struct ceph_options *new_opt,
2163d14c5d2SYehuda Sadeh 				struct ceph_client *client);
2173d14c5d2SYehuda Sadeh extern struct ceph_client *ceph_create_client(struct ceph_options *opt,
2183d14c5d2SYehuda Sadeh 					      void *private);
2193d14c5d2SYehuda Sadeh extern u64 ceph_client_id(struct ceph_client *client);
2203d14c5d2SYehuda Sadeh extern void ceph_destroy_client(struct ceph_client *client);
2213d14c5d2SYehuda Sadeh extern int __ceph_open_session(struct ceph_client *client,
2223d14c5d2SYehuda Sadeh 			       unsigned long started);
2233d14c5d2SYehuda Sadeh extern int ceph_open_session(struct ceph_client *client);
2243d14c5d2SYehuda Sadeh 
2253d14c5d2SYehuda Sadeh /* pagevec.c */
2263d14c5d2SYehuda Sadeh extern void ceph_release_page_vector(struct page **pages, int num_pages);
2273d14c5d2SYehuda Sadeh 
2283d14c5d2SYehuda Sadeh extern struct page **ceph_get_direct_page_vector(const char __user *data,
229b6aa5901SHenry C Chang 						 int num_pages,
230b6aa5901SHenry C Chang 						 bool write_page);
231b6aa5901SHenry C Chang extern void ceph_put_page_vector(struct page **pages, int num_pages,
232b6aa5901SHenry C Chang 				 bool dirty);
2333d14c5d2SYehuda Sadeh extern void ceph_release_page_vector(struct page **pages, int num_pages);
2343d14c5d2SYehuda Sadeh extern struct page **ceph_alloc_page_vector(int num_pages, gfp_t flags);
2353d14c5d2SYehuda Sadeh extern int ceph_copy_user_to_page_vector(struct page **pages,
2363d14c5d2SYehuda Sadeh 					 const char __user *data,
2373d14c5d2SYehuda Sadeh 					 loff_t off, size_t len);
2383d14c5d2SYehuda Sadeh extern int ceph_copy_to_page_vector(struct page **pages,
2393d14c5d2SYehuda Sadeh 				    const char *data,
2403d14c5d2SYehuda Sadeh 				    loff_t off, size_t len);
2413d14c5d2SYehuda Sadeh extern int ceph_copy_from_page_vector(struct page **pages,
2423d14c5d2SYehuda Sadeh 				    char *data,
2433d14c5d2SYehuda Sadeh 				    loff_t off, size_t len);
2443d14c5d2SYehuda Sadeh extern int ceph_copy_page_vector_to_user(struct page **pages, char __user *data,
2453d14c5d2SYehuda Sadeh 				    loff_t off, size_t len);
2463d14c5d2SYehuda Sadeh extern void ceph_zero_page_vector_range(int off, int len, struct page **pages);
2473d14c5d2SYehuda Sadeh 
2483d14c5d2SYehuda Sadeh 
2493d14c5d2SYehuda Sadeh #endif /* _FS_CEPH_SUPER_H */
250