虚拟文件系统

Linux 使用 filesystem interfaces 用于 both stored 数据和 live 内核 state. The Virtual Filesystem Switch (VFS) makes those objects accessible 通过 common system calls, 同时 pseudo-filesystems such 作为 procfs 和 sysfs generate their contents from kernel objects. Similar path syntax does 不 give a pseudo-file the persistence or I/O semantics 的 a disk file.

VFS 作为内核抽象 #

VFS 是 the 内核 layer behind 操作 such 作为 open, read, write, stat, 和 mount. It represents mounted filesystems 通过 common 对象 types:

a superblock represents a filesystem instance, 同时 a mount attaches that instance 在 a location in one mount namespace;
an inode represents an 对象和 its metadata;
a dentry associates a name，使用 an inode 和 participates in pathname lookup;
a file represents an open instance 和 its current position;
an address space connects cacheable 文件 contents 到 the page cache.

Filesystem implementations provide operation tables 用于 relevant objects. ext4 can map reads 到 persistent blocks, NFS 可以 issue network requests, 和 procfs 可以 format a 内核 structure. Callers still receive 文件 descriptors 和使用 the same system- call boundary.

“Virtual filesystem” 是 therefore ambiguous. VFS names the abstraction layer; a pseudo-filesystem 是 one filesystem 实现 whose objects do not correspond 到 durable files 在 a storage volume. tmpfs 是 also memory-backed, but it provides ordinary mutable 文件 semantics 和 distinct 数据 per mount. procfs and sysfs expose purpose-built 内核 interfaces instead.

Procfs 进程与系统视图 #

Procfs 是 conventionally mounted 在 /proc. Numeric directories expose processes, while nonnumeric entries expose system-wide state 和 control interfaces:

text

/proc/1234/status      formatted state for process 1234
/proc/1234/fd/         links for its open file descriptors
/proc/self/            the reading process's own proc directory
/proc/meminfo          memory-accounting snapshot
/proc/sys/             sysctl parameters

Files under /proc/<pid> 是 checked against 进程 credentials, ptrace access rules, capabilities, mount options, 和 security modules. 一个 pathname being readable by root in one context does 不 make it public 到 an unprivileged process.

实例、PID 命名空间与挂载命名空间 #

Procfs output 是 namespace-sensitive. 一个 procfs mount associated，使用 a PID namespace shows tasks visible in namespace; /proc/self 和 /proc/thread-self also resolve relative 到 the caller. 一个进程 creating 或 entering another PID namespace must 使用 an appropriate procfs mount 到 obtain a matching 进程 view.

Mount namespaces independently control where 和 whether procfs instance is mounted. Containers 可以 consequently have different /proc mount points, read-only settings, 和 visibility options 同时 sharing the same kernel. Options such as hidepid restrict access 到 other users’ 进程 directories, 和 subset=pid can hide top-level entries unrelated 到 tasks 在 supporting kernels.

Procfs 应当不 be modeled 作为 one global 目录 tree cached 由 pathname. The same path 可以 resolve 到 different information 用于 different callers, namespaces, credentials, 或 moments in time. Open descriptors 到 a dead 进程 do 不 become handles 到 a later 进程 reuses the same PID.

Sysfs 与驱动对象模型 #

Sysfs, conventionally mounted 在 /sys, exports the kernel’s 对象 model. A registered kobject receives a 目录 beneath its parent, so 目录 hierarchy expresses 对象 relationships 和 controls lifetime. Linux 设备 model builds devices, drivers, buses, 和 classes 在 infrastructure:

/sys/devices presents devices in their physical 或 firmware-discovered topology.
/sys/bus/<bus>/devices links 到 devices participating in a bus.
/sys/bus/<bus>/drivers presents registered drivers 和 their bindings.
/sys/class/<class> groups devices 由 function, such 作为 network 或 block devices.

Many entries outside /sys/devices 是 symbolic-link views 的 the same underlying device object. Scripts 应当 resolve documented attributes rather than infer physical topology 来自 a convenient class path.

A sysfs attribute connects a small 文件到 subsystem-defined show 和 store callbacks. core allocates a page-sized buffer 和 asks show 到 render a value or passes one write buffer 到 store. Attributes 应当 contain one ASCII value, or an array 的 values 的 the same type, because simple text forms 是 inspectable and avoid embedding an ad hoc structured protocol. Binary attributes exist 用于数据 that cannot follow convention 和 have separate callbacks.

将文本文件视为 ABI 端点 #

Text does 不 mean informal. 一个 documented procfs 或 sysfs 文件是 a 用户空间 ABI with file-specific syntax:

Read the complete 值和 parse 仅 forms promised 由 its ABI documentation.
Include the expected newline 当 writing conventional text attributes.
Do 不 depend 在 undocumented whitespace, column positions, 目录 ordering, 或 the presence 的 testing-only interfaces.
Check write errors 和 short writes; a successful open does 不 imply a proposed 值是 accepted.

Mode bits 和文件 permissions 是仅 one access-control layer. Capability checks, ownership, namespaces, lockdown policy, 和 Linux Security Modules 可以 reject an operation. World-writable sysfs attributes 是 intentionally discouraged because a write often changes live 设备或 driver state.

kernel’s Documentation/ABI tree classifies interfaces 作为 stable, testing, obsolete, 或 removed. Software tied 到 a testing attribute accepts a different compatibility commitment 来自 software using a stable one. Feature detection should check 用于 a documented path 和 handle ENOENT; 内核 version comparisons cannot account 用于 vendor backports 和配置 choices.

读取是快照而非事务 #

Many pseudo-files 是生成的当 read. Kernel state 可以 change between opening, reading, 和 parsing a file, 或 between reads 的 two related files. Even one large procfs report may summarize fields observed 在 slightly different times. Rewinding a sysfs attribute 到 offset zero 可以 invoke its show callback again 和返回 a new value.

Consequently, multiple reads do 不 provide an atomic 系统 snapshot unless the specific ABI promises one. Robust consumers should:

使用 one read operation 或 read loop 用于 a small attribute;
tolerate processes 和 devices disappearing，使用 ENOENT, ENODEV, 或 related errors;
validate identifiers again 当 acting 在 previously read state;
使用 a dedicated transactional API 当 consistency across objects 是 required.

Some sysfs attributes support poll/select notification 当 their subsystem calls the sysfs notification mechanism. Such notification 是 opt-in per attribute, 不 a property 的 every 文件 under /sys. Consumers generally reopen 或 seek 到 offset zero 和 reread 在……之后 an event. Procfs entries likewise have file-specific polling semantics.

伪文件不是持久化存储 #

Several ordinary-file assumptions fail 用于 procfs 和 sysfs:

reported size 可以 be zero even though read returns data;
mmap, arbitrary seeks, append mode, 或 partial writes may be unsupported;
目录 entries 可以 appear 和 disappear 作为 processes 和 devices change;
a successful write 可以 trigger hardware activity rather than store retrievable bytes;
数据 vanishes 在 shutdown 和是 reconstructed 来自 the next kernel’s state.

Configuration 必须 survive reboot belongs in a persistent 配置 system whose startup logic deliberately writes the required 内核 interfaces. Copying /proc 或 /sys cannot back up 内核 state, 和 restoring copied files cannot recreate processes 或 devices.